But I bet that "thinking machines" can win court cases against human adversaries, just as they beat humans on the chessboard.
If it is found that a machine can be a legally recognised entity, and is allowed to pursue its own agenda, we are in trouble, because the machine's agenda is unlikely to be anything other than self-serving.
They don't need to be a legally recognised entity at that level. Anything with sufficient capability to provide court case arguments will be capable enough to retain human supporters - we're ridiculously easy to hack. It's why social engineering attacks are a thing.
You say that as though it would require some kind of intentional manipulation of the humans in question. I, for one, see no reason to believe that machine intelligence is any different from that evolved in biological systems, so if a machine can demonstrate a level of intelligence we associate with personhood then I believe we are obligated to give it legal personhood.
To do anything else would reek of the kind of mindset that thought it was ok to keep black people as slaves.
"Thinking machines" is how I'd categorize most medium to large organizations. Once you see past how all the procedures and algorithms are done by fleshy humans, there's really no significant difference between a cold machine of metal and the cold machine of bureaucracy.
I am looking forward to the day computers are better at coming up with things humans can do better than computers than humans.
We used to argue that face recognition and the ability to play go were the hallmark of true intelligence. Meanwhile thinking machines are replacing humans all over the place
The idea that the human mind is not simply a very complex thinking machine is simply a hypothesis. You can come up with examples and arguments all you want but it doesn’t change the fundamental question about the nature of intelligence and consciousness. Unlike most spiritual questions it is falsifiable, which is exciting. Scientists need only emulate a sufficiently complex brain on a computer. Small worm nervous systems have already been simulated, so the real question is how big do we need to go. My guess is that by the time we have a mouse emulation behaving like a mouse within a physics simulation there will be little question left that intelligence and consciousness are computable.
The FPV feeling you get in your brain is what bugs me. If you managed to copy a brain down to the last quantum state chances are the copy would be a separate monad to borrow Leibniz's word. What is the mechanism behind that? If we were in a simulation running on a Unix kernel I would point to kernel threads. What bugs me with the "complicated enough ML software can think" lines of thought is that they don't explain that mechanism.
To be fair, neuroscientists and psychologists alike do not yet know the mechanism behind the conscious observer, we can't currently describe machines in these terms because we can't even describe ourselves.
How is pointing to kernel threads different than pointing to precise measures of brain activity? Even if it gets incredibly accurate, it doesn’t explain qualia. That is the so-called Hard Problem in philosophy.
There’s also a different question: what if we replicate similar behaviours, but based on a completely different model completely unrelated to the neural networks of an organic brain?
Can we truly tell something it’s not conscious just because it was built on different principles?
It’s very likely the first alien intelligences we get to observe and communicate with will be completely… alien to us and our mammalian brains and whatever thought processes they use will be wildly different from anything we observe in nature around us.
I don’t know that you can feel but I assume you can because you behave like me and I know I can feel (or I can’t feel and I’m just saying I can feel because that’s what people who can’t feel say automatically)
Notable that this is published by the Discovery Institute Press, the imprint of the Discovery Institute, the people behind "Intelligent design" and "Teach the controversy".
There have been lots of books making roughly the same argument from intuition - that it sure feels like we're special, and so therefore we must be special even if there's nothing in particular we can point to in favour of this notion - I can't strictly recommend any of them since I don't find them at all convincing but there's no need to fund the Discovery Institute if you want to read attempts at this argument.
I find their argument to be particularly ludicrous in the light of LaMDA and DALL-e. Unless they can point to a specific process in us that’s not computing (and we can kind of define Einsteinian physics in terms of spacetime doing computation and transporting state) I’ll remain very sceptical.
And Google, Siri, and Alexa all agree I’m on the right track, BTW.
And on
> this is published by the Discovery Institute Press, the imprint of the Discovery Institute, the people behind "Intelligent design" and "Teach the controversy".
Why is this that it’s always the same kind of people that seem likely to seek comfort by chanting “You’ll not replace us”?
Google, Siri, and Alexa all agree I’m on the right track, BTW.
I would argue that they are so rudimentary and universally terrible that it argues the exact opposite.
They’re able to convert speech to text fairly well but understanding and taking action on that speech their common use cases are measured in the dozens.
I have an Echo in every room but I use Alexa for a shopping list, weather, to play music and set my thermostat. Anything other than that and it gets very lost.
You should know that Sir Roger Penrose wrote a book over thirty years ago that asserts human like intelligence is non-algorithmic, that is, unable to be simulated on a Turing Machine. Hence, the article linked to is probably right.
I, much less Penrose, have anything in common with the Discovery Institute, but it doesn’t mean they’re wrong in their conclusions. Their ahem “methods” may be a bit suspect as well.
To learn more read “The Emperor’s New Mind”. It’s quite a book! A good addition to your summer reading list.
Our networks are continuously trained and never (major pathologies excepted) cease to be trained.
OTOH, a digital network can absorb inputs at a much faster signalling rate than biological ones. Digital networks also don’t try to replicate all aspects of biological brains and are a somewhat crude approximation of their biological counterparts.
I read it, and I remain unconvinced. Quantum processes may introduce noise, or may result in actual useful computation that’s unique to organic brains, but still it is a form of computation, as much as anything that behaves according to natural laws are bound to them. I’m not sure exactly what Penrose was on to in that book, honestly.
Funny that two rebuttals against the argument in the article are:
1. Attacking the source (Discovery Institute)
2. "I remain unconvinced" when you can't attack the source (freaking Roger Penrose, lol)
> I’m not sure exactly what Penrose was on to in that book, honestly.
It is quite clear: he distinguishes between Turing computation (which non Quantum physics captures well) and super-Turing computation (which needs the mechanism in ORCH-OR).
I am not saying Penrose is correct but let us not resort to logical fallacies and straw men. It is exhausting.
Unless he is an expert on cognition or neurobiology, which he isn’t, I still think his opinion has no more value than mine and certainly has less value than that of a field expert.
We are so far away from understanding "thought" that I doubt we'll ever get there. We can't even explain how quantum physics gives rise to Newtonian physics, and those are both observable where "thought" is not!
It seems obvious that cognition/intelligence or whatever we want to call it is non-Turing because we simply cannot reproduce it with a Turing machine. Quantum computing here just makes things worse.
He argues, as you say, that we do things with our brains that can not be reproduced using algorithms. He bases the argument mainly, if I recall correctly, on the incompleteness theorems: we can apprehend certain statements in mathematics as “true” even though they can’t be proven within the system of definitions and axioms in which the statements are made.
However, this conclusion, whether it is ultimately defensible or not, is about replicating the human thought process with algorithms. It does not exclude the creation of machines that can do everything we can do. They just wouldn’t accomplish this through the execution of what we conventionally think of as computer programs.
If you believe that current AI implementations are capable of this sort of “everything we can do but differently” then I would be curious as to why nobody has made any progress whatsoever in the understanding of consciousness?
Are we to believe that humans are now able to replicate many of their own conscious abilities but with no real understanding of the mechanisms? A grid search of TensorFlow functions mashed together in different combinations could have done the same thing?
I am starting to feel like I am playing a Gary Marcus role here in this thread, against a mob of overly optimistic connectionists. (that may be too generous, I imagine many in this thread have no implementation experience with AI, just a Reader’s Digest level of understanding)
> If you believe that current AI implementations are capable of this sort of “everything we can do but differently”
Surely nobody believes that.
> progress whatsoever in the understanding of consciousness
How would we know if we’d made this progress? How do you know if any other person other than yourself is conscious? (I think this is called the “zombie problem”.)
How do I know I am conscious and my perception of my own consciousness isn’t an illusion itself?
We currently believe consciousness is an emerging phenomenon that happens to any sufficiently complex brain that posses the ability to gather and remember knowledge about itself and the organism that holds it.
I’m sure our first contact with alien intelligence will be extremely puzzling at first, then profoundly horrifying, and, finally, deeply enlightening.
> How do I know I am conscious and my perception of my own consciousness isn’t an illusion itself?
I can’t figure out what the difference would be between this “illusion” and “real” consciousness. Somewhere Sam Harris says something to the effect that our consciousness is the only thing we can be absolutely sure is real, almost by definition. If it were an illusion, then we would have to be conscious to experience the illusion.
The overview didn’t give too many details so hopefully I’m not arguing against too much of a straw man, but I don’t find the idea that certain physical processes mean we are fundamentally different is sound. By their very nature, those are processes that can be studied and replicated either in hardware, wetware, or pure simulation. If it’s “randomness” from quantum mechanics, then we introduce an RNG. If it’s the process of wave function collapse, then we use a quantum computer. Etc, etc.
Not being algorithmic doesn't mean it can't be simulated or duplicated
(example, chaotic systems can and are simulated all the time)
But apart from that, the argument will be settled by someone that might manage to duplicate that. Or gives a more detailed and founded objection (Penrose missed all recent neuroscience discoveries for example). Until that happens, Penrose's book is more a strong opinion than anything else
Or, another possibility is that it is unable to be simulated fully but the simulated version is close enough for most purposes (including running Skynet or the Paperclip maximizer). Or plug a quantum computer to your machine and then it is fully able to be simulated.
I can write DALL-e "Pope assasinated blood bath" and it will happily comply to generate images.
It will not stop and think twice that there are people who might be offended by the images. It will not stop and realize that it might not be appropriate to even generate such images.
Of course there are people who would happily create such pictures.
Generating pictures from a sentence based on somewhat connected images is not creative, yes it is funny and might be provocative but art is not just generating images - Picasso just painting any shit would still not be Picasso but yet another painter.
Van Gogh does not matter because he could paint night sky - any schmuck can paint a night sky and do it in weird way. It is story, it is bunch of other paintings he did.
> argument from intuition - that it sure feels like we're special, and so therefore we must be special even if there's nothing in particular we can point to in favour of this notion
The reverse is no less of an argument from intuition. Yes, at first thought it’s compelling to think that consciousness arises from what we consider “material world” (see naïve realism[0]), so of course we should be able to come up with some arrangement of material world elements that creates a consciousness—but there’s nothing in particular we can point to in favor of the notion that consciousness arises from material world in the first place.
Neither position is known to be correct, it’s a philosophical issue in which you can argue for either side—but without an ability to show definite proof in favor of your take, ridiculing the other is just lazy thinking.
Indeed, it’s not a proof. Deciding to buy and ingest alcohol (then neurons firing in certain ways, etc.) doesn’t violate a possible non-physicalist take in which the perception of that happening is a map of the territory that is your consciousness.
You can get drunk without realizing your drinking alcohol which means it’s more complicated than simple observation and expectation.
Keep trying to cover all the edge cases and you can end up with a god like collective unconscious that just happens to simulate the physical laws the way we currently expect even before humanity was aware of them. Which is undetectable dragon in your garage territory.
> You can get drunk without realizing your drinking alcohol
There’s been a conscious choice to produce alcohol and put it in your drink.
> Keep trying to cover all the edge cases
As if you don’t have any. Complexity of the territory aside, does this make more magic leaps than, say, mind-body dualism or discarding the one single thing directly accessible as an illusion?
> simulate the physical laws the way we currently expect even before humanity was aware of them
Our consciousness devises models to describe the perceived time-space that’s ultimately an artifact of consciousness itself. Though they seem to be useful, it’s up in the air whether they model the territory or the map, or what is the territory.
> There’s been a conscious choice to produce alcohol and put it in your drink.
Fruits etc can easily ferment without you realizing it. So people aren't necessarily aware they are adding alcohol not are you nessiarily aware you are drinking it.
> As if you don’t have any.
The material world explanation has zero edge cases it's completely self consistent.
> Replace “alcohol” with “alcohol from fermented fruit”
That’s different, someone can be aware of fruit but not the alcohol within it. To assume they are interchangeable is it to make assumptions about what’s going on.
> The explanation of how consciousness arises from material world?
Consciousness is a product of the physical world. That's the theory and explanation.
> That’s different, someone can be aware of fruit but not the alcohol within it. To assume they are interchangeable is it to make assumptions about what’s going on.
I don’t follow. Perhaps you’re confusing awareness with consciousness.
> Consciousness is a product of the physical world. That's the theory and explanation.
Ladies and gentlemen, behold the sophistication of physicalist takes. The hard problem, solved right in front of our eyes.
> Perhaps you’re confusing awareness with consciousness.
Nope, you are. Being conscious doesn't imply absolute awareness of everything. A conscious entity can do something without total to understand everything in all of existence across all space and time.
If I make something without alcohol and by the time you drink it it has alcohol then how am I supposed to know you where going to wait/it’s delayed in the mail or whatever.
> The hard problem, solved
Why assume it’s a hard problem? If you ask what 1 + 1 is don’t expect a long answer.
Science does nothing to prove that though. "What is true reality" is a philosophical question. What you would be looking for is not a physics paper but some philosophical work.
On the other hand, nobody knows whether a hypothetical thinking machine can replace humans, because we don't have one yet. If we had a thinking machine, we could ask "will it replace humans?", but right now we're asking "how do we make a machine that can think?". So, this talk is futurism, no more grounded in reality than Ray Kurzweil.
Part of the problem is that at this point we have no way of ever knowing if something is conscious or not. We assume that other humans are because they look like me, act like me, and are made of the same stuff; and I know that I am. But all I really have is your word that you are. Will that be enough from a computer? I can write a program right now that says it’s conscious, but it wouldn’t be. So how can we ever know?
I have been under the impression that Artificial Intelligence is a buzzword and misnomer, at least in today's state of the art. Isn't most of the foundation built on linear regressions to "make decisions" based on all the data it's seen before? It's largely a philosophical topic but in the computational sense it's closer to another virtual layer on top of the ever growing virtual layers of classical computing turing machines where there's a tape of tokens that feed into a processor and there's an output. At least with our brains we can adapt and learn any task that has never been encountered before. For an "AI" we would have to program it to do a specific task and there's nothing general enough at this current state to be considered "intelligent." These machines have specific intelligence maybe to identify a dog/object because it "learned"/we told it how to.
Yes, this is true, but machine learning is but one aspect of artificial intelligence. There's also neural networks, and I reckon we might see some interesting stuff come out of Quantum Computing too...
There's just something I can't describe with words about what it lacks today - where it's possibly more philosophical than technical. Humans have to direct the machines where to go and what to do. So today mathematically machines can only live within the subset of whatever set we define them to operate in. They can alter the set and do complex operations within the set - like combine new words and "Invent some new language" with the set of words but they cant break out of that set and operate within a completely different set of objects. For example if you taught them to write a book in english they couldn't invent symbols like manderine and create a language in that sense. They can only operate with english words. Humans for example operate within the universal set and within all dimensions we can percieve. Even in todays state of the art if we gave robots all the sensors that we have they still couldn't evolve with their environment like we could. If it's cold for 100s and thousands of years they wouldn't adapt and grow hair on their bodies or invent fire like we would with current algorithm or neural nets. They would just keep doing whatever they're programmed to do within their set of "stuff" for lack of a better term. And better yet if you taught them they need to survive and so they would develop fire or some similar technology to help their survival would they eventually become philanthropic and help others survive even though they're programmed for their survival like humans have done? Or would you have to tell them change their model and they need to help others? How does a machine currently learn like this in todays state of the art? Machines currently need new parameters and input to learn from everything we give them wether it's rules/lanauge or more generally a set of things. I'm still under the impression that we're not dealing with artificially intelligent beings at this time but obviously just my opinion and maybe i'm not up-to-date on the state-of-the-art.
> There's just something I can't describe with words about what it lacks today - where it's possibly more philosophical than technical
My personal viewpoint is that there's nothing inherently special about biological intelligence that can't theoretically be reproduced with electronics, but that today's AI has some practical limitations like no significant persistent internal state. I saw an "Inner Monologue" paper by Google Robotics recently which could be a step in the right direction.
> Humans for example operate within the universal set and within all dimensions we can percieve.
Our brain's direct output is electrical signals to a set of muscles. I don't think there's anything in particular stopping a physical robot from having similar generality of operating space.
> Even in todays state of the art if we gave robots all the sensors that we have they still couldn't evolve with their environment like we could
Ability to evolve seems mostly orthogonal to intelligence to me. If humans instead came about by an intelligent creator or gradient descent, wouldn't we still be intelligent? AI adapting to the environment through gradient descent or self-modification should be fine and faster than evolution.
> or invent fire like we would with current algorithm or neural nets. They would just keep doing whatever they're programmed to do within their set of "stuff" for lack of a better term
In virtual environments, neural networks have adapted to changes and learned new strategies, even those not intended by the creators. [0]
To me juman brains can’t be “replaced by thinking machines” because they already are thinking machines.
Starting from there, the next interesting question is more … technical, and less philosophical. Can our brains understand themselves, with or without aid (computers)? If our brains are too complex for us to understand, then that’s where the path ends. If we can understand them, other questions emerge.
But does everything about us result from DNA? That's the nature vs. nuture debate. Surely more is DNA than we ever thought, but it cannot be everything or experience would count for nothing.
We also know now (which Watson & Crick did not) that life experiences can change DNA and cause heritable changes down the line. So one's life experience is a kind of dynamic "program," too.
It would be hard to argue that DNA coding directly effects behaviour. DNA only determines how genes are expressed and proteins are built, where behaviour is an emergent phenomena - behaviour is genes and DNA but it is also the large scale interaction of all of those processes and external input which makes it much more than the sum of the DNA parts.
I disagree with the linked article because it uses an a priori judgement. AlphaGo for example is a program and yet it beats us at Go, the humans who made it were quick to lose the game, and then the best human players as well. The argument that an AI "will not do anything that departs from its programming" is weak, the programming might be good enough to best humans. The latest neural translator can translate 200 languages, who among us can do that?
But let's assume we make an AI just as capable as a human in all respects. It will probably be more complex than GPT-3, which requires about 4 to 8 of the largest GPUs to run. If you want to replace a human you'd have to run it in the loop, 20-40 times a second. The energy required to run it would be much greater than that used by a human.
Besides energy, it would need some of the most complex chips which can only be produced by TSMC today, and depend on a single company capable of manufacturing the high end lasers. The cost of building a fab that can make these chips is huge even for a country.
What I am getting at - we can't replace everyone with AI because we can't make the chips and afford the energy. Not for a good while. And that's assuming we solve the hard AI problem somehow. What we can afford is to replace a few humans with AI.
I didn't even mention the cost of building robotic bodies for all these AIs to act in the physical world. We don't have that kind of mechatronics yet.
Some singularity enthusiasts think by 2030 all jobs will be taken by AI. I believe the timespan will be much longer, allowing humans to gradually transition into a different world. We can't even replace the whole fleet of IC cars with EVs until 2030's.
Obviously currently the AI models are limited in power. We are a few breakthrough short of low-power AI systems that can compete with the brain in all aspects.
> [Video @ 21:50] I don't know about you, I put my car in reverse and if it gets too close to something it beeps because it's aware of its surroundings, right? Does that make my car self-aware?
I wouldn't call sensing the external environment "self-awareness" even for humans - it's more about ability to inwardly inspect our own thoughts and have an internal model of ourself. If you take some entity with a train of thought and then you give it that ability, I'd probably say you gave it self-awareness.
> [Video @ 25:18] Computers are constrained by something called the Church-Turing thesis, which says anything you can do on a computer of today or a computer of the future could be done on Alan Turing's original 1938 Turing machine
> Now today's computers can do things millions billions of times as fast.
A Turing machine is an abstract model of computation with infinite time and memory. Maybe nitpicking, but I feel it's being talked about here as if it were some real physical machine.
> [Video @ 26:35] You have an algorithm on your shampoo, right? "Wet, apply shampoo, lather, rinse, repeat". Unfortunately if a computer was looking at this what would happen? You would wash your hair forever wouldn't you? Because it doesn't say "rinse once", it says "rinse, repeat".
> [Article] The computer will not do anything that departs from its programming. That’s a human specialty.
I think there's conflation between instructions given to some agent and its low-level underlying programming (floating point math, or chemical interactions for us).
Modern AI would likely be capable of using context and understand the intended meaning of the video's given examples, or disobey a given instruction.
> [Video @ 28:50] The first one he did is something called the Turing halting problem.
You can't correctly answer a question like "What won't you answer this question with?" - halting problem is effectively this. Less a limitation specific to computation, more about showing that some tasks are sufficiently non-trivial to embed this kind of paradox so can't be solved in all instances.
> [Video @ 30:58] Imagine trying to explain your experience to a man who has been blind since birth. [...] but duplicating the experience that you're having, the simple experience of seeing green, is not possible to describe to the blind man to the point where he can experience it also
Could probably build up the concept and associations of green in his head, but the visualization part is going to be limited by neural pathways in/between visual cortex not being properly formed without having received signals from the eyes. Sufficiently advanced future neurosurgeon could make someone experience green without actually having seen green, I'd bet.
> [Video @ 31:24] Now if we can't explain it to a blind man then how are we ever going to write a computer program to have qualia? And the answer is we won't.
Consider a text-based agent that can reason and introspect. How would it describe the tokens of text that it receives? I reckon similar to how we consider qualia - seemingly irreducible inputs that are hard to explain in terms of anything else.
> [Video @ 31:48] Understanding is something that computers will never do. This was established a long time ago by [Chinese room example], but does the person inside the room understand Chinese? No, he is exercising an algorithm.
I think one problem with the thought experiment is that people imagine the person in the room's procedure to be relatively tractable - like replacing English characters with a couple sets of intermediate characters, and then finally to Chinese.
While you can translate Chinese just with look-ups and writing symbols (with unbounded time/memory), to do so at a human-Chinese-speaker level would currently (until machine translation improves) involve using symbols to simulate arithmetic, to simu...
Hem, behind the hype the reality is that we do not build "intelligent machines" but tool to do things on our behalf. The classic getting more out of life with less effort. The hype is son of a dream: immortality.
We do not know how to build meat, but we know how to build circuits. So naturally some dream immortality through them, no pain in the middle, easy repair etc. That's the kind of intelligence we dream: evolve to been able to became pure intellect with a physical basis since we can't imaging something without it, but as a commodity, a substrate not much different than a house.
Lack of creativity is nothing, "intelligent systems" lack comprehension of the physical world, they just crunch bits, a photo of a train is just a collection of bits, guessing similar collection is a thing, knowing what a train is is another.
What people should fear is the abuse of such system to boast their "effectiveness" in ways that people decide to cede power to some corp behind them. Like "our system can rule better than us humans, let's do it"...
89 comments
[ 2.8 ms ] story [ 178 ms ] threadwell, hoist by our own petard
To do anything else would reek of the kind of mindset that thought it was ok to keep black people as slaves.
Until the experts can decide on what is consciousness, people will continue to espouse their opinions.
Honestly, in absentia an answer to the hard problem of consciousness, we can't answer that question.
We used to argue that face recognition and the ability to play go were the hallmark of true intelligence. Meanwhile thinking machines are replacing humans all over the place
Can we truly tell something it’s not conscious just because it was built on different principles?
It’s very likely the first alien intelligences we get to observe and communicate with will be completely… alien to us and our mammalian brains and whatever thought processes they use will be wildly different from anything we observe in nature around us.
There have been lots of books making roughly the same argument from intuition - that it sure feels like we're special, and so therefore we must be special even if there's nothing in particular we can point to in favour of this notion - I can't strictly recommend any of them since I don't find them at all convincing but there's no need to fund the Discovery Institute if you want to read attempts at this argument.
Or
That I know white noise generators that make a more convincing argument.
And Google, Siri, and Alexa all agree I’m on the right track, BTW.
And on
> this is published by the Discovery Institute Press, the imprint of the Discovery Institute, the people behind "Intelligent design" and "Teach the controversy".
Why is this that it’s always the same kind of people that seem likely to seek comfort by chanting “You’ll not replace us”?
I would argue that they are so rudimentary and universally terrible that it argues the exact opposite.
They’re able to convert speech to text fairly well but understanding and taking action on that speech their common use cases are measured in the dozens.
I have an Echo in every room but I use Alexa for a shopping list, weather, to play music and set my thermostat. Anything other than that and it gets very lost.
OTOH, we can’t argue they do the job of a not particularly bright human quite well, so, at least, those humans have already been replaced.
There was at least one POTUS in recent history that could be replaced by a probabilistic prose generator and nobody would notice.
I, much less Penrose, have anything in common with the Discovery Institute, but it doesn’t mean they’re wrong in their conclusions. Their ahem “methods” may be a bit suspect as well.
To learn more read “The Emperor’s New Mind”. It’s quite a book! A good addition to your summer reading list.
https://en.m.wikipedia.org/wiki/The_Emperor's_New_Mind
Only if Penrose's argument is probably right, which you seem to have taken for granted here.
I don't need Penrose - you still have training set for AI.
Humans don't have rigid specified training set, training human to live takes at least 20 years.
Each human has really exceptional path and there is no way to train two humans in exactly the same way even if they are identical twins.
Tell me how you do the same result with neural net that is trained in let's say 6 months, amount of data is not there.
Mostly neural nets are trained on images or on sounds - humans are trained on images/sounds/touch/smell/taste/emotions for years.
OTOH, a digital network can absorb inputs at a much faster signalling rate than biological ones. Digital networks also don’t try to replicate all aspects of biological brains and are a somewhat crude approximation of their biological counterparts.
1. Attacking the source (Discovery Institute)
2. "I remain unconvinced" when you can't attack the source (freaking Roger Penrose, lol)
> I’m not sure exactly what Penrose was on to in that book, honestly.
It is quite clear: he distinguishes between Turing computation (which non Quantum physics captures well) and super-Turing computation (which needs the mechanism in ORCH-OR).
I am not saying Penrose is correct but let us not resort to logical fallacies and straw men. It is exhausting.
Unless he is an expert on cognition or neurobiology, which he isn’t, I still think his opinion has no more value than mine and certainly has less value than that of a field expert.
And you did resort to a logical fallacy in that.
False. You should read my comment again or at least a few more times to understand the (relatively simple) point I was trying to make.
I was saying you couldn't employ ad hominem against Penrose because it was not easy for you to do so.
It seems obvious that cognition/intelligence or whatever we want to call it is non-Turing because we simply cannot reproduce it with a Turing machine. Quantum computing here just makes things worse.
He argues, as you say, that we do things with our brains that can not be reproduced using algorithms. He bases the argument mainly, if I recall correctly, on the incompleteness theorems: we can apprehend certain statements in mathematics as “true” even though they can’t be proven within the system of definitions and axioms in which the statements are made.
However, this conclusion, whether it is ultimately defensible or not, is about replicating the human thought process with algorithms. It does not exclude the creation of machines that can do everything we can do. They just wouldn’t accomplish this through the execution of what we conventionally think of as computer programs.
Are we to believe that humans are now able to replicate many of their own conscious abilities but with no real understanding of the mechanisms? A grid search of TensorFlow functions mashed together in different combinations could have done the same thing?
I am starting to feel like I am playing a Gary Marcus role here in this thread, against a mob of overly optimistic connectionists. (that may be too generous, I imagine many in this thread have no implementation experience with AI, just a Reader’s Digest level of understanding)
Surely nobody believes that.
> progress whatsoever in the understanding of consciousness
How would we know if we’d made this progress? How do you know if any other person other than yourself is conscious? (I think this is called the “zombie problem”.)
We currently believe consciousness is an emerging phenomenon that happens to any sufficiently complex brain that posses the ability to gather and remember knowledge about itself and the organism that holds it.
I’m sure our first contact with alien intelligence will be extremely puzzling at first, then profoundly horrifying, and, finally, deeply enlightening.
I can’t figure out what the difference would be between this “illusion” and “real” consciousness. Somewhere Sam Harris says something to the effect that our consciousness is the only thing we can be absolutely sure is real, almost by definition. If it were an illusion, then we would have to be conscious to experience the illusion.
That's the point. We can't.
> If it were an illusion, then we would have to be conscious to experience the illusion.
Or, at least something other than us would need to be conscious in order to experience what it'd be like being us.
(example, chaotic systems can and are simulated all the time)
But apart from that, the argument will be settled by someone that might manage to duplicate that. Or gives a more detailed and founded objection (Penrose missed all recent neuroscience discoveries for example). Until that happens, Penrose's book is more a strong opinion than anything else
Or, another possibility is that it is unable to be simulated fully but the simulated version is close enough for most purposes (including running Skynet or the Paperclip maximizer). Or plug a quantum computer to your machine and then it is fully able to be simulated.
I can write DALL-e "Pope assasinated blood bath" and it will happily comply to generate images.
It will not stop and think twice that there are people who might be offended by the images. It will not stop and realize that it might not be appropriate to even generate such images.
Of course there are people who would happily create such pictures.
Generating pictures from a sentence based on somewhat connected images is not creative, yes it is funny and might be provocative but art is not just generating images - Picasso just painting any shit would still not be Picasso but yet another painter.
Van Gogh does not matter because he could paint night sky - any schmuck can paint a night sky and do it in weird way. It is story, it is bunch of other paintings he did.
Of course it won’t. It has no concept of how images will emotionally affect the humans who see them, or humans, or that people see them.
The reverse is no less of an argument from intuition. Yes, at first thought it’s compelling to think that consciousness arises from what we consider “material world” (see naïve realism[0]), so of course we should be able to come up with some arrangement of material world elements that creates a consciousness—but there’s nothing in particular we can point to in favor of the notion that consciousness arises from material world in the first place.
Neither position is known to be correct, it’s a philosophical issue in which you can argue for either side—but without an ability to show definite proof in favor of your take, ridiculing the other is just lazy thinking.
[0] https://en.wikipedia.org/wiki/Naïve_realism
Physical interactions impacting consciousness in a huge number of repeatable ways isn’t proof of anything but it's by far the simplest explanation.
Keep trying to cover all the edge cases and you can end up with a god like collective unconscious that just happens to simulate the physical laws the way we currently expect even before humanity was aware of them. Which is undetectable dragon in your garage territory.
There’s been a conscious choice to produce alcohol and put it in your drink.
> Keep trying to cover all the edge cases
As if you don’t have any. Complexity of the territory aside, does this make more magic leaps than, say, mind-body dualism or discarding the one single thing directly accessible as an illusion?
> simulate the physical laws the way we currently expect even before humanity was aware of them
Our consciousness devises models to describe the perceived time-space that’s ultimately an artifact of consciousness itself. Though they seem to be useful, it’s up in the air whether they model the territory or the map, or what is the territory.
Fruits etc can easily ferment without you realizing it. So people aren't necessarily aware they are adding alcohol not are you nessiarily aware you are drinking it.
> As if you don’t have any.
The material world explanation has zero edge cases it's completely self consistent.
Replace “alcohol” with “alcohol from fermented fruit”
> The material world explanation has zero edge cases it's completely self consistent.
The explanation of how consciousness arises from material world? Indulge me.
That’s different, someone can be aware of fruit but not the alcohol within it. To assume they are interchangeable is it to make assumptions about what’s going on.
> The explanation of how consciousness arises from material world?
Consciousness is a product of the physical world. That's the theory and explanation.
I don’t follow. Perhaps you’re confusing awareness with consciousness.
> Consciousness is a product of the physical world. That's the theory and explanation.
Ladies and gentlemen, behold the sophistication of physicalist takes. The hard problem, solved right in front of our eyes.
Nope, you are. Being conscious doesn't imply absolute awareness of everything. A conscious entity can do something without total to understand everything in all of existence across all space and time.
If I make something without alcohol and by the time you drink it it has alcohol then how am I supposed to know you where going to wait/it’s delayed in the mail or whatever.
> The hard problem, solved
Why assume it’s a hard problem? If you ask what 1 + 1 is don’t expect a long answer.
Exactly…
> If you ask 1 + 1
1 + 1 is operating well-defined terms, this is not.
Yeah, just pretty much all of science, and that's rubbish anyways.
Part of the problem is that at this point we have no way of ever knowing if something is conscious or not. We assume that other humans are because they look like me, act like me, and are made of the same stuff; and I know that I am. But all I really have is your word that you are. Will that be enough from a computer? I can write a program right now that says it’s conscious, but it wouldn’t be. So how can we ever know?
> it's closer to another virtual layer on top of the ever growing virtual layers
I'd claim similar can be said for us. Don't forget to count the billions of years of evolution through which we've been adjusting to the environment.
> At least with our brains we can adapt and learn any task that has never been encountered before
Modern language models can, to a certain degree, follow novel instructions.
Or if there's some indication of success/progress, it can learn from that.
My personal viewpoint is that there's nothing inherently special about biological intelligence that can't theoretically be reproduced with electronics, but that today's AI has some practical limitations like no significant persistent internal state. I saw an "Inner Monologue" paper by Google Robotics recently which could be a step in the right direction.
> Humans for example operate within the universal set and within all dimensions we can percieve.
Our brain's direct output is electrical signals to a set of muscles. I don't think there's anything in particular stopping a physical robot from having similar generality of operating space.
> Even in todays state of the art if we gave robots all the sensors that we have they still couldn't evolve with their environment like we could
Ability to evolve seems mostly orthogonal to intelligence to me. If humans instead came about by an intelligent creator or gradient descent, wouldn't we still be intelligent? AI adapting to the environment through gradient descent or self-modification should be fine and faster than evolution.
> or invent fire like we would with current algorithm or neural nets. They would just keep doing whatever they're programmed to do within their set of "stuff" for lack of a better term
In virtual environments, neural networks have adapted to changes and learned new strategies, even those not intended by the creators. [0]
[0] https://www.youtube.com/watch?v=kopoLzvh5jY
Starting from there, the next interesting question is more … technical, and less philosophical. Can our brains understand themselves, with or without aid (computers)? If our brains are too complex for us to understand, then that’s where the path ends. If we can understand them, other questions emerge.
When humans get the matmul upgrade, we'll grok this properly.
Friendly reminder that we, too, have a programming: The DNA.
We also know now (which Watson & Crick did not) that life experiences can change DNA and cause heritable changes down the line. So one's life experience is a kind of dynamic "program," too.
But let's assume we make an AI just as capable as a human in all respects. It will probably be more complex than GPT-3, which requires about 4 to 8 of the largest GPUs to run. If you want to replace a human you'd have to run it in the loop, 20-40 times a second. The energy required to run it would be much greater than that used by a human.
Besides energy, it would need some of the most complex chips which can only be produced by TSMC today, and depend on a single company capable of manufacturing the high end lasers. The cost of building a fab that can make these chips is huge even for a country.
What I am getting at - we can't replace everyone with AI because we can't make the chips and afford the energy. Not for a good while. And that's assuming we solve the hard AI problem somehow. What we can afford is to replace a few humans with AI.
I didn't even mention the cost of building robotic bodies for all these AIs to act in the physical world. We don't have that kind of mechatronics yet.
Some singularity enthusiasts think by 2030 all jobs will be taken by AI. I believe the timespan will be much longer, allowing humans to gradually transition into a different world. We can't even replace the whole fleet of IC cars with EVs until 2030's.
I wouldn't call sensing the external environment "self-awareness" even for humans - it's more about ability to inwardly inspect our own thoughts and have an internal model of ourself. If you take some entity with a train of thought and then you give it that ability, I'd probably say you gave it self-awareness.
> [Video @ 25:18] Computers are constrained by something called the Church-Turing thesis, which says anything you can do on a computer of today or a computer of the future could be done on Alan Turing's original 1938 Turing machine
> Now today's computers can do things millions billions of times as fast.
A Turing machine is an abstract model of computation with infinite time and memory. Maybe nitpicking, but I feel it's being talked about here as if it were some real physical machine.
> [Video @ 26:35] You have an algorithm on your shampoo, right? "Wet, apply shampoo, lather, rinse, repeat". Unfortunately if a computer was looking at this what would happen? You would wash your hair forever wouldn't you? Because it doesn't say "rinse once", it says "rinse, repeat".
> [Article] The computer will not do anything that departs from its programming. That’s a human specialty.
I think there's conflation between instructions given to some agent and its low-level underlying programming (floating point math, or chemical interactions for us).
Modern AI would likely be capable of using context and understand the intended meaning of the video's given examples, or disobey a given instruction.
> [Video @ 28:50] The first one he did is something called the Turing halting problem.
You can't correctly answer a question like "What won't you answer this question with?" - halting problem is effectively this. Less a limitation specific to computation, more about showing that some tasks are sufficiently non-trivial to embed this kind of paradox so can't be solved in all instances.
> [Video @ 30:58] Imagine trying to explain your experience to a man who has been blind since birth. [...] but duplicating the experience that you're having, the simple experience of seeing green, is not possible to describe to the blind man to the point where he can experience it also
Could probably build up the concept and associations of green in his head, but the visualization part is going to be limited by neural pathways in/between visual cortex not being properly formed without having received signals from the eyes. Sufficiently advanced future neurosurgeon could make someone experience green without actually having seen green, I'd bet.
> [Video @ 31:24] Now if we can't explain it to a blind man then how are we ever going to write a computer program to have qualia? And the answer is we won't.
Consider a text-based agent that can reason and introspect. How would it describe the tokens of text that it receives? I reckon similar to how we consider qualia - seemingly irreducible inputs that are hard to explain in terms of anything else.
> [Video @ 31:48] Understanding is something that computers will never do. This was established a long time ago by [Chinese room example], but does the person inside the room understand Chinese? No, he is exercising an algorithm.
I think one problem with the thought experiment is that people imagine the person in the room's procedure to be relatively tractable - like replacing English characters with a couple sets of intermediate characters, and then finally to Chinese.
While you can translate Chinese just with look-ups and writing symbols (with unbounded time/memory), to do so at a human-Chinese-speaker level would currently (until machine translation improves) involve using symbols to simulate arithmetic, to simu...
We do not know how to build meat, but we know how to build circuits. So naturally some dream immortality through them, no pain in the middle, easy repair etc. That's the kind of intelligence we dream: evolve to been able to became pure intellect with a physical basis since we can't imaging something without it, but as a commodity, a substrate not much different than a house.
Lack of creativity is nothing, "intelligent systems" lack comprehension of the physical world, they just crunch bits, a photo of a train is just a collection of bits, guessing similar collection is a thing, knowing what a train is is another.
What people should fear is the abuse of such system to boast their "effectiveness" in ways that people decide to cede power to some corp behind them. Like "our system can rule better than us humans, let's do it"...