The information isn't just in the base pairs. It's in the entire evolved ecology (with optional industrial add-on pack) that makes the base pairs do something useful.
How much of that could you take away and still get a human brain?
Craig Venter says here[1] that Franz Och (of Google Translate fame) has done some analysis for his new company Human Longevity Inc that shows that about 50% of the genome is responsible for the brain. So the number is quite big and much larger than I would have thought. Excited to see more results from HLI.
70 per cent of protein-coding human genes are related to genes found in the zebrafish (Danio rerio), and 84 per cent of genes known to be associated with human disease have a zebrafish counterpart.
Given unlimited computing power, AIXItl [1] is a pretty simple algorithm that can provably solve all problems, in some sense, at least as well as any other algorithm. The idea is to simply dovetail over all possible algorithms and select the ones that fit observations best. That includes humans, if you believe as I do that humans are computable.
With limited computing power, it's likely that the best algorithms for many different problems (possibly including "intelligence" however you define it) won't be simple, just like the best known algorithms for integer multiplication [2] aren't simple. In particular, AIXI variants will be hard to approximate, precisely because they dovetail over all possible algorithms.
That said, it's very likely that the best algorithms for intelligence will be simpler and faster than humans, because humans are the stupidest possible creatures that can build a civilization (otherwise it would've happened earlier in our evolution).
>Given unlimited computing power, AIXItl [1] is a pretty simple algorithm that can provably solve all problems at least as well, in some sense, as any other algorithm.
Technically, AIXI_{tl} solves all problems at least as well as any other algorithm in a given order of asymptotic time and length, those being the t and l. The problem is that it has a "trivial additive constant" which will make it take longer than the lifetimes of many stars to achieve its much-vaunted "optimal" asymptotic performance.
In short, AIXI is the most utterly brute-force kind of "intelligence" you could possibly build, whose notion of "scaled down" bounded-rational inference is still astronomically intractable. It basically just amounts to throwing computing power at the problem, more computing power than anyone actually has.
Mind, Schmidhuber and Hutter still get massive props for managing to cut out the philoso-wank that usually accompanies discussions of cognition and instead saying, "Let's just pose a very general inference problem and write an algorithm that solves it."
Yeah, that's why I said "given unlimited computing power". But theoretical work sometimes has a way of becoming practical and scary. Check out the work of Joel Veness on Monte Carlo AIXI, which learned to play Pac-Man on a single desktop computer, presumably faster than "the lifetimes of many stars".
Writing a bot that plays Pac-Man is many things, but I wouldn't call it "scary". Especially considering the rules of Pac-Man had to be imperatively coded in relative detail [1], that some of the utilities were factored out into a basic reinforcement learning agent doesn't immediately convince me we've hit AGI at all.
Having to rely on playing an arcade game is a weak case in general. We don't call Counter-Strike modders AI researchers. The fear angle is nothing but woo, but I'd like to hear you elaborate nonetheless.
>Especially considering the rules of Pac-Man had to be imperatively coded in relative detail [1]
See, chessboards are meaningless outside of human experience. You only know how to play pac man without explicit instructions because you have a general concept of a game and what games look like when you win or lose them. Whenever you try to solve an AI problem in a narrow domain like this then you necessarily have to compensate by giving it a miniature model of the specific problem being solved.
Humans after all don't just intuitively know the rules of games before playing them. You also cannot just describe the rules of the game to the AI because the AI does not understand language.
I understand this. I'm also saying this is done pretty routinely in the video game industry from everything to the most basic pathfinding AI to more complex agents that factor in the terrain, particle effects and other aspects of the environment hardcoded into their processes to exhibit "intelligent" behavior, though typically only within that box. Game modders then engage in plenty of work from reverse engineering to what can realistically even be called AI programming.
MC-AIXI seems to be more generic because it factors this out into a reinforcement learner, but this in of itself is hardly a cause for fear and that AGI is coming, as the GP alluded. I'd wager today's AAA titles have some damned complicated heuristics of their own.
It did happen, several times! But the randomness of weather, volcanoes, disease etc squashed them. Ours is just the lucky combination of critical mass and congenial weather patterns.
On the other hand, Neanderthal had quite a bigger brain than us; but fewer of the social impulses. We're likely dumber but more socialized, which reinforces your theme I guess!
AIXI solves a useful class of problems, but definitely does not solve all problems [1].
The main class of problems it fails to solve are ones where it has to model itself as an agent in the world.
For example, AIXI is a bit accidentally-suicidal. Its world models have no term for "I am here"; e.g. it will fail to realize that disconnecting the power to save money would stop the inference process instead of informing it. It needs many many examples of dying in order to learn to avoid it, but you only get one chance. (Evolution worked around this problem by making copies, and preferentially keeping copies that were less accidentally suicidal.)
AIXItl though has the problem of self destruction. It is still modeled as an agent an so it would eventually destroy itself during its dovetailing. The agent model of AI is a bit too dualistic to be correct.
> humans are the stupidest possible creatures that can build a civilization
Kind of a tangent, but there are conditions that are not easily described as "intelligence" that are necessary for building civilization. Namely, the ability to cooperate in very large groups. Chimpanzees outperform human 4 year-olds in a variety of intelligence tests, but adult chimpanzees are not observed to cooperate to the same degree as humans, or in very large groups. There is also some archaeological evidence to suggest that this is one of the reasons humans were able to outcompete Neaderthals - larger tribes. Individual Neanderthals may well have had higher IQs than their contemporary human rivals.
I saw Shane Legg's talk recently and I understood the main idea but I had a question I wish I could ask him:
He said general intelligence is 'as long as you provide a goal, the intelligence should optimize over that goal based on experience'.
But my problem is: general intelligence is supposed to come up with its own goals.
Take for a example a couch potato who has some steady stream of monthly stipend, and sits in front of tv all day. This person is in possession of a functioning general intelligence yet has no goal. One day this person decides to assign to self a goal of going out and getting a degree and so on and so forth. That's also part of that person's general intelligence.
It's not clear to me that AIXI is self-driven in that sense. It would be great if someone could comment on that.
I think the idea is to build something usefull (learn from experience and optimize goals)instead of something dangerous (own goals).
And I dont think that own goals are a neccessary condition for general inteligence. But it depends on the definition of general intelligence, of course.
> In particular, AIXI variants will be hard to approximate, precisely because they dovetail over all possible algorithms.
For this reason, I suspect that invoking AIXItl is begging the question. Positing unlimited computing power allows one to sidestep the difficult question: which algorithm(s)? In fact, this is not just a difficult question, it is the question (the one that AI research is trying to answer.) Absent the convenience of unlimited computing power, I suspect that AIXItl is a dead end, no matter how elegant it is.
Yes, there is a simple algorithm for intelligence. It's just a few lines of code. However, it requires a terabyte-sized table of data which has yet to be filled in.
This ignores that genetic code is meaningless without the translation machinery (ribosomes and the entire physical world of physics and chemistry). Genetic code without the translation machinery and the physical world it is translated into is meaningless.
A multi-cellular embodied intelligence can acquire information about the patterns of the physical world by growing in it - the code is compact because much of the necessary information can be acquired during embodied growth.
A disembodied intelligence such as AIs can't rely on external storage and acquisition of information during a cellular growth stage, and thus must likely be considerably more complex from the get-go.
It also ignores the possibilities (1), that cells and tissues are extremely complex (and thus require a lot of genetic code), while the overall algorithm the cells and tissues perform might still be very simple, and conversely (2), that chimpanzees might be already extremely intelligent and the difference is rather a matter of embodiment (in particular the ability to produce language), and perhaps minor tweaks to neurons, to the cortex and/or other brain regions.
If this were a correct interpretation, even Bacteria could be seen as already incredibly intelligent. Going Zero to One (dead matter -> prokaryotes) is a significantly bigger step than going One to Five (prokaryotes -> humans).
Careful, modern prokaryotes also have billions of years of evolution behind them. The first organism where likely inefficient and would not last in our current hyper competitive ecosystems. Arguably, the first organism where really more just free floating proteins without even a cell wall.
His numbers are also based off the difference between humans and chimps for some reason? We can't replicate the intelligence of a chimp on a computer, so why that is a valid basis of comparison is beyond me.
Quite. You would suspect that if we were able to produce an AI with the intelligence of a chimp then a human-level AI would be in reach. The harder step is probably getting to the level of chimp intelligence; you'd have managed to create consciousness and solved the hard problem for a start.
Moravec's paradox is not an experimental result. It's just one guys musings on intelligence. It says a lot more about the misguided things people thought about intelligence in the early days of AI than the actual inherent difficulty of achieving human-level AI.
According to the linked article, several notable researchers were of the same opinion (and one might also say that relativity was just one guy's musings - though admittedly experimental evidence wasn't long in coming.)
I do, however, agree with your second sentence, and I think the paradox is also weakened by the fact that when 'high-level' mental tasks are computerized (chess-playing, for example), it is often through brute-force calculation that is not a model of how humans use their general intelligence to perform the task.
Mind if I go a little off-topic here? This paradox reminds me of some other conclusion I've had, which is the fact that energy requirements of many physical processes are completely counterintuitive. For instance, it probably takes more energy to light a light bulb for a day, than it takes a small processor to run a set of calculations equivalent to a lifetime's worth of arithmetic done by every human being alive today.
>Kind of renders all of his information theoretic calculations moot.
The stuff in the article isn't information theory. Information theory, in the senses of Shannon and Chaitin, is actually very useful for studying cognition, because it tells us about how much we can try to learn, via what methods, from noisy sensory data.
> Chimps are remarkable thinkers in their own right. Maybe the key to intelligence lies mostly in the mental abilities (and genetic information) that chimps and humans have in common. If this is correct, then human brains might be just a minor upgrade to chimpanzee brains, at least in terms of the complexity of the underlying principles.
Of course this still kills the argument that it's at most 125MiB in my opinion. Moravec's paradox [1] implies a lot of the actually-hard tasks related to inference and learning are already solved in chimps, because they can see and hear and balance just fine. I think if we actually understood how chimps reasoned, in high-level and low-level terms, then we could easily tweak and improve that process to function beyond human level.
I believe Moravec's paradox is a consequence of the fact that we do not consciously recognise the effort our brains/bodies do in sensorimotor computation. When we add up numbers consciously in our head, using arithmetic, it takes us quite some effort, but in the meanwhile our visual cortex is doing similar operations (unconsciously) millions of times more efficiently.
It's as if the conscious arithmetic runs on some kind of stack of virtualization layers, using millions of neurons to build a mental image of the concept of "the number five", whereas a handful of neurons would suffice to do the actual arithmetic.
I often wonder if those rare people who can do calculations in their head faster than calculators, may have - unconsciously - found a way to unlock the native hardware of their brain, bypassing all of the symbol abstractions required by us muggles. A bit like GPGPU vs CPU for certain algorithms - only orders of magnitude more pronounced.
Yeah, the genetic code alone is like having a bunch of source code with no compiler, no spec to write a compiler, and the language is something you've never seen before and it has no similarities to anything you had seen before.
It's like being giving a bunch of Brainfuck source--if you'd never heard of Brainfuck.
>> This ignores that genetic code is meaningless without the translation machinery
Not really. The translation machinery is also defined in the DNA. Granted, there is a bootstrapping issue. The chimp argument seemed silly to me. Why not just estimate so fraction of the human genome as defining the brain? In that case 3G base pairs is a bit under one gigabyte of information defining an entire human. Still small in comparison to the quantity of synapses.
The translation machinery is nowhere represented in the DNA. Its the cell soup, that came from mother and her mother before her, back to the primordial soup. Dna is the paper tape; the egg is the computer (or autofac, or self-replicating machine etc). That paper tape has nowhere the instructions for making the egg; only detail for how it operates.
I think the DNA does contain instructions for making the egg.
For example, if executing the DNA instructions produces a human female, she'll be born with a lifetime supply of eggs, the developmental instructions for which were part of her startup DNA. Those instructions were definitely not part of the "cell soup", as is shown by mitochondrial donation assisted reproduction technology (aka 'three-parent babies').
The egg made the eggs. The DNA said "make an egg". That's the idea. Like a paper tape that you feed into a factory, with one punch, labeled "Man/Mouse". Punch 'Man' and feed it in, and the factory makes a man. So, a man is encoded in one bit? No.
Cells are factories, incredibly complex ones. DNA is a set of punches in a tape.
>> The translation machinery is nowhere represented in the DNA.
Why yes, yes it is. There are genes that code the proteins that make up the ribosome for example. The ribosomes in turn interpret the DNA string to make proteins. This leads to a bit of a chicken and egg problem which I mentioned as the issue of bootstrapping. But the DNA does code everything in the cell, or at least defines the machinery needed to make everything.
The embodied cognition perspective provides such an interesting counterpoint to the notion that a brain is just a computer in these philosophical debates over AI vs humanity.
> A multi-cellular embodied intelligence can acquire information about the patterns of the physical world by growing in it - the code is compact because much of the necessary information can be acquired during embodied growth.
Not sure why you think "embodied growth" is essential here; having a compact code base that can be supplemented by information during maturation obviously has some utility, but why would you think embodied growth is critical to it?
> A disembodied intelligence such as AIs can't rely on external storage and acquisition of information during a cellular growth stage
AIs aren't disembodied. They have physical substrates, just like natural intelligences.
I don't see why something like cellular growth is necessary for "external storage and acquisition of information" during development. Even if it were, however, just as natural multicellular organisms are coordinated colonies of self-reproducing cells, artificial organisms (intelligent or otherwise) could be coordinated colonies of self-reproducing machines.
AIs are not disembodied; they obviously are embodied in physical machines. And there's no reason an AI couldn't acquire information during development,
I think the GP's point is just that there is unaccounted-for information about existing in the physical world encoded into that embodiment. AI will also have embodiment, and there will necessarily be information stored there, but a great deal of operational information will have to be learned and explicitly expressed in the AI itself. To use your example, an artificial organism could be made of coordinated colonies of machines, but cells already have a huge head start in the way they bind to one another, transport materials within and without, respond to various signaling agents, etc. Down to the basic molecular building blocks of the cellular machinery has been selected to (attempt to) optimize operation across all actions that cell might take.
The argument isn't that that embodiment is necessary for the artificial system to learn how to operate in a way that's as capable or better, just that there's a whole lot of information there not accounted for in the genetic code alone.
I hold the same belief. I also think that we don't pay enough attention to the form of processes whereas nature seems to place lots of emphasis on it. In fact, the more it goes, the more a form factor appears to be somehow determinant. Molecular biology with enzymes, genetics with histones, neurobiology with firing patterns, etc. We can see that now because we have access to finer grained observation. Form could be a higher-order type of content and could also be approximated with an algorithm.
This all feeds in a previous discussion about intuition. Top-down or bottom-up?
In the 16th century only a foolish optimist could have imagined that all these objects' motions could be explained by a simple set of principles. But in the 17th century Newton formulated his theory of universal gravitation, which not only explained all these motions, but also explained terrestrial phenomena such as the tides and the behaviour of Earth-bound projecticles.
Newton's gravity was a theory about a single fundamental force - and it turned out to be not 100% accurate. Rather than be an example of how intelligence might be similarly possible to explain with a simple theory, I think it offers a perfect counter-example of why it almost certainly cannot.
Comets, and the planets and the tides, and apples falling from trees were found to be caused by the same underlying phenomena, despite seeming very different. Newton's insight that they were the same was not overturned by relativity, so the point stands quite well.
Newton's insight that they were the same was not overturned by relativity
I almost feel like you're replying to someone else because I never said what you are contradicting.
I merely said that Newton's gravity is a simplification that doesn't take all variables into account. So to point to Newton's gravity as an example of, "Sometimes simple things can be right!" actually makes it more of a counter-example.
I was pointing out that the complex explanation is that comets have their own driving force, and tides have a separate driving force, and planets have their own reason for moving as they do, etc. Compared to that kind of hodge-podge, relativity is still a really simple theory. While relativity increased the complexity of the explanation of gravity, it didn't bring it anywhere near the level of complexity that might have been assumed before Newton.
The author's argument is that our intelligence is likely not solely possible by the interaction of a large number of separate principles, but rather by some simple principle that manifests itself in different ways. I think if we find some simple principle that explains intelligence, and then later have to make a correction to the math of that principle, the resulting explanation will still be simpler than the idea that intelligence is made up of many different principles all of which have to be present.
If we had a theory of intelligence that worked as well for reasoning as Newton's theory of gravity worked for modelling physics, we could ask the machines built with it to help us in figuring out the model accounting for the equivalent of relativistic effects.
Simplicity is always only a little ways off. If I sat you next to god and told you to pair program and write an OS equivalent to Windows XP, with you driving the keyboard- you'd be sitting there for quite a while.
But if I told you guys to pair program an AGI, you could probably finish in a day or so.
A simple algorithm is a matter of finding just one or two basic discoveries. It could happen tomorrow! It probably won't, but that's the nature of discovery.
Well, supposing I'm not programming with god, rather I'm with a bunch of programmer friends, I still doubt the first artificial general intelligence (assuming we happen to be the people who make it) will be a beautifully simple piece of software. It'll probably be horribly hacked together and hard to maintain, and only in time will a simpler way to do things emerge.
For example, I'd reckon we'll find that a lot of things that today only make sense to solve with a neural net are actually perfectly representable using a simple(ish) algorithm, but it's unlikely we'll find that out until after we create an artificial general intelligence and know more about the problem of intelligence.
In other words, we'll solve the problem, and then realise how much easier it could have been.
> "I believe it's not in serious doubt that an intelligent computer is possible" -
Yes, of course it is possible, if we change the meaning of "intelligent" to fit what a computer does. :)
Generally, I think the problem here - as in many other situations - starts with the lack of rigor that computer science has when "naming" fields or, most exactly, appropriating concepts from other fields.
* Computer Vision: "vision" has a defined meaning, even "organic-related". Signal processing and a bunch of algorithms aren't exactly what vision is and what we define as "vision";
* Machine learning: does the computer really learns, as understood by the common definition of "learning"?
* Context awareness: if we consider common definitions, neither a few sensor readings are "context", nor acting according to said readings is "awareness".
This might sound like nitpicking, but I truly think it's a real problem. Talking about vision, learning or intelligence put's the stake at what people or the general scientific community commonly perceive as those concepts.
What's most frustrating to me is the contrast. Computer science research is thorough; why are we not thorough on naming our fields of study? Why do we appropriate concepts by "analogy", knowingly that analogies aren't truly exact?
> Yes, of course it is possible, if we change the meaning of "intelligent" to fit what a computer does. :)
I don't think that's what he's doing. Otherwise he could just define intelligence to mean what computers do now. Clearly he has at least roughly the same idea of intelligence you or I do.
> "I believe it's not in serious doubt that an intelligent computer is possible"
Do you have an objection to this specific statement when taken to mean general human-like intelligence being exhibited by computers? I have not heard anyone defend the opposite viewpoint often, and I'm interested if there are any compelling arguments why it's not possible.
> Do you have an objection to this specific statement when taken to mean general human-like intelligence being exhibited by computers? I have not heard anyone defend the opposite viewpoint often, and I'm interested if there are any compelling arguments why it's not possible.
A few aspects which relate and shape intelligence:
1) Socialization;
2) Embodiment (the fact that I have a body and the body has feedback loops - affects and is affected by cognition);
3) Emotions (we have "hunches" and those hunches are surprisingly many times adequate);
4) Language (our sign system affects our perception of the world).
Will a computer be able to be a sociable, embodied person? I tend to think that it won't, so that's roughly why I believe a computer won't ever have "intelligence" (unless, again, we redefine intelligence).
1) 2) 3) Are I think research goals of robotics. None are simple but I don't personally think they require anything that isn't possible to accomplish with computer + physical objects beyond the computer itself.
While I think there is some grandiose and hype that gets attached to these fields by the media, you have to recognize that these names represent the end goal of the field, not the current status of where we are today.
No, computer vision researchers haven't achieved what you and I perceive as vision, no more than a biochemical research in cancer cures have found a cure to cancer.
But I would argue that these lofty named fields are a good thing. Academia is already plagued by it's uncertainty and lack of an inherent "end game." Let's not make it worse by renaming machine learning to "computer pattern recognition." Pattern recognition is just a possible piece of the puzzle or step in the process to creating AI, not the actual holy grail of achievement.
I see your point but - according to the way I was thought - you shouldn't use operative concepts so "freely".
If you use e.g. "vision" in your research as a basic concept, then you should do a thorough state of the art inquiry and state exactly what you understand by it (using, of course, the state of the art). I don't believe this was done when someone thought of naming these fields as they did. I also don't think that beginning researchers do this basic "background" check; they try to achieve "vision" (or "intelligence") without a clear (and at least commonly accepted) definition of "vision" (or "intelligence").
As I said above, I can see your point, but I wouldn't do this in the scientific community. Nothing wrong about saying in the news that you want to make a computer see, but e.g. having a journal with a misdefined or malappropriated concept in its name is - IMHO - wrong.
> I believe it's not in serious doubt that an intelligent computer is possible - although it may be extremely complicated, and perhaps far beyond current technology - and current naysayers will one day seem much like the vitalists.
Physicists (and this guy is one of them) certainly do have a robust set of taboos against considering consciousness as a quantum phenomena (whatever that might mean.) It's unfortunate because there is a huge resurgence going on in the physics of quantum many-body entanglement, quantum computers, foundations of quantum, etc. Alot of this meandering was just sidelined in the 1940s when world affairs intervened. However, in light of these new discoveries it's time to revisit the old vitalist arguments. Imho.
If consciousness is a quantum phenomenon it has coherence scales orders of magnitude beyond what we can achieve in labs at ~1 Kelvin. That is, it's pretty unlikely that quantum effects are having macro scale effects on the brain.
That's a good point. We should probably not trust this "lab tinkering" because it's so new - nature already figured everything out already. I'll go back to hunting and gathering now...
Let's say, for the sake of argument, that intelligence requires quantum effects to work. I don't see how going back to vitalist arguments will aid us in any way. Researching the phenomena, following evidence, making inferences from that evidence... that's how progress on that subject would be made. Not by going back to arguments made by people with less information than we have now.
Specifically, if we follow the evidence, and find that somehow, beyond all odds, the vitalists were correct (once you add "quantumness!" to the mix), then it's clearly only by lucky guess they were right. It would be something we could look back at and say "wow, that's an interesting coincidence" rather than "wow, those vitalists had it right all along, we should have paid closer attention".
Its often by intuition that we know where to look. Lets not be arrogant and claim "ok, you guessed it right, but its only true if I measure it and prove its true". That's correct but disingenuous.
That's what I meant by rational inference, but if we look at it from the point of view of intuition, I'm saying our intuitions will be better because we have much better information now. We have no particular reason to weight heavily the intuitions of the vitalists in our search.
Intuition can be correct, and it can be wildly wrong.
Science is how we separate the correct, the mildly wrong, and the wildly wrong intuitions.
Its not an alternative to intuition -- intuition is where hypotheses come from. (OTOH, vitalism isn't a hypothesis, simply an intuition that some aspects of living things are outside the domain of science. Its not testable -- if correct, its completely indistinguishable from some of the relevant aspects of living things simply not being sufficiently well modeled by the current state of science at any given time. So, its an intuition that is neither confirmable nor useful in any way.)
I guess all scientific progress has been by coincidence then? That's what I mean by disingenuous; some credit has to be given to those that are right. There are an infinite number of wrong intuitions.
No, scientific progress comes from empirically chasing down and disproving false intuitions so that only the few true ones remain. Intuition creates hypothesis, empiricism makes them useful since nearly all intuitions are wrong. It's the empiricists that deserve the credit, they are why science works and of course are generally the ones with the intuitions that are correct as well.
And you can't say vitalists were right, they've made no predictions: calling something magic woo woo isn't a hypothesis that can be proven right.
Which begs the question: where did the correct intuitions come from? Can't be random; there are billions of wrong ideas and only 1 right one. So some weight has to be given to intuitions, and some props to those who had the right ones. Because that's how progress gets made.
> Which begs the question: where did the correct intuitions come from?
Logic.
> Can't be random; there are billions of wrong ideas and only 1 right one.
It's not random, it's deductions based on previously known truths.
> So some weight has to be given to intuitions, and some props to those who had the right ones. Because that's how progress gets made.
Props are given to those people, they're called scientists. There are no correct intuitions from the vitalists, they're claim is magic... that's not in any way useful.
> I guess all scientific progress has been by coincidence then?
Correct hypotheses are, in a sense, a product of fortune, and scientific progress requires them to happen sometimes, so, in a sense, that's true. Without a certain minimal degree of luck, you would have no scientific progress.
The scientific method can be viewed as simply a method of filtering out the intuitions without utility and promoting those (if any) that have utility. It is a way to avoid wasting the good fortune of the intuition of forming useful hypothesis, and avoiding the waste demonstrably fruitless intuitions.
> some credit has to be given to those that are right
But vitalism is so vague as to be untestable, and justifies no useful models. "Some (unspecified elements of) living things are not governed by physical law" is indistinguishable in effect from "We have not yet discovered the physical laws governing (some elements of) living things".
> Without a certain minimal degree of luck, you would have no scientific progress.
A minor disagreement on this point. I think often our intuitions are shaped by experience, and lead us to conclusions that we don't have a rational line of inference to (at least one we can explain). Science acts as a good filter, saying "we don't care how you came up with it, test it and we'll agree if it turns out it's right". This isn't a random guess, but it's far from reasoning you could convince anyone else with. This is different from a lucky guess. Your brain has been seeing a lot of the data, and is picking up on patterns, in other words your intuition is correlated with the phenomena, and that's what leads you to the theory you want to test.
Second, sometimes purely rational arguments lead us to theories from other data. This isn't random chance, but it is exploiting regularities in the world around us to predict new things. For example, kepler's laws of motion weren't the direct result of observation. You can't observe equations that guide the motion of the planets. In between observations, the planets might have jumped all over the place. But it was assumed that the planets weren't doing that, and simple equations were inferred from the data available that fit the evidence. There is a step here between "observation" and "prediction" that doesn't involve a lucky guess.
If intelligence is based on magic juju, it's achievable by a computer that uses magic juju. To have a situation where intelligent humans are possible but intelligent computers are impossible, you need a decree from Cthulhu saying humans are special. I don't think that's likely.
> Physicists (and this guy is one of them) certainly do have a robust set of taboos against considering consciousness as a quantum phenomena
They really don't, but they do have taboos again ignoring the fact that there's no evidence for such a hypothesis, as do all scientists. Science is empirical and scientists don't generally just like to make shit up because it sounds cool.
Just from skimming the article it seems to me the problem with AI and probably just the whole matter of what people define as intelligence is that it assumes that the universe itself is not intelligent. Maybe the universe itself is intelligent but that consciousness is some other phenomena that gives intelligence an individual characteristic? It's just a random thought, I don't really know which way to go with it.
I can relate. While I consider myself an agnostic with a slight atheist streak, I still like to muse on these sorts of thoughts occasionally. The universe does seem somehow innately intelligent and alive. Everything is in motion. Everything appears to have some kind of structure. Gotta be careful not to fall back on the "G" word though (not that you did) :).
Personally, I think there might be a Creator, but that doesn't seem to be relevant as to the problem of consciousness. To me, when I see that physical processes act in a manner that is complex then it indicates that what we're trying to call intelligence is really just form of stateful structure. And the rest of intelligence is something else entirely different (and probably unrelated).
"The Universe" is huge and empty and completely hostile to life. Most matter is in stars which are not empty and completely hostile to life. (And which are huge compared to us, but which are like specks of dust compared to "The Universe").
Our planet is a speck of dust rotating at a fortunate distance around another speck of (shining) dust. By mass, the majority of the planet is useless to life. Everything alive on Earth exists in a very thin skin between lifeless rock and lifeless, airless, freezing space.
Then, on any timescale meaningful to "The Universe" individuals are tiny blips. Species are around longer, but not much. Vast majority of species are killed off by cosmic events regularly (we are on something like the 6th major extinction event, IIRC).
All of humanities struggles with itself, with ignorance, all of it's artistic expression, have taken place within the context of that thin skin of warm fluid surrounding Earth over a shockingly short time frame.
The best that can be said is that we now know the score: that "The Universe" is fascinating, vast, hostile, empty, silent, and our entire species is the rough equivalent not to a mewling baby, but rather to, perhaps, a single microscopic insect gestating in its egg, just beginning to break the shell and look around.
And yet here we are, living and experiencing the universe.
Most matter that you interact with is not in your body, and much of it is not in a usable form to your body, yet it still sustains you. Why should this not hold true at a larger scale?
We cannot survive in the colds of space, yet it is the tendency of matter to clump together, creating that space, which also gives us a sun and a planet and an atmosphere that was necessary for our existence. We cannot survive in the lifeless molten rock, yet we also could not have formed without its heat. It's usually at the edges of things where the most interesting and varied phenomena arise.
Much of the universe is hostile to human life in its current form, yet it is also the universe that gave rise to it. Whether human newborn or microscopic insect hatchling, the experienced world is at its most incomprehensible, and the organism at its most vulnerable, at the beginning. Growth allows for greater experience, and experience can lead to greater growth. Our knowledge of what's possible for a species extends only backwards, and only on one single speck of dust.
The universe has no inherent vastness, hostility, emptiness or silence without a conscious being to assign it those properties. I choose to imbue it with aspects of magnificence, intelligence, organization, perhaps even benevolence. We are not in a cosmic battle to conquer a dumb, hostile substrate; we are embedded in a rich, nurturing agar whose purpose it is, at least in part, to help us to flourish.
>The universe has no inherent vastness, hostility, emptiness or silence without a conscious being to assign it those properties.
Agreed.
I choose to imbue it with aspects of magnificence, intelligence, organization, perhaps even benevolence.
Well, that's nice, but it's only possible at the space scale of, at most, the Earth. Or at least human scale - since it is you who are doing the choosing. I mean, it's fine to change the subject, but at least be aware that you are doing so.
>We are not in a cosmic battle to conquer a dumb, hostile substrate; we are embedded in a rich, nurturing agar whose purpose it is, at least in part, to help us to flourish.
Again, I feel like you are confusing space- and time- scales. Earth's biosphere is nurturing agar, no doubt, over a short time scale and small space scale. But the solar system, galaxy, let alone the universe at large, is most definitely not, and nor is the Earth's biosphere on any large time scale.
We have arisen in this universe, apparently the first intelligent life on this planet, and on any other planet. If there is intelligent life elsewhere, we are separated by impossible vastness: to get to another star, in any form, we must harness immense energy. More than is reasonable. And once there, the lifeboat must somehow found a colony, a self-sustaining biosphere, which itself is a gargantuan task.
Perhaps, though, this is a benevolent universe: maybe those great chasms of space protect intelligences from each other. It means that if any of us manage to make the journey, and we discover each other, it will be a meeting of great good fortune, and the travellers will always be at the mercy of the inhabitants of the destination star.
There is hope for long-term survival of intelligence, but it is very, very small.
> Well, that's nice, but it's only possible at the space scale of, at most, the Earth. Or at least human scale - since it is you who are doing the choosing.
Hmm, my intent was to expressly claim those things at large scales, or more accurately, at all scales. It's just that it is easier for us to understand e.g. the organization and benevolence only in the things which are of the most immediate usefulness: we understand the biosphere as a resource because we can use that now, whereas we don't really understand how to harness the full power of a star because it's too far beyond our current capabilities.
> Earth's biosphere is nurturing agar, no doubt, over a short time scale and small space scale. But the solar system, galaxy, let alone the universe at large, is most definitely not, and nor is the Earth's biosphere on any large time scale.
I agree with you about Earth's biosphere space-wise though perhaps not time-wise. But if Earth is a cradle of intelligent life, surely we are not meant to stay here indefinitely? Even now the wild idea of mining asteroids, once relegated to science fiction, is reaching the point of feasibility. What was previously useless (not to mention unknown) to us is becoming useful, and will act as a stepping stone (or a catapult) to enable some other far-fetched ideas to come closer to manifestation. I believe this is a pattern that does not necessarily end.
> If there is intelligent life elsewhere, we are separated by impossible vastness: to get to another star, in any form, we must harness immense energy. More than is reasonable. And once there, the lifeboat must somehow found a colony, a self-sustaining biosphere, which itself is a gargantuan task.
There was a time, or more accurately various times for various peoples, that crossing an ocean was almost beyond imagination. As we continue to gain in knowledge and understanding of how the universe works, especially our understanding of spacetime, these problems become far less intractable. Interstellar travel is just a bigger ocean. The scale of the metaphor is logarithmic but so is the pace of the growth of our collective intelligence. My personal belief is that we will find a way around the immense time and energy requirements that we currently believe are necessary for interstellar travel. Terraforming is the harder problem: understanding ecosystems is IMO orders of magnitude harder than understanding fundamental physics, and building them from scratch seems like a pipe dream even for a dreamer like me. But if we figure out how to travel smarter then it becomes more a question of finding the most Earth-like planets that already have much or most of what we need. Still not easy, but probably easier and faster than trying to terraform Mars.
> Perhaps, though, this is a benevolent universe: maybe those great chasms of space protect intelligences from each other.
This seems plausible to me also. Seeds need a volume to grow in that is often tens of millions times larger than the seed itself. There could be a cosmic ecosystem that we are still too young to see or understand, and we may not catch a glimpse of it until we have spread through the stars.
> There is hope for long-term survival of intelligence, but it is very, very small.
If you mean small hope for human intelligence, then I would agree. I think this has less to do with the harshness of the universe out there and more to do with our own species' maladaptive behavior patterns. We are in a species-scale moment of crisis, which is to say one that has been going on for hundreds or thousands of years and may last a few hundred more, but there's no question in my mind that we need to get our shit together now if we want the species (and/or its children) to survive. This is why I think it is important to view the universe through a lens of possibility: it gives us a direction to steer towards. A more hopeful outlook inspires activity to push towards a better outcome.
If there were a simple algorithm for intelligence, and assuming the presence of intelligence confers a competitive advantage, then why wouldn't natural selection have created many examples of it by now, after some billion years of multicellular life? Yet it hasn't.
In fact intelligence manifests rarely in nature and only in long lived species with big complex brains which spend years educating their young to develop their cognitive skills. This implies that intelligence requires both a nontrivial nervous system and substantial nurturing thereof.
Ergo, I conclude that intelligence itself is inherently complex and unlikely to arise simply or spontaneously.
Every animal is intelligent. Using the google dictionary definition, "the ability to acquire and apply knowledge and skills."
What they are are intelligent about varies, and the assumption that intelligence arises rarely is dangerous to ecological and environmental decision making. Rather, we don't know how to communicate with animals well and we don't think the same ways.
> If there were a simple algorithm for intelligence, and assuming the presence of intelligence confers a competitive advantage, then why wouldn't natural selection have created many examples of it by now, after some billion years of multicellular life?
A reasonable hypothesis would be that intelligence comes with high cost, such as high energy consumption, high heat production, longer period of infant stage, adaptation of body parts that will reduce their strength and interfere with their functionality, etc.
All those negative effects more than offset the benefit of higher intelligence in the short term. Because natural selection does not plan but only pushes evolution toward local maximum points, most organism never evolves high intelligence.
> then why wouldn't natural selection have created many examples of it by now, after some billion years of multicellular life? Yet it hasn't.
It has, all animals display intelligence; I think you need to update your definition of the word, you're using it as if only humans have it and we haven't believed that in a very long time.
There are algorithms that make algorithms. (Even if we assume that the brain doesn't run on algorithms.) Genetic programming has made algorithms, for instance.
Generate random strings and feed them though a compiler. You have can create algorithms. That's not a very good algorithm but it shows algorithms that create algorithms exist.
Now most modern AI algorithms, such as neural net, don't create algorithms (at least not directly), yes. But that might be indication that AI has further to go.
A k-nearest neighbor classifier could be an example of a "simple" algorithm for intelligence, assuming of course, that you've collected a tremendous number of accurate (input, intelligent response) pairs!
For me, the question addressed by the article isn't even the interesting one. The author talks about "intelligence" as in how we process environmental information and perform reasoning tasks. From that perspective, it does seem obvious that some form of computer will one day be able to do it. I don't really care if the algorithms for doing so are simple or complicated.
For me, the real question of interest is not about that kind of intelligence, but of "awareness". Computers may process environmental information and perform all sorts of logical operations based on it. But there is zero reason to think that any computer has any awareness whatsoever. It has no experience. There is something it is like to experience the color yellow, as opposed to merely classifying observed light as being of a certain wavelength; no computer has begun, in any way, to make that leap.
It seems like a lot of people, including the author, don't even notice that this is a question. They seem to assume that computer-like information processing is all that happens in a human mind, and that awareness itself doesn't even exist as something to ponder. So they equate any doubt about whether a computer can do what we do as a form of vitalism.
But with just a little bit of introspection, the fact that there is something it is like to experience the color yellow, as opposed to classifying a visual stimulus as being of a certain wavelength, or logically classifying it as having the attribute of "yellowness", is absolutely clear -- and absolutely fundamental to our existence as human beings (although many would assume that many other forms of life also have this basic ability to experience, if not the same ability to process information in computer-like ways).
This is why, for me, the Turing Test is orthogonal to the question of whether a machine is actually conscious. I have no problem imagining that a software system, even on hardware that is basically the same as what we have now but bigger and faster, will one day be able to pass the Turing Test. But that doesn't prove it has any experience of the color yellow; it merely means it can mimic the output of an entity that does.
If a call center operator follows a "answering" algorithm strictly, is it a computer?
These thought experiments are very interesting and thought provoking. But - still - it's a mathematician asking if something is alive/human. Not the most authoritative source on the subject matter.
The original computers were women in warehouses calculating actuary tables. They had a set of criteria to perform evaluations of numbers.
In short, you can reduce a subset of the human experience to algorithmic emulation, which, in effect, reduces a human to a computer. It's sort of what it means to do any job to be frank.
> In short, you can reduce a subset of the human experience to algorithmic emulation, which, in effect, reduces a human to a computer.
You're underestimating everything associated with being human. The computer won't make mistakes because a familiar is sick. He won't perform slower than usual because he slept less. He won't have a urge to poop affecting is performance, also. This is somewhat what I meant: these thought experiments are interesting, but if you look closer there's really not much to them. Human computers were, of course, computing. They didn't, however, became "not human".
> The computer won't make mistakes because a familiar is sick.
Race conditions cause mistakes. Seems similar to having your mind elsewhere.
> He won't perform slower than usual because he slept less.
Didn't clean up memoery so now your computer is resource starved and running slower.
> He won't have a urge to poop affecting is performance,
GC pause.
I feel like people often focus too much on the "implementation" when it comes to consciousness. Beyond what instinct gives us, which is really just a pre defined set of conditionals, everything we know and understand is learned, programmed in a way. I don't see why it's unfeasible that given enough time and the right environment that an AI could do the same.
Those are beautiful analogies. They might look good on a magazine, explaining layman how a computer behaves. However, science-wise, there's nothing to it. Or will the loss of a familiar trigger a psychiatric condition on the computer?
> Beyond what instinct gives us, which is really just a pre defined set of conditionals,
I don't think even neuroscientists have a definite answer to this. However, computer people - as always - do. That's, I think, the problem. A KISS approach is great; being oblivious of complexity isn't.
Pull the cat5 from a critical nfs server and you'll see plenty of psychiatric conditions, both in computers and people. The analogies were a bit tongue in cheek, but I think they relate well enough. Why does a human have an urge to poop? Is it because of "human-ness" or because there's waste that needs to be disposed of.
I don't think computer people think they have the definite answer, it's just easy to apply familiar abstractions to things when they map well. Plus https://en.wikipedia.org/wiki/Fixed_action_pattern seems quite like a conditional to me, and neuroscientists seems to agree. Comparing things at a higher level doesn't mean being oblivious to the lower levels, it just means it's outside the scope you care about. If we could 1:1 replicate every single possible input/output/decision/etc that happens in the human brain with transistors and have the exact same properties as a human when interacting with it, does it really matter if the path to get to those interactions is different? To me, it doesn't. We don't define people by which neurons fire in their brain, we define them by the actions and decisions those chemical reactions cause, and I don't see why it should be any different if they're made of silicon.
It's a nice notion but I've yet to see an existence proof that a machine can do that. It's not about just passing the Turing test and fooling humans, I'm sure that will happen, but I mean the more nuanced qualities of how the machine fails, how it is able to tweak and be more thoughtful of the situation. When NLP / smart text systems make a tiny mistake, it's quite egregious / silly and we can immediately detect it (but since the machine lacks knowledge of social constructs, physics of the world, idioms, and other idiosyncrasies that we have, it doesn't know what it doesn't know).
Responding to humans using knowledge accumulated by organically living in a society for years, is different than generating a body of text.
No one says it is impossible.
It is more a question of whether our current efforts are enough to get us to the point.
Can we get to another galaxy? Probably, if we have enough resources and we discover enough to do it, and we actually go through with it. But it can seem unlikely, even given an unlimited amount of time.
Not to mention that we as a species may be facing a severe catastrophe in the future as a result of how we have populated this planet in the last 200 years. Hopefully we can avoid it, but once again are you sure?
"Algorithms that can write heartfelt posts like yours will be developed one day."
I don't find it hard to assume that algorithms will one day be able to write posts that an observer may classify as seeming to be heartfelt.
But that doesn't mean that the machine actually experienced anything in writing it. That it's "heartfelt"-seeming post was actually felt by the algorithm at all.
"This is literally the point of Turing's thought experiment: if a machine gives you answers indistinguishable from a human, is it alive?"
Turing assumed it was. I think the issues are orthogonal. If it is indistinguishable by a human, all that means that I can see is that it's a very effective mimic, not that it has awareness.
And how do you know that a human experienced anything while writing a "heartfelt-seeming" post?
Do you think that everyone reading that experienced it as heartfelt?
Did you have the "awareness" that it was heartfelt from birth, or did you only know that because you can recognize that it's much like other writing described as "heartfelt", that is, is it something you learned?
If you did in fact learn it, which I assume you did, since we're not born knowing language or even what heartfelt means, why do think that an AI couldn't learn the same? Do you think that how you "experience" it is universal to all humans? Or does it come from things you've encountered in the past.
Regarding your argument about being aware of the color yellow upthread, have you asked someone that is red-green colorblind how they tell the difference or how they experience red and green? It's clearly not the same as people who can see the full "visible spectrum", so do they not have awareness of those colors?
In the past when I asked a friend of mine who is red-green colorblind how he can distinguish the difference, he told me it was because he'd seen things people call red or green in the past. I asked if it was the same as the difference between colors, he said yes, but more like the difference between teal and a slightly different teal. So if you were never told what yellow is, would you be aware of it, or just know it as a slightly different orange? Personally, I know what colors are only because I know what other things of the same color look like, the same wavelength hits my retina so I know that it's the same color. How is this any different from an AI recognizing that?
"And how do you know that a human experienced anything while writing a "heartfelt-seeming" post?"
Because I wrote that original post, and I had experiences while writing it. I know that from direct observation and I don't need other evidence.
"Regarding your argument about being aware of the color yellow upthread, have you asked someone that is red-green colorblind how they tell the difference or how they experience red and green? It's clearly not the same as people who can see the full "visible spectrum", so do they not have awareness of those colors??"
Yes, I was wondering if someone would bring that up, but my point wasn't that they have the exact same experience of yellow as me, but rather that they have an experience of a color (or sound, or taste, or anything) at all. And sure, some people experience fewer colors. That doesn't change the argument.
"I asked if it was the same as the difference between colors, he said yes, but more like the difference between teal and a slightly different teal." Interesting!
> Because I wrote that original post, and I had experiences while writing it.
Well I said "a" heartfelt seeming post, not that specific one. If an AI told me the same thing, what difference would that make? I have nothing but the word of the author. How do I know you're not just an AI, and does it make any difference if you used neurons or transistors to come up with the concepts expressed? To me, the difference is meaningless.
> but rather that they have an experience of a color
But what defines that experience? Is it not just a wavelength of light hitting your eye, and your brain interpreting it?
I also find it interesting that you addressed the questions that were easy to refute, and left most of them unanswered. I am genuinely curious on your thoughts, as I find your conclusions as mystifying as you find mine.
EDIT:
> And sure, some people experience fewer colors.
Ah, but do they experience fewer colors, or do they experience them differently? Since the wavelength of a color isn't what defines experiencing it, then what does? And surely you can't say they don't experience it just because of the wavelength.
> So if you were never told what yellow is, would you be aware of it, or just know it as a slightly different orange?
This question was asked and answered concerning blue and green:
"The Himba tribe from northern Namibia, for instance, does not classify green and blue separately, the way Westerners do, but it does differentiate among various shades of what we call green. And when tested, members of the tribe, who are likely to have trouble with blue-green distinctions that most Westerners make easily, readily distinguish among greens that tend to look the same to Western eyes."
Yes. I'm not extremely well-read in the literature but I have read some of it. I'm aware of the term "qualia" but was trying to use the most direct, simple possible language to communicate.
> But that doesn't prove it has any experience of the color yellow; it merely means it can mimic the output of an entity that does.
Which is all we might be doing with each other as well. That awareness could merely be an illusion and there's even some evidence to think it is. Experiments show your brain makes a decision before you are consciously aware of it despite you thinking you made the decision consciously. That leads to the conclusion that awareness is simply a lie your brains tells itself to feel in control.
How do you know other humans can truly experience "yellowness"? What if we're all just pretending and you're the only one?
Let's not talk about realistic AI for a moment, let's talk about hypothetically building a human protein by protein (or whatever building block you're comfortable with saying doesn't have qualia on its own). We're going to write the computer program to build this human. Nanobots will carry it out, every part of the process is understood and nothing is left for biology to do for us. We're just putting together lego blocks and at the end is a human indistinguishable from any other human.
Does this person experience qualia? How would you know? Apparently you can't just ask, since this isn't a valid thing to do with a computer (it's not really experiencing qualia! It's just saying that!).
Let's just say you come up with some objective qualia test. Now we can build a slightly more deficient human protein by protein and test it for qualia. In fact, we can do a qualia binary search and...
Once you take apart the argument like this, it's easy to see that arguments that computers can't experience qualia are arguments against materialism or reductionism, or are patent attempts to sneak souls and other woo into the discussion.
I believe his point is, such a process wouldn't ever be feasible if we can't "create" awareness. You'd have a sack of meat that would not interact. Or if it would, it'd be a vegetable capable of computation but little else (that's my random assumption).
If you forget your fear of "soul" and other terms not ratified by science journals, you'll find out there are plenty of theories in metaphysics and spirituality on what it means to be and what awareness is. It requires you to at the very least read those philosophical/theological concepts instead of dismissing it as "woo". Cattle dwellers in India wrote many things about it eons ago. Joseph Campbell's work is a nice place to start.
> you'll find out there are plenty of theories in metaphysics and spirituality on what it means to be and what awareness is
I have encountered these theories. In the past I have been under the sway of them. They can be interesting to think about, but without any evidence for them there isn't any reason to give them particular consideration. (I am including rational, mathematical, or logical arguments based on experimental evidence as evidence here)
Without evidence, how do we pick which of these myriad theories to believe? How do we decide what the actual answer is? Science and materialism have repeatedly produced predictions that pan out in experiments. Theories of souls and sprituality have repeatedly had to retreat from testable predictions.
To take another tack, how did these spiritual theories come about? If they're true, then the ancient theorist must have seen some phenomena in the real world, and extrapolated correctly from that evidence. If that's the case, these ancient theorists were unparalleled geniuses, but we don't need to worry about ancient spiritual theories themselves, we'll rediscover them in time through our plodding (but safe) scientific method. If we don't rediscover them, then they were interesting theories that didn't pan out.
"They can be interesting to think about, but without any evidence for them there isn't any reason to give them particular consideration. (I am including rational, mathematical, or logical arguments based on experimental evidence as evidence here)"
I'm glad we're discussing this because you're an example of the kind of thinking that is a mystery to me.
The fact that I have qualia when looking at, for example, the color yellow, is directly knowable by means of just a little introspection. This is not "rational, mathematical, or logical" or based on "experimental evidence". It is simply direct observation. Observation is where the scientific processes you name start. Observation does not depend on those processes. They depend on it and build from it. Observations are the most fundamental thing.
From the direct observation of your experience of yellowness, which I think you must agree is something more than merely classifying the wavelength as being in a certain range, I would think you would know that qualia exist. That is enough evidence.
"How do we decide what the actual answer is? Science and materialism have repeatedly produced predictions that pan out in experiments. Theories of souls and sprituality have repeatedly had to retreat from testable predictions."
So, let's not have a theory of souls or spirituality at this point. We don't know enough. That doesn't mean that those things don't exist, it only means we don't understand them.
Suppose a cave man-level creature on another planet encounters an iPhone because we somehow send one there. This cave man creature knows how to build fires and has used reasoning to make weapons, etc. In doing those things, he probably tried different ideas about how best to make fires and rejected the ones that didn't work well. In other words, he has a rudimentary understanding of the scientific method.
Now he tries to figure out how an iPhone works. He constructs theories based on what he knows. The theories are all shown, over hundreds of years, to be wrong. Therefore, he is told, iPhones don't exist because when he constructs theories about them, they are wrong.
But he directly observes the existence of the iPhone because he holds it in his hand. He knows they exist. He doesn't need a scientific theory to prove that they exist. Maybe, eventually, he will have a correct scientific theory as to how they work; it will just take more time. Or maybe he never will. Either way, iPhones exist.
Awareness is, in my view, exactly the same, and it is mystifying to me that people such as yourself come to a different conclusion.
Oh, I'm not saying qualia doesn't exist. Our experiences are definitely an actual phenomenon. And we can't currently explain exactly how it works and we should try to find out what produces the experience of qualia.
What we shouldn't do, is say humans have qualia and jump to the conclusion that machines can't have qualia, and can't be "aware" in the same sense a human can be. I apologize if that wasn't your intended argument.
OK, good, so we have in common that we both believe qualia exist. I didn't get that from your preceding posts.
"What we shouldn't do, is say humans have qualia and jump to the conclusion that machines can't have qualia, and can't be "aware" in the same sense a human can be. I apologize if that wasn't your intended argument."
No, we can't make that jump. But my intuition as a programmer (and one who is interested in A.I. and does A.I. in the sense that some of my mathematical ideas were incorporated into a number of spam filter products) is that they don't. There is simply nothing there that would "add" awareness to the algorithm. All there is is information processing. It can classify something as either having the attribute of yellowness or not. But something beyond that functionality is required to have it experience the qualia of yellowness.
My point is that we have no idea what that something extra is. Somehow, people have it. But we don't know how. There are hypotheses, like those of Roger Penrose who discuss quantum phenomena as possibly being behind it, as made manifest in certain microscope structures in the brain. Suppose Penrose is right. Then, maybe one day a quantum computer can incorporate the same phenomena and be conscious for the same reasons we are.
But I do not believe that, for example, a spam filter as we know it today has the slightest experience of qualia. There is just no "room" for it. Everything in the spam filter has a specific function for the classification task and has nothing to do with experiencing qualia. The same is true for the most complicated deep learning network.
Maybe software as we know it can induce qualia as some kind of an emergent phenomenon, and the brain also induces it through some kind of emergent phenomenon. But I tend to hypothesize there would be something that causes that, some mechanism that the emergent phenomenon grows out of, that is not embodied in something like a current-day deep learning network.
My hypothesis is that to create a machine that experiences qualia, we will first understand the mechanism behind our own experience of qualia, and then reproduce that mechanism in a machine. I hypothesize that this would be a new and extremely fundamental, and transforming, scientific discovery.
And if we do reproduce that mechanism in a machine, and THEN that machine passes the Turing Test without us programming mechanisms to merely MIMIC human behavior, then I'll tend to assume that we've created a truly conscious entity -- though we may never be able to completely prove it.
On the other hand, there's no reason to assume we'll ever discover how it works -- for instance, we can't rule out the idea that there is a God who doles out souls to human bodies, and that we'll never have the means to understand God or souls. I don't believe that's the case -- but plenty of people do, and we can't presently prove they're wrong.
I don't disagree except that in order to produce a sufficiently human-like intelligence that can pass the turing test, we'll most likely need to have pulled the human experience apart enough to understand where the experience of qualia comes from. I think it's hugely unlikely we'll be able to create an intelligence that can carry on a conversation that can fool any human, and still not have a single clue what qualia is or where it comes from. We are a long way from that conversationalist, so it's not surprising we have no good understanding of qualia.
>My hypothesis is that to create a machine that experiences qualia, we will first understand the mechanism behind our own experience of qualia, and then reproduce that mechanism in a machine.
From my experience with machine learning I think that we (humans) dont need to understand a problem to be able to build a tool (AI) which can solve that problem. The necessity of understanding problems in order to solve them is obsolete :-P
> Now he tries to figure out how an iPhone works. He constructs theories based on what he knows. The theories are all shown, over hundreds of years, to be wrong. Therefore, he is told, iPhones don't exist because when he constructs theories about them, they are wrong.
But he directly observes the existence of the iPhone because he holds it in his hand. He knows they exist. He doesn't need a scientific theory to prove that they exist.
Yes, that's the point. He has "experimental evidence" (aka something observable) right there in his hand. Understanding of how it works is immaterial for the question of existence (if you're not a constructivist :) So then we use that as analogy to look back at the question at hand:
> From the direct observation of your experience of yellowness, which I think you must agree is something more than merely classifying the wavelength as being in a certain range, I would think you would know that qualia exist. That is enough evidence.
Enough evidence of what? Certainly evidence of internal thoughts and feelings. I don't see how you can demonstrate anything beyond that, with or without an understand of how that "thing" might work.
> If you forget your fear of "soul" and other terms not ratified by science journals, you'll find out there are plenty of theories in metaphysics and spirituality on what it means to be and what awareness is. It requires you to at the very least read those philosophical/theological concepts instead of dismissing it as "woo".
I've read them. They are, from a scientific point-of-view, all woo. Almost by definition. They necessarily appeal -- if they can be epistemologically justified at all -- to a non-utilitarian epistemology. Science is based on a radically utilitarian epistemology: claims are justified to the extent that they are demonstrably useful in explaining and predicting observed phenomena. Theories -- in the non-scientific sense, obviously -- that involve entities that produce results in the physical universe but which are not governed by physical law or controlled by physical events as they are conjectured to be outside of the physical universe -- are non-utilitarian. They are irrelevant to any predictive model of observed behavior, and it is impossible to distinguish by observation between the existence of such non-physical, uncaused causes and processes that are purely physical. Either (a) there exists an adequate scientific model of the process, in which case the model being an accurate model of an underlying physical process is indistinguishable from the uncaused cause merely mimicking the behavior of the model at the times it has been observed, or (b) there exists no adequate scientific model, in which case the physical process not yet being discovered is indistinguishable from a non-physical cause.
That's not to say that such conjectures cannot be true, just that they can never be validated or distinguished from either known or unknown physical processes.
Say we create a machine that behaves like a human to all outward measures. There is no way to distinguish between:
(1) Human and machine alike being purely physical,
(2) Humans having metaphysical features like a "soul" and internal awareness and machine simply mimicking behavior,
(3) Humans and machines alike having metaphysical features like a "soul" (perhaps because God chose to reward human ingenuity by creating a "soul" for the machine.)
Since the context wasn't moral theory, the special definition of "utilitarian" has in the context of moral philosophy wasn't intended; "utilitarian" has a general sense outside of that context which is what was intended.
I avoided "Pragmatist" specifically because I'm familiar enough to know it refers to a philosophical school of thought that touches on the epistemological questions I was referring to, but not quite certain that the exact position I was referring to was within its scope (I knew it was along the same general lines.)
And yet there is something unbelievably fascinating about being conscious. Many of our words, even existence itself, are predicated on it.
Consider this - if conscious observers didn't exist, in what sense would anything else exist?
If I told you that right now there is a universe out there which you can never detect nor deduce its existence from any evidence in the one you're exposed to. You won't be able to verify its existence in your lifetime. But, I tell you, it EXISTS! What doss that mean?
If you say it's meaningless, you'd be onto something. Furthermore if I told you that, if you wash your hands you'll be killing 10 innocent little bunnies in that universe. Will you stop washing your hands? No. You will err on the side of that undetectabld universe and bunnies not existing - after all how have I come to possess this knowledge if you don't?
If anything, you'd say that universe can only exist in the sense that some conscious observer in the future will VERIFY ITS EXISTENCE.
Now, if you subscribe to atheism, then in order to even build a machine, you need an intelligent animal that arose organically. So, in what sense would OUR universe exist if humans never arose?
YOU SEE even the concept of existence is tied into our subjective experience! It is the dual of it and equally mysterious. Why is there something rather than nothing, indeed?
> if conscious observers didn't exist, in what sense would anything else exist?
> [A] universe can only exist in the sense that some conscious observer in the future will VERIFY ITS EXISTENCE.
> So, in what sense would OUR universe exist if humans never arose?
I think things exist independently of whether anyone is observing them. That's because "observing" is just a description of a bunch of particles interacting with each other. Some of those particles are part of a larger pattern of particles we call human brains. There's nothing particularly unique about our brains in terms of physics.
At a higher level, we can look at the interactions between our brains and the environment and think about brains as a different kind of thing. A pattern recognizer, a prediction maker, an effective evolutionary adaptation that leads to the propagation of the genes that create it. But there's no reason to suspect it creates universes.
I absolutely agree that a lot about our brains is mysterious and there are a lot of questions that need answers. The next few decades are going to be very interesting as we discover many of those secrets. I just don't think we'll discover that among those secrets is the ability to make universes real or not.
Certainly, we can only observe the universe we are in and phenomena that are within our light cone. But that's a limitation on us, not on universes.
What you're saying is that there's no difference between something that exists and is unobserved by a conscious mind, and something that doesn't exist. I'm talking about the definition of existence that makes the sentence before this one make sense. It's true there are no observational differences between the two, but nevertheless there is a conceptual difference.
To illustrate the difference, look at the universe before intelligence evolved. The most reasonable scenario is that this universe existed, even though there were no observers. Something was going on before anyone was looking, because that process led to the creation of intelligence.
It's possible that the universe didn't truly exist until consciousness existed to be aware of it. That it popped into existence once consciousness did, and everything just happened to be consistent with a universe that already existed beforehand for billions of years. But that's a pretty surprising theory. It's much more likely the universe just already existed.
That's not what I am saying. You and I think this universe existed before humans arose because we are aware that it exists now, and that gives us a good starting point for extrapolating. If that last part wasn't there -- eg about some other universe in a multiverse -- we wouldn't know what it means for it to "exist".
Or do we? You'd have to define the word "exist" in some terms that relate it to things we are familiar with. Otherwise it is meaningless. That's my point. Trying to define existence is like trying to explain consciousness. If you think consciousness explained by souls is "woo" perhaps you should consider the nature of your concept of existence! You take it for granted and use it in a sentence without really exploring its definition.
"How do you know other humans can truly experience "yellowness"? What if we're all just pretending and you're the only one?"
To me, that's kind of a silly question. Yes, it might be true. Of course, it might be that there are no other human beings and I'm just a brain in a vat. I just find that hypothesis to be too unlikely to care about.
"We're just putting together lego blocks and at the end is a human indistinguishable from any other human.
Does this person experience qualia?"
I would guess so.
"How would you know?" For me, this is a little like your first question. I wouldn't necessarily know. But if this entity is physically indistinguishable from a human at even the most infinitesimal level of physical reality, I'd assume it experiences.
But, I may be wrong here, because it might be that there is something non-physical called a "soul" that resides in humans, and a physically identical entity might not be provided with one by God or whoever/whatever the soul-provider is. That's not what I tend to assume, but I have no evidence either way.
Your question "How would you know" is an interesting one because it assumes that it matters whether or not an observer knows. Maybe some entities have awareness and others don't, and an observer can't tell the difference. So what? That wouldn't imply, in any way, that the phenomenon of awareness doesn't exist.
"Once you take apart the argument like this, it's easy to see that arguments that computers can't experience qualia are arguments against materialism or reductionism, or are patent attempts to sneak souls and other woo into the discussion."
That doesn't mean that such arguments/hypotheses aren't true. It just means they may not be testable using the scientific method. Much as many people don't like the multiverse theory because it may not be testable using the scientific method. But that merely means that there are limits to the truths that can be discovered by the scientific method; it doesn't change the reality of what is true.
It's a bit reminiscent of Godel's completeness theorem. Many great mathematicians assumed that there were no truths that weren't provable within the systems they were studying; Godel proved that such assumptions were simply wrong.
I think we're on the same page here. Usually qualia are brought out as a proof that there is some kind of ineffable quality that can't be replicated by machines, but can only be experienced first-hand by truly conscious beings. I see now that that was not your argument, and you're interested in an answer to what produces the feeling of qualia.
> to experience the color yellow, as opposed to classifying a visual stimulus as being of a certain wavelength
I've been wondering about this for a few years, can't find anything about that because it's too generic a search term. Is there a name, or references about this ?
experience the color yellow, as opposed to classifying a visual stimulus as being of a certain wavelength
Purple is the weird one as it doesn't correspond to a wavelength, but can only be represented by a set.
Edit - all colours correspond to sets of sets, if you think about it, as you can experience exactly the same colour for essentially an infinite set of different sets of frequencies.
I once met David Chalmers when he was teaching a philosophy course at NYU. And I asked him after the class ... What can you say about consciousness in the "hard probem of consciousness"? I am having trouble making any statements about it which are objective.
I have begun to suspect that our concepts of consciousness and morality are the result of our language and the way we have defined them. With morality is has become clear to me that personal duties are relative to personal goals. That is, "X should do Y" is incoherent, it really implies "X should do Y if X wants Z to happen" which is what the colloquial understanding is. Always being careful to mention Z removes any feeling of mysticism from morality!!
Perhaps with statements about consciousness, the situation is similar. I couldn't really formulate any sort of mystical question once I was carefully adding the SUBJECT of each sentence explicitly in.
The main question I have about consciousness is not about the processing - that can be explained mechanistically. But about the final result of perceiving this orchestra - why am I in my brain and not anyone else's? What is this extra element that is required for the conscious experience and can we ever be sure a machine we built is really subjectively perceiving anything in that metaphysical sense?
Doesn't seem right - morality is often intuitive or blind. Else why did Socrates say "The unexamined life is not worth living?" Why, because most (almost all) people don't think about Z.
Just because they don't think about or stress Z doesn't mean it isn't there. Sometimes Z is a subconscious brain heuristic developed as a response to competing interests, threats, social consequences etc.
> It seems like a lot of people, including the author, don't even notice that this is a question. They seem to assume that computer-like information processing is all that happens in a human mind, and that awareness itself doesn't even exist as something to ponder.
This seems to be a misjudgement. There are actually plenty of people who have studied traditional philosophy and dualistic approaches to consciousness for years and eventually arrived at a naturalistic conclusion after they've studied neuroscience. I would actually be very much interested in the numbers for the opposite case: How many subscribers to dualism have knowledge in neuroscience, human behavioral biology, the theory of evolution and physics?
My problem with that argument, which I have heard quite often, is that we cannot do anything useful with a term (awareness/experience of yellow in this case) that has neither extension nor intension.
What you experience when you see yellow is what you experience when you see yellow. Nothing more can be said about it and it cannot be compared to anyone elses experience of yellow, human or machine.
So if you make that experience a condition for calling another being intelligent, you are effectively saying that nothing but yourself can ever be intelligent.
You need to break down that thing you call experience of yellow or awareness of yellow and explain what properties it has (intension) or you need to be able to show that it influences the behaviour of humans in a way in which machines cannot be influenced by it (extension).
"So if you make that experience a condition for calling another being intelligent, you are effectively saying that nothing but yourself can ever be intelligent."
I think you are confusing the principles of whether an observer can determine whether an entity has an attribute with the question of whether it does in fact have that attribute. But those are entirely different issues because, for one thing, the observer may simply not have the tools yet to make such a judgement.
Matter is made of atoms. For a long time we did not have the tools to determine that. But that didn't change the fact that it was, and is, true.
Now suppose the human race died out before it ever developed such tools. Obviously, it would still be true.
Now suppose that due to our limited intellectual abilities or for some other reason, it never was, or will be, possible to ever develop such tools. To me, it's obvious that that has no bearing on the truth of the proposition any more than if we didn't have time to develop the tools.
Similarly, the question of whether you have awareness is independent of whether I or any other third party can prove you do.
I don't necessarily disagree with that, but your argument is nothing more than an appeal to believe in the existence of such properties based on how you yourself feel when you look at something yellow. I don't find that very convincing.
But I do agree with you in that the question you asked is worth asking. Not even mentioning the issue is certainly a frequent mistake made by authors taking a purely technological perspective.
"I don't necessarily disagree with that, but your argument is nothing more than an appeal to believe in the existence of such properties based on how you yourself feel when you look at something yellow. I don't find that very convincing."
No... I'm not inviting you to be convinced based on my experience. I'm inviting you to notice how you yourself feel when you yourself look at something yellow. My bet is that you'll experience that which is referred to in the literature as "qualia".
My own experience isn't any more convincing than yours because I don't know what properties are necessary for me to feel the way I feel. I also don't know if my feeling is in any way similar to yours or that of a fly on the wall.
So if someone built a very complex machine that was able to communicate with me like a human, and if that machine claimed to feel something very specific at the sight of yellow, it would not feel right for me to say "no, you're not feeling anything like a human would feel" because I don't know how other humans feel. I have no way of comparing my own feelings with those of other humans or non humans.
Everyone knows qualia exist. The dispute is over whether they're components of the mind's perceptive systems, or whether they're metaphysically queer. The latter position is obviously wrong and requires reams of magical thinking, but being magical, it's more interesting to talk about, so human philosophers, being dumb and untrained humans who don't get the most basic principles of naturalistic wizardry, concentrate on it, to the neglect of the obviously correct and far more useful former line of inquiry, which would actually augment their spellcraft quite a lot.
> But that doesn't prove it has any experience of the color yellow; it merely means it can mimic the output of an entity that does.
I disagree: it would be mimicking the output of an entity that is mimicking....
My point is that these qualia arguments are begging the question, because they posit some 'real' entity that is somehow different from all the 'mimics', despite the fact that no-one is saying what the difference between them is.
The really interesting question is: why do you humans obsess over subjective consciousness rather than general intelligence, when the former is simpler and stupider, while carrying a burden of moral value, while the latter is more versatile and useful, while not carrying undesirable moral weight?
Which one do you really want to build in a computer? The intelligence, plainly.
The author discusses concepts around intelligence without articulating what definition of intelligence he is adopting. When he discusses the innovation in understanding planetary motion, or chemical substances, it represents examples of where scientists could place observed data amongst a common frame of reference (ie, a meaningful definition to contextualizes observation leads the way to a theory which explains the data).
A more meaningful direction for article, imo, would have been to adopt a concrete definition of intelligence, and then talk about how ferrets, chimps, babies and adults fall on this continuum. Perhaps, connectomics, molecular biology, psychology or neuro-anatomy help us approximate the 'atomic number', giving us the opportunity to speculate about the underlying equivalent of quantum mechanics.
Like the article states, evidence is lacking, and we're somewhere between 1-100 nobel prizes away. But, the power of a concrete definition helps point to what we know we don't know, and discussion about whether the definition of intelligence itself needs to be re-framed to answer a question like this.
I enjoy the planetary motion analogy immensely. However, a curious complexity is evident in biological systems - it's physics imbued with 'meaning'. A cell that behaves one way instead of another will die. This is true from e. coli up to skin cells. I suspect this adds a lot of complexity to the ultimate 'algorithm' of intelligence.
In fact, it's probably prudent to stipulate whether the goal is to specify the developmental algorithm that gives rise to intelligence, or intelligence itself. The generating process might be a lot more simple than the finished product, in the same way that the output of a pseudorandom number generator is very complex in terms of entropy (but not kolmogorov complexity, I guess), but the generating algorithm is very simple. Or the Rado graph, which is sorta maximally complex in the sense that it contains all finite and countably infinite graphs as induced subgraphs, yet it has a simple generating scheme. As a last example, consider No Man's Sky - relatively simple algorithm compared to the astonishing complexity of the worlds.
I do believe there is a relatively simple high-level description of neural development. But it's curious how it's relatively robust to genetic manipulations. I remember early in graduate school listening to a lecture about a particular mouse model for autism, caused by just one gene. The lecturer excitedly told us that it had a very high rate of behavioral manipulation - 75% or so. Not coming from a mouse-model background I was astonished that it was 'only' 75%. What happens in the other 25% - the gene doesn't just magically reappear, there are other compensatory mechanisms. What drives those? I suspect that's the more abstract level of description that a simple algorithm might describe.
Since I did not see it in the article or in the comments here: Intelligence is compression / dimensionality reduction. Finding the essential parts of a problem/object/concept and creating a code book for it. The better you can compress things, the better you understand them. If you know that the left wing of a plane is the same as the right wing of a plane, but rotated/flipped, then you would not need to store (redundant) information about both wings. The upper bound to an intelligent action is the Kolmogorov complexity (upper bound to compression) of the problem it aims to solve.
Another imperfect analogy: say someone (maybe from an earlier time) was looking to reverse engineer a modern CPU but could not see what was going on inside it very well. Surely the whole thing would be very confusing. The CPU has many many clever designs based on the needs of people who buy it and use it, but the whole thing is based on how you can make AND/NOT using transistors. Maybe if the person was to discover this, then they could start getting past the complexity of reconstructing some of the mechanisms of the CPU; and they could start making their own CPU for their own needs.
Take it one step further: say person from the past sees OS X running on a computer. They take the CPU and try to reverse-engineer OS X just from the CPU. Good luck.
The NFL theorems have been misused by advocates of 'intelligent design' -- they do not factor against evolution since evolution doesn't demand a global maximum, only a local one. (There are other reasons but this is the most fundamental.) But the theorems themselves are incredibly interesting.
In short they show that averaged over the space of all possible search landscapes, no search algorithm performs better than any other.
Obviously the universe does not provide "all possible search landscapes," since the universe has structure. Some search landscape structures and meta-structures are pathological and rare in nature. But nature does provide a huge diversity of them. The NFL theorem is to me powerful evidence that something as general as human or even animal intelligence must be a superposition of multiple "algorithms" rather than just one.
It also, I think, explains why all single-algorithm or single-approach methods of AI end up being domain specific. Obviously they'll be domain specific -- they only work against search landscapes with a certain structure!
The NFL theorems are a big reason I am a short term AI skeptic. I don't think human-level or beyond AI is impossible, but I think we are quite far from realizing it. I'll become more optimistic when AI researchers start studying biology more closely and deeply, since biological systems are the only existing examples of truly intelligent systems.
Solomonoff induction seems to be an useful innate bias for the current approaches at general intelligence. It's basically a formalization of Occam's razor that states that instead of being completely random, the surrounding environment is mostly lawful and simple, and therefore most of the processes in it can be described by algorithms with short rather than long source code.
Thanks, those look interesting. Will give them a read when I have time. My naive reaction though is: a lawful universe does not necessarily imply well-behaved fitness landscapes. Many lawful and even very simple processes give rise to chaotic and complex results.
175 comments
[ 3.0 ms ] story [ 211 ms ] threadI don't mean to nitpick, but epigenetics is a thing.
Also... how up-to-date on the research literature is this essay supposed to be?
How much of that could you take away and still get a human brain?
Comment at 13:12 [1] https://youtu.be/D3nIIKwwiLc?t=11m43s
http://www.sci-news.com/genetics/article01036.html
70 per cent of protein-coding human genes are related to genes found in the zebrafish (Danio rerio), and 84 per cent of genes known to be associated with human disease have a zebrafish counterpart.
With limited computing power, it's likely that the best algorithms for many different problems (possibly including "intelligence" however you define it) won't be simple, just like the best known algorithms for integer multiplication [2] aren't simple. In particular, AIXI variants will be hard to approximate, precisely because they dovetail over all possible algorithms.
That said, it's very likely that the best algorithms for intelligence will be simpler and faster than humans, because humans are the stupidest possible creatures that can build a civilization (otherwise it would've happened earlier in our evolution).
[1] http://www.hutter1.net/ai/paixi.htm [2] https://en.wikipedia.org/wiki/F%C3%BCrer%27s_algorithm
Technically, AIXI_{tl} solves all problems at least as well as any other algorithm in a given order of asymptotic time and length, those being the t and l. The problem is that it has a "trivial additive constant" which will make it take longer than the lifetimes of many stars to achieve its much-vaunted "optimal" asymptotic performance.
In short, AIXI is the most utterly brute-force kind of "intelligence" you could possibly build, whose notion of "scaled down" bounded-rational inference is still astronomically intractable. It basically just amounts to throwing computing power at the problem, more computing power than anyone actually has.
Mind, Schmidhuber and Hutter still get massive props for managing to cut out the philoso-wank that usually accompanies discussions of cognition and instead saying, "Let's just pose a very general inference problem and write an algorithm that solves it."
Having to rely on playing an arcade game is a weak case in general. We don't call Counter-Strike modders AI researchers. The fear angle is nothing but woo, but I'd like to hear you elaborate nonetheless.
[1] https://github.com/moridinamael/mc-aixi/blob/master/src/pacm...
See, chessboards are meaningless outside of human experience. You only know how to play pac man without explicit instructions because you have a general concept of a game and what games look like when you win or lose them. Whenever you try to solve an AI problem in a narrow domain like this then you necessarily have to compensate by giving it a miniature model of the specific problem being solved.
Humans after all don't just intuitively know the rules of games before playing them. You also cannot just describe the rules of the game to the AI because the AI does not understand language.
MC-AIXI seems to be more generic because it factors this out into a reinforcement learner, but this in of itself is hardly a cause for fear and that AGI is coming, as the GP alluded. I'd wager today's AAA titles have some damned complicated heuristics of their own.
This would be true if intelligence were the only requirement of building a civilization.
It's entirely possible there is some fish at the bottom of the ocean smarter than humans. But it's hard to build a civilization using fins.
Many insects have quite elaborate 'civilizations' or at least colonies, and it's doubtful that each individual bug is very intelligent.
On the other hand, Neanderthal had quite a bigger brain than us; but fewer of the social impulses. We're likely dumber but more socialized, which reinforces your theme I guess!
The main class of problems it fails to solve are ones where it has to model itself as an agent in the world.
For example, AIXI is a bit accidentally-suicidal. Its world models have no term for "I am here"; e.g. it will fail to realize that disconnecting the power to save money would stop the inference process instead of informing it. It needs many many examples of dying in order to learn to avoid it, but you only get one chance. (Evolution worked around this problem by making copies, and preferentially keeping copies that were less accidentally suicidal.)
1: http://lesswrong.com/lw/jg1/solomonoff_cartesianism/
Kind of a tangent, but there are conditions that are not easily described as "intelligence" that are necessary for building civilization. Namely, the ability to cooperate in very large groups. Chimpanzees outperform human 4 year-olds in a variety of intelligence tests, but adult chimpanzees are not observed to cooperate to the same degree as humans, or in very large groups. There is also some archaeological evidence to suggest that this is one of the reasons humans were able to outcompete Neaderthals - larger tribes. Individual Neanderthals may well have had higher IQs than their contemporary human rivals.
He said general intelligence is 'as long as you provide a goal, the intelligence should optimize over that goal based on experience'.
But my problem is: general intelligence is supposed to come up with its own goals.
Take for a example a couch potato who has some steady stream of monthly stipend, and sits in front of tv all day. This person is in possession of a functioning general intelligence yet has no goal. One day this person decides to assign to self a goal of going out and getting a degree and so on and so forth. That's also part of that person's general intelligence.
It's not clear to me that AIXI is self-driven in that sense. It would be great if someone could comment on that.
For this reason, I suspect that invoking AIXItl is begging the question. Positing unlimited computing power allows one to sidestep the difficult question: which algorithm(s)? In fact, this is not just a difficult question, it is the question (the one that AI research is trying to answer.) Absent the convenience of unlimited computing power, I suspect that AIXItl is a dead end, no matter how elegant it is.
A multi-cellular embodied intelligence can acquire information about the patterns of the physical world by growing in it - the code is compact because much of the necessary information can be acquired during embodied growth.
A disembodied intelligence such as AIs can't rely on external storage and acquisition of information during a cellular growth stage, and thus must likely be considerably more complex from the get-go.
"Living systems are cognitive systems, and living as a process is a process of cognition. This statement is valid for all organisms, with or without a nervous system." https://en.wikipedia.org/wiki/Santiago_Theory_of_Cognition
If this were a correct interpretation, even Bacteria could be seen as already incredibly intelligent. Going Zero to One (dead matter -> prokaryotes) is a significantly bigger step than going One to Five (prokaryotes -> humans).
https://en.wikipedia.org/wiki/Moravec%27s_paradox
Kind of renders all of his information theoretic calculations moot.
I do, however, agree with your second sentence, and I think the paradox is also weakened by the fact that when 'high-level' mental tasks are computerized (chess-playing, for example), it is often through brute-force calculation that is not a model of how humans use their general intelligence to perform the task.
The stuff in the article isn't information theory. Information theory, in the senses of Shannon and Chaitin, is actually very useful for studying cognition, because it tells us about how much we can try to learn, via what methods, from noisy sensory data.
> Chimps are remarkable thinkers in their own right. Maybe the key to intelligence lies mostly in the mental abilities (and genetic information) that chimps and humans have in common. If this is correct, then human brains might be just a minor upgrade to chimpanzee brains, at least in terms of the complexity of the underlying principles.
Of course this still kills the argument that it's at most 125MiB in my opinion. Moravec's paradox [1] implies a lot of the actually-hard tasks related to inference and learning are already solved in chimps, because they can see and hear and balance just fine. I think if we actually understood how chimps reasoned, in high-level and low-level terms, then we could easily tweak and improve that process to function beyond human level.
1: https://en.wikipedia.org/wiki/Moravec%27s_paradox
It's as if the conscious arithmetic runs on some kind of stack of virtualization layers, using millions of neurons to build a mental image of the concept of "the number five", whereas a handful of neurons would suffice to do the actual arithmetic.
I often wonder if those rare people who can do calculations in their head faster than calculators, may have - unconsciously - found a way to unlock the native hardware of their brain, bypassing all of the symbol abstractions required by us muggles. A bit like GPGPU vs CPU for certain algorithms - only orders of magnitude more pronounced.
It's like being giving a bunch of Brainfuck source--if you'd never heard of Brainfuck.
Not really. The translation machinery is also defined in the DNA. Granted, there is a bootstrapping issue. The chimp argument seemed silly to me. Why not just estimate so fraction of the human genome as defining the brain? In that case 3G base pairs is a bit under one gigabyte of information defining an entire human. Still small in comparison to the quantity of synapses.
For example, if executing the DNA instructions produces a human female, she'll be born with a lifetime supply of eggs, the developmental instructions for which were part of her startup DNA. Those instructions were definitely not part of the "cell soup", as is shown by mitochondrial donation assisted reproduction technology (aka 'three-parent babies').
Cells are factories, incredibly complex ones. DNA is a set of punches in a tape.
Why yes, yes it is. There are genes that code the proteins that make up the ribosome for example. The ribosomes in turn interpret the DNA string to make proteins. This leads to a bit of a chicken and egg problem which I mentioned as the issue of bootstrapping. But the DNA does code everything in the cell, or at least defines the machinery needed to make everything.
Not sure why you think "embodied growth" is essential here; having a compact code base that can be supplemented by information during maturation obviously has some utility, but why would you think embodied growth is critical to it?
> A disembodied intelligence such as AIs can't rely on external storage and acquisition of information during a cellular growth stage
AIs aren't disembodied. They have physical substrates, just like natural intelligences.
I don't see why something like cellular growth is necessary for "external storage and acquisition of information" during development. Even if it were, however, just as natural multicellular organisms are coordinated colonies of self-reproducing cells, artificial organisms (intelligent or otherwise) could be coordinated colonies of self-reproducing machines.
AIs are not disembodied; they obviously are embodied in physical machines. And there's no reason an AI couldn't acquire information during development,
The argument isn't that that embodiment is necessary for the artificial system to learn how to operate in a way that's as capable or better, just that there's a whole lot of information there not accounted for in the genetic code alone.
It turns out that Newton's ideas on gravity are a mathematical approximation that at least ignored relativistic effects. See: http://physics.stackexchange.com/questions/52165/newtonian-g...
Newton's gravity was a theory about a single fundamental force - and it turned out to be not 100% accurate. Rather than be an example of how intelligence might be similarly possible to explain with a simple theory, I think it offers a perfect counter-example of why it almost certainly cannot.
I almost feel like you're replying to someone else because I never said what you are contradicting.
I merely said that Newton's gravity is a simplification that doesn't take all variables into account. So to point to Newton's gravity as an example of, "Sometimes simple things can be right!" actually makes it more of a counter-example.
The author's argument is that our intelligence is likely not solely possible by the interaction of a large number of separate principles, but rather by some simple principle that manifests itself in different ways. I think if we find some simple principle that explains intelligence, and then later have to make a correction to the math of that principle, the resulting explanation will still be simpler than the idea that intelligence is made up of many different principles all of which have to be present.
But if I told you guys to pair program an AGI, you could probably finish in a day or so.
A simple algorithm is a matter of finding just one or two basic discoveries. It could happen tomorrow! It probably won't, but that's the nature of discovery.
For example, I'd reckon we'll find that a lot of things that today only make sense to solve with a neural net are actually perfectly representable using a simple(ish) algorithm, but it's unlikely we'll find that out until after we create an artificial general intelligence and know more about the problem of intelligence.
In other words, we'll solve the problem, and then realise how much easier it could have been.
Yes, of course it is possible, if we change the meaning of "intelligent" to fit what a computer does. :)
Generally, I think the problem here - as in many other situations - starts with the lack of rigor that computer science has when "naming" fields or, most exactly, appropriating concepts from other fields.
* Computer Vision: "vision" has a defined meaning, even "organic-related". Signal processing and a bunch of algorithms aren't exactly what vision is and what we define as "vision";
* Machine learning: does the computer really learns, as understood by the common definition of "learning"?
* Context awareness: if we consider common definitions, neither a few sensor readings are "context", nor acting according to said readings is "awareness".
This might sound like nitpicking, but I truly think it's a real problem. Talking about vision, learning or intelligence put's the stake at what people or the general scientific community commonly perceive as those concepts.
What's most frustrating to me is the contrast. Computer science research is thorough; why are we not thorough on naming our fields of study? Why do we appropriate concepts by "analogy", knowingly that analogies aren't truly exact?
I don't think that's what he's doing. Otherwise he could just define intelligence to mean what computers do now. Clearly he has at least roughly the same idea of intelligence you or I do.
> "I believe it's not in serious doubt that an intelligent computer is possible"
Do you have an objection to this specific statement when taken to mean general human-like intelligence being exhibited by computers? I have not heard anyone defend the opposite viewpoint often, and I'm interested if there are any compelling arguments why it's not possible.
A few aspects which relate and shape intelligence:
1) Socialization; 2) Embodiment (the fact that I have a body and the body has feedback loops - affects and is affected by cognition); 3) Emotions (we have "hunches" and those hunches are surprisingly many times adequate); 4) Language (our sign system affects our perception of the world).
Will a computer be able to be a sociable, embodied person? I tend to think that it won't, so that's roughly why I believe a computer won't ever have "intelligence" (unless, again, we redefine intelligence).
There's not enough energy to simulate human intelligence at the same speed that humans think, without just essentially producing biological computers.
And that's more of bio-engineering than anything relating to computing.
No, computer vision researchers haven't achieved what you and I perceive as vision, no more than a biochemical research in cancer cures have found a cure to cancer.
But I would argue that these lofty named fields are a good thing. Academia is already plagued by it's uncertainty and lack of an inherent "end game." Let's not make it worse by renaming machine learning to "computer pattern recognition." Pattern recognition is just a possible piece of the puzzle or step in the process to creating AI, not the actual holy grail of achievement.
If you use e.g. "vision" in your research as a basic concept, then you should do a thorough state of the art inquiry and state exactly what you understand by it (using, of course, the state of the art). I don't believe this was done when someone thought of naming these fields as they did. I also don't think that beginning researchers do this basic "background" check; they try to achieve "vision" (or "intelligence") without a clear (and at least commonly accepted) definition of "vision" (or "intelligence").
As I said above, I can see your point, but I wouldn't do this in the scientific community. Nothing wrong about saying in the news that you want to make a computer see, but e.g. having a journal with a misdefined or malappropriated concept in its name is - IMHO - wrong.
Physicists (and this guy is one of them) certainly do have a robust set of taboos against considering consciousness as a quantum phenomena (whatever that might mean.) It's unfortunate because there is a huge resurgence going on in the physics of quantum many-body entanglement, quantum computers, foundations of quantum, etc. Alot of this meandering was just sidelined in the 1940s when world affairs intervened. However, in light of these new discoveries it's time to revisit the old vitalist arguments. Imho.
I would like to point out though, that this is a comparison between a few years of tinkering in the lab with billions of years of evolution.
Specifically, if we follow the evidence, and find that somehow, beyond all odds, the vitalists were correct (once you add "quantumness!" to the mix), then it's clearly only by lucky guess they were right. It would be something we could look back at and say "wow, that's an interesting coincidence" rather than "wow, those vitalists had it right all along, we should have paid closer attention".
Science is how we separate the correct, the mildly wrong, and the wildly wrong intuitions.
Its not an alternative to intuition -- intuition is where hypotheses come from. (OTOH, vitalism isn't a hypothesis, simply an intuition that some aspects of living things are outside the domain of science. Its not testable -- if correct, its completely indistinguishable from some of the relevant aspects of living things simply not being sufficiently well modeled by the current state of science at any given time. So, its an intuition that is neither confirmable nor useful in any way.)
And you can't say vitalists were right, they've made no predictions: calling something magic woo woo isn't a hypothesis that can be proven right.
Logic.
> Can't be random; there are billions of wrong ideas and only 1 right one.
It's not random, it's deductions based on previously known truths.
> So some weight has to be given to intuitions, and some props to those who had the right ones. Because that's how progress gets made.
Props are given to those people, they're called scientists. There are no correct intuitions from the vitalists, they're claim is magic... that's not in any way useful.
Correct hypotheses are, in a sense, a product of fortune, and scientific progress requires them to happen sometimes, so, in a sense, that's true. Without a certain minimal degree of luck, you would have no scientific progress.
The scientific method can be viewed as simply a method of filtering out the intuitions without utility and promoting those (if any) that have utility. It is a way to avoid wasting the good fortune of the intuition of forming useful hypothesis, and avoiding the waste demonstrably fruitless intuitions.
> some credit has to be given to those that are right
But vitalism is so vague as to be untestable, and justifies no useful models. "Some (unspecified elements of) living things are not governed by physical law" is indistinguishable in effect from "We have not yet discovered the physical laws governing (some elements of) living things".
A minor disagreement on this point. I think often our intuitions are shaped by experience, and lead us to conclusions that we don't have a rational line of inference to (at least one we can explain). Science acts as a good filter, saying "we don't care how you came up with it, test it and we'll agree if it turns out it's right". This isn't a random guess, but it's far from reasoning you could convince anyone else with. This is different from a lucky guess. Your brain has been seeing a lot of the data, and is picking up on patterns, in other words your intuition is correlated with the phenomena, and that's what leads you to the theory you want to test.
Second, sometimes purely rational arguments lead us to theories from other data. This isn't random chance, but it is exploiting regularities in the world around us to predict new things. For example, kepler's laws of motion weren't the direct result of observation. You can't observe equations that guide the motion of the planets. In between observations, the planets might have jumped all over the place. But it was assumed that the planets weren't doing that, and simple equations were inferred from the data available that fit the evidence. There is a step here between "observation" and "prediction" that doesn't involve a lucky guess.
They really don't, but they do have taboos again ignoring the fact that there's no evidence for such a hypothesis, as do all scientists. Science is empirical and scientists don't generally just like to make shit up because it sounds cool.
(The reason is very simple, namely that the rules for the universe are probably very simple.)
Our planet is a speck of dust rotating at a fortunate distance around another speck of (shining) dust. By mass, the majority of the planet is useless to life. Everything alive on Earth exists in a very thin skin between lifeless rock and lifeless, airless, freezing space.
Then, on any timescale meaningful to "The Universe" individuals are tiny blips. Species are around longer, but not much. Vast majority of species are killed off by cosmic events regularly (we are on something like the 6th major extinction event, IIRC).
All of humanities struggles with itself, with ignorance, all of it's artistic expression, have taken place within the context of that thin skin of warm fluid surrounding Earth over a shockingly short time frame.
The best that can be said is that we now know the score: that "The Universe" is fascinating, vast, hostile, empty, silent, and our entire species is the rough equivalent not to a mewling baby, but rather to, perhaps, a single microscopic insect gestating in its egg, just beginning to break the shell and look around.
Most matter that you interact with is not in your body, and much of it is not in a usable form to your body, yet it still sustains you. Why should this not hold true at a larger scale?
We cannot survive in the colds of space, yet it is the tendency of matter to clump together, creating that space, which also gives us a sun and a planet and an atmosphere that was necessary for our existence. We cannot survive in the lifeless molten rock, yet we also could not have formed without its heat. It's usually at the edges of things where the most interesting and varied phenomena arise.
Much of the universe is hostile to human life in its current form, yet it is also the universe that gave rise to it. Whether human newborn or microscopic insect hatchling, the experienced world is at its most incomprehensible, and the organism at its most vulnerable, at the beginning. Growth allows for greater experience, and experience can lead to greater growth. Our knowledge of what's possible for a species extends only backwards, and only on one single speck of dust.
The universe has no inherent vastness, hostility, emptiness or silence without a conscious being to assign it those properties. I choose to imbue it with aspects of magnificence, intelligence, organization, perhaps even benevolence. We are not in a cosmic battle to conquer a dumb, hostile substrate; we are embedded in a rich, nurturing agar whose purpose it is, at least in part, to help us to flourish.
Agreed.
I choose to imbue it with aspects of magnificence, intelligence, organization, perhaps even benevolence.
Well, that's nice, but it's only possible at the space scale of, at most, the Earth. Or at least human scale - since it is you who are doing the choosing. I mean, it's fine to change the subject, but at least be aware that you are doing so.
>We are not in a cosmic battle to conquer a dumb, hostile substrate; we are embedded in a rich, nurturing agar whose purpose it is, at least in part, to help us to flourish.
Again, I feel like you are confusing space- and time- scales. Earth's biosphere is nurturing agar, no doubt, over a short time scale and small space scale. But the solar system, galaxy, let alone the universe at large, is most definitely not, and nor is the Earth's biosphere on any large time scale.
We have arisen in this universe, apparently the first intelligent life on this planet, and on any other planet. If there is intelligent life elsewhere, we are separated by impossible vastness: to get to another star, in any form, we must harness immense energy. More than is reasonable. And once there, the lifeboat must somehow found a colony, a self-sustaining biosphere, which itself is a gargantuan task.
Perhaps, though, this is a benevolent universe: maybe those great chasms of space protect intelligences from each other. It means that if any of us manage to make the journey, and we discover each other, it will be a meeting of great good fortune, and the travellers will always be at the mercy of the inhabitants of the destination star.
There is hope for long-term survival of intelligence, but it is very, very small.
Hmm, my intent was to expressly claim those things at large scales, or more accurately, at all scales. It's just that it is easier for us to understand e.g. the organization and benevolence only in the things which are of the most immediate usefulness: we understand the biosphere as a resource because we can use that now, whereas we don't really understand how to harness the full power of a star because it's too far beyond our current capabilities.
> Earth's biosphere is nurturing agar, no doubt, over a short time scale and small space scale. But the solar system, galaxy, let alone the universe at large, is most definitely not, and nor is the Earth's biosphere on any large time scale.
I agree with you about Earth's biosphere space-wise though perhaps not time-wise. But if Earth is a cradle of intelligent life, surely we are not meant to stay here indefinitely? Even now the wild idea of mining asteroids, once relegated to science fiction, is reaching the point of feasibility. What was previously useless (not to mention unknown) to us is becoming useful, and will act as a stepping stone (or a catapult) to enable some other far-fetched ideas to come closer to manifestation. I believe this is a pattern that does not necessarily end.
> If there is intelligent life elsewhere, we are separated by impossible vastness: to get to another star, in any form, we must harness immense energy. More than is reasonable. And once there, the lifeboat must somehow found a colony, a self-sustaining biosphere, which itself is a gargantuan task.
There was a time, or more accurately various times for various peoples, that crossing an ocean was almost beyond imagination. As we continue to gain in knowledge and understanding of how the universe works, especially our understanding of spacetime, these problems become far less intractable. Interstellar travel is just a bigger ocean. The scale of the metaphor is logarithmic but so is the pace of the growth of our collective intelligence. My personal belief is that we will find a way around the immense time and energy requirements that we currently believe are necessary for interstellar travel. Terraforming is the harder problem: understanding ecosystems is IMO orders of magnitude harder than understanding fundamental physics, and building them from scratch seems like a pipe dream even for a dreamer like me. But if we figure out how to travel smarter then it becomes more a question of finding the most Earth-like planets that already have much or most of what we need. Still not easy, but probably easier and faster than trying to terraform Mars.
> Perhaps, though, this is a benevolent universe: maybe those great chasms of space protect intelligences from each other.
This seems plausible to me also. Seeds need a volume to grow in that is often tens of millions times larger than the seed itself. There could be a cosmic ecosystem that we are still too young to see or understand, and we may not catch a glimpse of it until we have spread through the stars.
> There is hope for long-term survival of intelligence, but it is very, very small.
If you mean small hope for human intelligence, then I would agree. I think this has less to do with the harshness of the universe out there and more to do with our own species' maladaptive behavior patterns. We are in a species-scale moment of crisis, which is to say one that has been going on for hundreds or thousands of years and may last a few hundred more, but there's no question in my mind that we need to get our shit together now if we want the species (and/or its children) to survive. This is why I think it is important to view the universe through a lens of possibility: it gives us a direction to steer towards. A more hopeful outlook inspires activity to push towards a better outcome.
Thank...
In fact intelligence manifests rarely in nature and only in long lived species with big complex brains which spend years educating their young to develop their cognitive skills. This implies that intelligence requires both a nontrivial nervous system and substantial nurturing thereof.
Ergo, I conclude that intelligence itself is inherently complex and unlikely to arise simply or spontaneously.
What they are are intelligent about varies, and the assumption that intelligence arises rarely is dangerous to ecological and environmental decision making. Rather, we don't know how to communicate with animals well and we don't think the same ways.
http://www.nature.com/news/2010/100909/full/news.2010.458.ht...
http://scienceblogs.com/mixingmemory/2007/08/17/metatool-use...
http://blogs.scientificamerican.com/running-ponies/catch-the...
http://ngm.nationalgeographic.com/2015/05/dolphin-intelligen...
A reasonable hypothesis would be that intelligence comes with high cost, such as high energy consumption, high heat production, longer period of infant stage, adaptation of body parts that will reduce their strength and interfere with their functionality, etc.
All those negative effects more than offset the benefit of higher intelligence in the short term. Because natural selection does not plan but only pushes evolution toward local maximum points, most organism never evolves high intelligence.
It has, all animals display intelligence; I think you need to update your definition of the word, you're using it as if only humans have it and we haven't believed that in a very long time.
Generate random strings and feed them though a compiler. You have can create algorithms. That's not a very good algorithm but it shows algorithms that create algorithms exist.
Now most modern AI algorithms, such as neural net, don't create algorithms (at least not directly), yes. But that might be indication that AI has further to go.
For me, the real question of interest is not about that kind of intelligence, but of "awareness". Computers may process environmental information and perform all sorts of logical operations based on it. But there is zero reason to think that any computer has any awareness whatsoever. It has no experience. There is something it is like to experience the color yellow, as opposed to merely classifying observed light as being of a certain wavelength; no computer has begun, in any way, to make that leap.
It seems like a lot of people, including the author, don't even notice that this is a question. They seem to assume that computer-like information processing is all that happens in a human mind, and that awareness itself doesn't even exist as something to ponder. So they equate any doubt about whether a computer can do what we do as a form of vitalism.
But with just a little bit of introspection, the fact that there is something it is like to experience the color yellow, as opposed to classifying a visual stimulus as being of a certain wavelength, or logically classifying it as having the attribute of "yellowness", is absolutely clear -- and absolutely fundamental to our existence as human beings (although many would assume that many other forms of life also have this basic ability to experience, if not the same ability to process information in computer-like ways).
This is why, for me, the Turing Test is orthogonal to the question of whether a machine is actually conscious. I have no problem imagining that a software system, even on hardware that is basically the same as what we have now but bigger and faster, will one day be able to pass the Turing Test. But that doesn't prove it has any experience of the color yellow; it merely means it can mimic the output of an entity that does.
This is literally the point of Turing's thought experiment: if a machine gives you answers indistinguishable from a human, is it alive?
These thought experiments are very interesting and thought provoking. But - still - it's a mathematician asking if something is alive/human. Not the most authoritative source on the subject matter.
The original computers were women in warehouses calculating actuary tables. They had a set of criteria to perform evaluations of numbers.
In short, you can reduce a subset of the human experience to algorithmic emulation, which, in effect, reduces a human to a computer. It's sort of what it means to do any job to be frank.
> In short, you can reduce a subset of the human experience to algorithmic emulation, which, in effect, reduces a human to a computer.
You're underestimating everything associated with being human. The computer won't make mistakes because a familiar is sick. He won't perform slower than usual because he slept less. He won't have a urge to poop affecting is performance, also. This is somewhat what I meant: these thought experiments are interesting, but if you look closer there's really not much to them. Human computers were, of course, computing. They didn't, however, became "not human".
Race conditions cause mistakes. Seems similar to having your mind elsewhere.
> He won't perform slower than usual because he slept less.
Didn't clean up memoery so now your computer is resource starved and running slower.
> He won't have a urge to poop affecting is performance,
GC pause.
I feel like people often focus too much on the "implementation" when it comes to consciousness. Beyond what instinct gives us, which is really just a pre defined set of conditionals, everything we know and understand is learned, programmed in a way. I don't see why it's unfeasible that given enough time and the right environment that an AI could do the same.
> Beyond what instinct gives us, which is really just a pre defined set of conditionals,
I don't think even neuroscientists have a definite answer to this. However, computer people - as always - do. That's, I think, the problem. A KISS approach is great; being oblivious of complexity isn't.
I don't think computer people think they have the definite answer, it's just easy to apply familiar abstractions to things when they map well. Plus https://en.wikipedia.org/wiki/Fixed_action_pattern seems quite like a conditional to me, and neuroscientists seems to agree. Comparing things at a higher level doesn't mean being oblivious to the lower levels, it just means it's outside the scope you care about. If we could 1:1 replicate every single possible input/output/decision/etc that happens in the human brain with transistors and have the exact same properties as a human when interacting with it, does it really matter if the path to get to those interactions is different? To me, it doesn't. We don't define people by which neurons fire in their brain, we define them by the actions and decisions those chemical reactions cause, and I don't see why it should be any different if they're made of silicon.
So you'll consider it impossible up to the point when such a machine exists?
No one says it is impossible.
It is more a question of whether our current efforts are enough to get us to the point.
Can we get to another galaxy? Probably, if we have enough resources and we discover enough to do it, and we actually go through with it. But it can seem unlikely, even given an unlimited amount of time.
Not to mention that we as a species may be facing a severe catastrophe in the future as a result of how we have populated this planet in the last 200 years. Hopefully we can avoid it, but once again are you sure?
I don't find it hard to assume that algorithms will one day be able to write posts that an observer may classify as seeming to be heartfelt.
But that doesn't mean that the machine actually experienced anything in writing it. That it's "heartfelt"-seeming post was actually felt by the algorithm at all.
"This is literally the point of Turing's thought experiment: if a machine gives you answers indistinguishable from a human, is it alive?"
Turing assumed it was. I think the issues are orthogonal. If it is indistinguishable by a human, all that means that I can see is that it's a very effective mimic, not that it has awareness.
Do you think that everyone reading that experienced it as heartfelt?
Did you have the "awareness" that it was heartfelt from birth, or did you only know that because you can recognize that it's much like other writing described as "heartfelt", that is, is it something you learned?
If you did in fact learn it, which I assume you did, since we're not born knowing language or even what heartfelt means, why do think that an AI couldn't learn the same? Do you think that how you "experience" it is universal to all humans? Or does it come from things you've encountered in the past.
Regarding your argument about being aware of the color yellow upthread, have you asked someone that is red-green colorblind how they tell the difference or how they experience red and green? It's clearly not the same as people who can see the full "visible spectrum", so do they not have awareness of those colors?
In the past when I asked a friend of mine who is red-green colorblind how he can distinguish the difference, he told me it was because he'd seen things people call red or green in the past. I asked if it was the same as the difference between colors, he said yes, but more like the difference between teal and a slightly different teal. So if you were never told what yellow is, would you be aware of it, or just know it as a slightly different orange? Personally, I know what colors are only because I know what other things of the same color look like, the same wavelength hits my retina so I know that it's the same color. How is this any different from an AI recognizing that?
Because I wrote that original post, and I had experiences while writing it. I know that from direct observation and I don't need other evidence.
"Regarding your argument about being aware of the color yellow upthread, have you asked someone that is red-green colorblind how they tell the difference or how they experience red and green? It's clearly not the same as people who can see the full "visible spectrum", so do they not have awareness of those colors??"
Yes, I was wondering if someone would bring that up, but my point wasn't that they have the exact same experience of yellow as me, but rather that they have an experience of a color (or sound, or taste, or anything) at all. And sure, some people experience fewer colors. That doesn't change the argument.
"I asked if it was the same as the difference between colors, he said yes, but more like the difference between teal and a slightly different teal." Interesting!
Well I said "a" heartfelt seeming post, not that specific one. If an AI told me the same thing, what difference would that make? I have nothing but the word of the author. How do I know you're not just an AI, and does it make any difference if you used neurons or transistors to come up with the concepts expressed? To me, the difference is meaningless.
> but rather that they have an experience of a color
But what defines that experience? Is it not just a wavelength of light hitting your eye, and your brain interpreting it?
I also find it interesting that you addressed the questions that were easy to refute, and left most of them unanswered. I am genuinely curious on your thoughts, as I find your conclusions as mystifying as you find mine.
EDIT: > And sure, some people experience fewer colors.
Ah, but do they experience fewer colors, or do they experience them differently? Since the wavelength of a color isn't what defines experiencing it, then what does? And surely you can't say they don't experience it just because of the wavelength.
This question was asked and answered concerning blue and green:
"The Himba tribe from northern Namibia, for instance, does not classify green and blue separately, the way Westerners do, but it does differentiate among various shades of what we call green. And when tested, members of the tribe, who are likely to have trouble with blue-green distinctions that most Westerners make easily, readily distinguish among greens that tend to look the same to Western eyes."
http://6thfloor.blogs.nytimes.com/2012/09/04/its-not-easy-se...
Which is all we might be doing with each other as well. That awareness could merely be an illusion and there's even some evidence to think it is. Experiments show your brain makes a decision before you are consciously aware of it despite you thinking you made the decision consciously. That leads to the conclusion that awareness is simply a lie your brains tells itself to feel in control.
Let's not talk about realistic AI for a moment, let's talk about hypothetically building a human protein by protein (or whatever building block you're comfortable with saying doesn't have qualia on its own). We're going to write the computer program to build this human. Nanobots will carry it out, every part of the process is understood and nothing is left for biology to do for us. We're just putting together lego blocks and at the end is a human indistinguishable from any other human.
Does this person experience qualia? How would you know? Apparently you can't just ask, since this isn't a valid thing to do with a computer (it's not really experiencing qualia! It's just saying that!).
Let's just say you come up with some objective qualia test. Now we can build a slightly more deficient human protein by protein and test it for qualia. In fact, we can do a qualia binary search and...
Once you take apart the argument like this, it's easy to see that arguments that computers can't experience qualia are arguments against materialism or reductionism, or are patent attempts to sneak souls and other woo into the discussion.
If you forget your fear of "soul" and other terms not ratified by science journals, you'll find out there are plenty of theories in metaphysics and spirituality on what it means to be and what awareness is. It requires you to at the very least read those philosophical/theological concepts instead of dismissing it as "woo". Cattle dwellers in India wrote many things about it eons ago. Joseph Campbell's work is a nice place to start.
I have encountered these theories. In the past I have been under the sway of them. They can be interesting to think about, but without any evidence for them there isn't any reason to give them particular consideration. (I am including rational, mathematical, or logical arguments based on experimental evidence as evidence here)
Without evidence, how do we pick which of these myriad theories to believe? How do we decide what the actual answer is? Science and materialism have repeatedly produced predictions that pan out in experiments. Theories of souls and sprituality have repeatedly had to retreat from testable predictions.
To take another tack, how did these spiritual theories come about? If they're true, then the ancient theorist must have seen some phenomena in the real world, and extrapolated correctly from that evidence. If that's the case, these ancient theorists were unparalleled geniuses, but we don't need to worry about ancient spiritual theories themselves, we'll rediscover them in time through our plodding (but safe) scientific method. If we don't rediscover them, then they were interesting theories that didn't pan out.
I'm glad we're discussing this because you're an example of the kind of thinking that is a mystery to me.
The fact that I have qualia when looking at, for example, the color yellow, is directly knowable by means of just a little introspection. This is not "rational, mathematical, or logical" or based on "experimental evidence". It is simply direct observation. Observation is where the scientific processes you name start. Observation does not depend on those processes. They depend on it and build from it. Observations are the most fundamental thing.
From the direct observation of your experience of yellowness, which I think you must agree is something more than merely classifying the wavelength as being in a certain range, I would think you would know that qualia exist. That is enough evidence.
"How do we decide what the actual answer is? Science and materialism have repeatedly produced predictions that pan out in experiments. Theories of souls and sprituality have repeatedly had to retreat from testable predictions."
So, let's not have a theory of souls or spirituality at this point. We don't know enough. That doesn't mean that those things don't exist, it only means we don't understand them.
Suppose a cave man-level creature on another planet encounters an iPhone because we somehow send one there. This cave man creature knows how to build fires and has used reasoning to make weapons, etc. In doing those things, he probably tried different ideas about how best to make fires and rejected the ones that didn't work well. In other words, he has a rudimentary understanding of the scientific method.
Now he tries to figure out how an iPhone works. He constructs theories based on what he knows. The theories are all shown, over hundreds of years, to be wrong. Therefore, he is told, iPhones don't exist because when he constructs theories about them, they are wrong.
But he directly observes the existence of the iPhone because he holds it in his hand. He knows they exist. He doesn't need a scientific theory to prove that they exist. Maybe, eventually, he will have a correct scientific theory as to how they work; it will just take more time. Or maybe he never will. Either way, iPhones exist.
Awareness is, in my view, exactly the same, and it is mystifying to me that people such as yourself come to a different conclusion.
What we shouldn't do, is say humans have qualia and jump to the conclusion that machines can't have qualia, and can't be "aware" in the same sense a human can be. I apologize if that wasn't your intended argument.
"What we shouldn't do, is say humans have qualia and jump to the conclusion that machines can't have qualia, and can't be "aware" in the same sense a human can be. I apologize if that wasn't your intended argument."
No, we can't make that jump. But my intuition as a programmer (and one who is interested in A.I. and does A.I. in the sense that some of my mathematical ideas were incorporated into a number of spam filter products) is that they don't. There is simply nothing there that would "add" awareness to the algorithm. All there is is information processing. It can classify something as either having the attribute of yellowness or not. But something beyond that functionality is required to have it experience the qualia of yellowness.
My point is that we have no idea what that something extra is. Somehow, people have it. But we don't know how. There are hypotheses, like those of Roger Penrose who discuss quantum phenomena as possibly being behind it, as made manifest in certain microscope structures in the brain. Suppose Penrose is right. Then, maybe one day a quantum computer can incorporate the same phenomena and be conscious for the same reasons we are.
But I do not believe that, for example, a spam filter as we know it today has the slightest experience of qualia. There is just no "room" for it. Everything in the spam filter has a specific function for the classification task and has nothing to do with experiencing qualia. The same is true for the most complicated deep learning network.
Maybe software as we know it can induce qualia as some kind of an emergent phenomenon, and the brain also induces it through some kind of emergent phenomenon. But I tend to hypothesize there would be something that causes that, some mechanism that the emergent phenomenon grows out of, that is not embodied in something like a current-day deep learning network.
My hypothesis is that to create a machine that experiences qualia, we will first understand the mechanism behind our own experience of qualia, and then reproduce that mechanism in a machine. I hypothesize that this would be a new and extremely fundamental, and transforming, scientific discovery.
And if we do reproduce that mechanism in a machine, and THEN that machine passes the Turing Test without us programming mechanisms to merely MIMIC human behavior, then I'll tend to assume that we've created a truly conscious entity -- though we may never be able to completely prove it.
On the other hand, there's no reason to assume we'll ever discover how it works -- for instance, we can't rule out the idea that there is a God who doles out souls to human bodies, and that we'll never have the means to understand God or souls. I don't believe that's the case -- but plenty of people do, and we can't presently prove they're wrong.
From my experience with machine learning I think that we (humans) dont need to understand a problem to be able to build a tool (AI) which can solve that problem. The necessity of understanding problems in order to solve them is obsolete :-P
Yes, that's the point. He has "experimental evidence" (aka something observable) right there in his hand. Understanding of how it works is immaterial for the question of existence (if you're not a constructivist :) So then we use that as analogy to look back at the question at hand:
> From the direct observation of your experience of yellowness, which I think you must agree is something more than merely classifying the wavelength as being in a certain range, I would think you would know that qualia exist. That is enough evidence.
Enough evidence of what? Certainly evidence of internal thoughts and feelings. I don't see how you can demonstrate anything beyond that, with or without an understand of how that "thing" might work.
I've read them. They are, from a scientific point-of-view, all woo. Almost by definition. They necessarily appeal -- if they can be epistemologically justified at all -- to a non-utilitarian epistemology. Science is based on a radically utilitarian epistemology: claims are justified to the extent that they are demonstrably useful in explaining and predicting observed phenomena. Theories -- in the non-scientific sense, obviously -- that involve entities that produce results in the physical universe but which are not governed by physical law or controlled by physical events as they are conjectured to be outside of the physical universe -- are non-utilitarian. They are irrelevant to any predictive model of observed behavior, and it is impossible to distinguish by observation between the existence of such non-physical, uncaused causes and processes that are purely physical. Either (a) there exists an adequate scientific model of the process, in which case the model being an accurate model of an underlying physical process is indistinguishable from the uncaused cause merely mimicking the behavior of the model at the times it has been observed, or (b) there exists no adequate scientific model, in which case the physical process not yet being discovered is indistinguishable from a non-physical cause.
That's not to say that such conjectures cannot be true, just that they can never be validated or distinguished from either known or unknown physical processes.
Say we create a machine that behaves like a human to all outward measures. There is no way to distinguish between:
(1) Human and machine alike being purely physical,
(2) Humans having metaphysical features like a "soul" and internal awareness and machine simply mimicking behavior,
(3) Humans and machines alike having metaphysical features like a "soul" (perhaps because God chose to reward human ingenuity by creating a "soul" for the machine.)
I avoided "Pragmatist" specifically because I'm familiar enough to know it refers to a philosophical school of thought that touches on the epistemological questions I was referring to, but not quite certain that the exact position I was referring to was within its scope (I knew it was along the same general lines.)
Consider this - if conscious observers didn't exist, in what sense would anything else exist?
If I told you that right now there is a universe out there which you can never detect nor deduce its existence from any evidence in the one you're exposed to. You won't be able to verify its existence in your lifetime. But, I tell you, it EXISTS! What doss that mean?
If you say it's meaningless, you'd be onto something. Furthermore if I told you that, if you wash your hands you'll be killing 10 innocent little bunnies in that universe. Will you stop washing your hands? No. You will err on the side of that undetectabld universe and bunnies not existing - after all how have I come to possess this knowledge if you don't?
If anything, you'd say that universe can only exist in the sense that some conscious observer in the future will VERIFY ITS EXISTENCE.
Now, if you subscribe to atheism, then in order to even build a machine, you need an intelligent animal that arose organically. So, in what sense would OUR universe exist if humans never arose?
YOU SEE even the concept of existence is tied into our subjective experience! It is the dual of it and equally mysterious. Why is there something rather than nothing, indeed?
> [A] universe can only exist in the sense that some conscious observer in the future will VERIFY ITS EXISTENCE.
> So, in what sense would OUR universe exist if humans never arose?
I think things exist independently of whether anyone is observing them. That's because "observing" is just a description of a bunch of particles interacting with each other. Some of those particles are part of a larger pattern of particles we call human brains. There's nothing particularly unique about our brains in terms of physics.
At a higher level, we can look at the interactions between our brains and the environment and think about brains as a different kind of thing. A pattern recognizer, a prediction maker, an effective evolutionary adaptation that leads to the propagation of the genes that create it. But there's no reason to suspect it creates universes.
I absolutely agree that a lot about our brains is mysterious and there are a lot of questions that need answers. The next few decades are going to be very interesting as we discover many of those secrets. I just don't think we'll discover that among those secrets is the ability to make universes real or not.
Certainly, we can only observe the universe we are in and phenomena that are within our light cone. But that's a limitation on us, not on universes.
Can you define what existence means as you use the term?
To illustrate the difference, look at the universe before intelligence evolved. The most reasonable scenario is that this universe existed, even though there were no observers. Something was going on before anyone was looking, because that process led to the creation of intelligence.
It's possible that the universe didn't truly exist until consciousness existed to be aware of it. That it popped into existence once consciousness did, and everything just happened to be consistent with a universe that already existed beforehand for billions of years. But that's a pretty surprising theory. It's much more likely the universe just already existed.
Or do we? You'd have to define the word "exist" in some terms that relate it to things we are familiar with. Otherwise it is meaningless. That's my point. Trying to define existence is like trying to explain consciousness. If you think consciousness explained by souls is "woo" perhaps you should consider the nature of your concept of existence! You take it for granted and use it in a sentence without really exploring its definition.
To me, that's kind of a silly question. Yes, it might be true. Of course, it might be that there are no other human beings and I'm just a brain in a vat. I just find that hypothesis to be too unlikely to care about.
"We're just putting together lego blocks and at the end is a human indistinguishable from any other human. Does this person experience qualia?"
I would guess so.
"How would you know?" For me, this is a little like your first question. I wouldn't necessarily know. But if this entity is physically indistinguishable from a human at even the most infinitesimal level of physical reality, I'd assume it experiences.
But, I may be wrong here, because it might be that there is something non-physical called a "soul" that resides in humans, and a physically identical entity might not be provided with one by God or whoever/whatever the soul-provider is. That's not what I tend to assume, but I have no evidence either way.
Your question "How would you know" is an interesting one because it assumes that it matters whether or not an observer knows. Maybe some entities have awareness and others don't, and an observer can't tell the difference. So what? That wouldn't imply, in any way, that the phenomenon of awareness doesn't exist.
"Once you take apart the argument like this, it's easy to see that arguments that computers can't experience qualia are arguments against materialism or reductionism, or are patent attempts to sneak souls and other woo into the discussion."
That doesn't mean that such arguments/hypotheses aren't true. It just means they may not be testable using the scientific method. Much as many people don't like the multiverse theory because it may not be testable using the scientific method. But that merely means that there are limits to the truths that can be discovered by the scientific method; it doesn't change the reality of what is true.
It's a bit reminiscent of Godel's completeness theorem. Many great mathematicians assumed that there were no truths that weren't provable within the systems they were studying; Godel proved that such assumptions were simply wrong.
I've been wondering about this for a few years, can't find anything about that because it's too generic a search term. Is there a name, or references about this ?
ps: ha, these in-thread answers already give pointers https://news.ycombinator.com/item?id=9864468 https://news.ycombinator.com/item?id=9864479
<closed/>
Purple is the weird one as it doesn't correspond to a wavelength, but can only be represented by a set.
Edit - all colours correspond to sets of sets, if you think about it, as you can experience exactly the same colour for essentially an infinite set of different sets of frequencies.
I have begun to suspect that our concepts of consciousness and morality are the result of our language and the way we have defined them. With morality is has become clear to me that personal duties are relative to personal goals. That is, "X should do Y" is incoherent, it really implies "X should do Y if X wants Z to happen" which is what the colloquial understanding is. Always being careful to mention Z removes any feeling of mysticism from morality!!
Perhaps with statements about consciousness, the situation is similar. I couldn't really formulate any sort of mystical question once I was carefully adding the SUBJECT of each sentence explicitly in.
The main question I have about consciousness is not about the processing - that can be explained mechanistically. But about the final result of perceiving this orchestra - why am I in my brain and not anyone else's? What is this extra element that is required for the conscious experience and can we ever be sure a machine we built is really subjectively perceiving anything in that metaphysical sense?
Thoughts?
This seems to be a misjudgement. There are actually plenty of people who have studied traditional philosophy and dualistic approaches to consciousness for years and eventually arrived at a naturalistic conclusion after they've studied neuroscience. I would actually be very much interested in the numbers for the opposite case: How many subscribers to dualism have knowledge in neuroscience, human behavioral biology, the theory of evolution and physics?
What you experience when you see yellow is what you experience when you see yellow. Nothing more can be said about it and it cannot be compared to anyone elses experience of yellow, human or machine.
So if you make that experience a condition for calling another being intelligent, you are effectively saying that nothing but yourself can ever be intelligent.
You need to break down that thing you call experience of yellow or awareness of yellow and explain what properties it has (intension) or you need to be able to show that it influences the behaviour of humans in a way in which machines cannot be influenced by it (extension).
I think you are confusing the principles of whether an observer can determine whether an entity has an attribute with the question of whether it does in fact have that attribute. But those are entirely different issues because, for one thing, the observer may simply not have the tools yet to make such a judgement.
Matter is made of atoms. For a long time we did not have the tools to determine that. But that didn't change the fact that it was, and is, true.
Now suppose the human race died out before it ever developed such tools. Obviously, it would still be true.
Now suppose that due to our limited intellectual abilities or for some other reason, it never was, or will be, possible to ever develop such tools. To me, it's obvious that that has no bearing on the truth of the proposition any more than if we didn't have time to develop the tools.
Similarly, the question of whether you have awareness is independent of whether I or any other third party can prove you do.
But I do agree with you in that the question you asked is worth asking. Not even mentioning the issue is certainly a frequent mistake made by authors taking a purely technological perspective.
No... I'm not inviting you to be convinced based on my experience. I'm inviting you to notice how you yourself feel when you yourself look at something yellow. My bet is that you'll experience that which is referred to in the literature as "qualia".
So if someone built a very complex machine that was able to communicate with me like a human, and if that machine claimed to feel something very specific at the sight of yellow, it would not feel right for me to say "no, you're not feeling anything like a human would feel" because I don't know how other humans feel. I have no way of comparing my own feelings with those of other humans or non humans.
:-p
I disagree: it would be mimicking the output of an entity that is mimicking....
My point is that these qualia arguments are begging the question, because they posit some 'real' entity that is somehow different from all the 'mimics', despite the fact that no-one is saying what the difference between them is.
Which one do you really want to build in a computer? The intelligence, plainly.
A more meaningful direction for article, imo, would have been to adopt a concrete definition of intelligence, and then talk about how ferrets, chimps, babies and adults fall on this continuum. Perhaps, connectomics, molecular biology, psychology or neuro-anatomy help us approximate the 'atomic number', giving us the opportunity to speculate about the underlying equivalent of quantum mechanics.
Like the article states, evidence is lacking, and we're somewhere between 1-100 nobel prizes away. But, the power of a concrete definition helps point to what we know we don't know, and discussion about whether the definition of intelligence itself needs to be re-framed to answer a question like this.
In fact, it's probably prudent to stipulate whether the goal is to specify the developmental algorithm that gives rise to intelligence, or intelligence itself. The generating process might be a lot more simple than the finished product, in the same way that the output of a pseudorandom number generator is very complex in terms of entropy (but not kolmogorov complexity, I guess), but the generating algorithm is very simple. Or the Rado graph, which is sorta maximally complex in the sense that it contains all finite and countably infinite graphs as induced subgraphs, yet it has a simple generating scheme. As a last example, consider No Man's Sky - relatively simple algorithm compared to the astonishing complexity of the worlds.
I do believe there is a relatively simple high-level description of neural development. But it's curious how it's relatively robust to genetic manipulations. I remember early in graduate school listening to a lecture about a particular mouse model for autism, caused by just one gene. The lecturer excitedly told us that it had a very high rate of behavioral manipulation - 75% or so. Not coming from a mouse-model background I was astonished that it was 'only' 75%. What happens in the other 25% - the gene doesn't just magically reappear, there are other compensatory mechanisms. What drives those? I suspect that's the more abstract level of description that a simple algorithm might describe.
https://en.wikipedia.org/wiki/Rado_graph
http://www.hutter1.net/ai/pfastprg.htm (The Fastest and Shortest Algorithm for All Well-Defined Problems)
https://en.wikipedia.org/wiki/No_free_lunch_theorem
The NFL theorems have been misused by advocates of 'intelligent design' -- they do not factor against evolution since evolution doesn't demand a global maximum, only a local one. (There are other reasons but this is the most fundamental.) But the theorems themselves are incredibly interesting.
In short they show that averaged over the space of all possible search landscapes, no search algorithm performs better than any other.
Obviously the universe does not provide "all possible search landscapes," since the universe has structure. Some search landscape structures and meta-structures are pathological and rare in nature. But nature does provide a huge diversity of them. The NFL theorem is to me powerful evidence that something as general as human or even animal intelligence must be a superposition of multiple "algorithms" rather than just one.
It also, I think, explains why all single-algorithm or single-approach methods of AI end up being domain specific. Obviously they'll be domain specific -- they only work against search landscapes with a certain structure!
The NFL theorems are a big reason I am a short term AI skeptic. I don't think human-level or beyond AI is impossible, but I think we are quite far from realizing it. I'll become more optimistic when AI researchers start studying biology more closely and deeply, since biological systems are the only existing examples of truly intelligent systems.
Hutter has written a bit about NFL: http://arxiv.org/pdf/1111.3846.pdf http://arxiv.org/pdf/1105.5721.pdf