OK, just an incremental improvement over lcd, oled, e-paper etc, but the characteristics enable deployment in many new situations. Plus it's a "why didn't I think of that?" technology.
https://www.clearinkdisplays.com/technology
Photonic computing is interesting; last year I attended a python meetup, and two of the presenters were writing Python tooling for designing photonic chips (called IPKISS). Seemed well suited for sensor data processing.
Wow IPKISS looks like something I would want to use. Ive been trying to write some of this myself and it's annoying. Thanks a bunch!
I'm working (like, on the side) on building a fully photonic computer. The progress is very slow, however some things fall into place quite a bit. Most photonic computing designs still use bits, mine would use signals which makes the computer much more powerful.
Robotics. Not humanoids necessarily (I find them creepy to tell you the truth), but the ones that will act in the background making mundane and repetitive tasks easier - garbage collection, cleaning, watering plants, restocking grocery store aisles, further automating industrial tasks, etc.
CRISPR. The ability to edit DNA in living animals (especially in the germ line) will cure genetic diseases, make us live longer, and increase our ability to feed the third world.
Electric vehicles, especially as the range increases and price and charging times drop. I hope that an ancillary benefit will be reexamination of how we generate grid power with a focus on renewables and nuclear.
CRISPR-CAS9 for genome editing. Briefly, it allows us to identify particular bits of DNA and modify one or more bases. It's a very new biotechnology (research essentially began within the last 10 years) which applies to essentially all species we've tested it on. It is cheap and so easy to perform that I could probably teach 99% of HN to do it in a week's training or less.
It's still not as accurate as we need it to be, but that won't stop biologists from modifying plants for improved performance as crops, bugs for less disease spreading, or other mammals to increase the pace of biological discovery. In time (and in China first, most likely), we will see the use of this genome-editing technology on humans flourish.
Thinking specifically of humans, it will allow us to eradicate inherited genetic diseases like sickle-cell anemia, to target cancer cells very specifically, and will hopefully revolutionize treatment in many sorts of diseases. Now, when it comes to enhancing humans and their germ-line (eggs/sperm), which allows enhancement on down the generations. this is the most controversial point. It will probably be done on the black market if we try to regulate it, and it would probably be immoral to disallow people from giving their children all the advantages possible.
But where does that stop? Do we allow a bunch of 6'5" mega-geniuses with blonde hair, blue eyes, and all the right genotypes to take over society? Will that occur? Who's to say. Either way, utopian or dystopian or somewhere in between, I think genome editing is the most influential class of technologies on the horizon.
Agree. I think the hunt for ever-newer computing tech is misplaced (blockchain lol) but CRISPR might actually cure cancer. That's way more important than Yet Another Logistics Company at this point.
I am not sure that human genome edition is something I would consider "exciting". It adds a lot of ethical questions, I am thinking about the kind of uses "I want my child to have blue eyes" in particular.
People are just having sex without thinking twice.
I'm married and in the appropriate age and i have thought about having kids a ton.
I'm not sure if i'm even allowed to create a new thinking human being only for the motivation of having a kid.
The crispr thing takes all our medical advantages just an additional level up.
You see people argue about if you should do a gen test for down syndrome (don't get me wrong here! A human being is a human being. But yes there is a consensus in our society that you become a human after a period of 3 month, at least in germany) and other complications but would never think about denying a doctor to use a incubator for a baby which would never ever be alive today without the medical advances.
There is an unbelievable huge hypocritical thinking in this. Yes tons of human beings would not be alive today without medicine but on the other hand the responsibility coming from all this knowledge doesn't matter?
I believe that in a moral/ethical right way you would need to do your BEST for your kid and this means making sure that by using crispr you do that.
I have no idea where this journey will bring us and yes there is tons of things to figure out but we will be able to overtake natures randomness for future live. And we have to.
Randomness already has nothing to do with (western) society. Modern healthcare and society in general allow for the survival and reproduction of individuals who would have simply died a few thousand years ago. The whole concept of "survival of the fittest" no longer applies to our species.
There is nothing wrong with this - as you said, everyone wants the best for their child. The big question is, what exactly _is_ the best? From my point of view, it seems that if designer babies were to become easily available, the average parent would just look on Facebook as to what the latest "influencers" fancy this week - and that would make the world an awful lot worse.
If someone has a blood cancer (i.e. they are damaged at a DNA level) and can only currently be cured by a danagerous stem cell transplant and then the ability comes along to fix their DNA back to what it previously was at the genetic level wouldnt you agree it would be more unethical to not allow them access to that treatment?
You don't think ridding ourselves of genetic diseases is an extremely large good? Millions of humans have incurable genetic diseases which vastly limit their quality of life and even lifespan itself. To me, this ALONE would be exciting. But when you add on the benefits to editing crops, livestock, microbes, and human traits not tied to disease, the ramifications become downright mind-boggling. I don't think it's exciting in a purely positive way, I honestly don't know if it's going to be good or bad on the whole. I'd wager that it starts off very good and slowly becomes less good, since the best use cases will all be targeted first.
"What would a spelling-checked person, one with no genetic typos, be like? Since no such person has ever existed, we have to speculate. I figure that kind of guy would win the decathlon, steal your shirt and your girl -- and you still couldn't help liking him."
This is all optimistic speculation. Being able to remove diseases from our genome is not equivalent to being able to create mega-geniuses with blonde hair. The "spelling-checked" persons DNA is just that, their DNA. It says nothing about how they're going to be raised, environmental factors etc. Before we can even start to think about "engineering" ourselves we still need to figure out how we're engineered to begin with and how that engineering changes throughout the span of our lives.
How would that work in practice? Of course, you could eliminate all hereditary diseases, make people smarter, remove certain genetic predispositions; but I'm not convinced that the "perfection" thing is possible.
The most obvious being, what exactly is perfection? If you ask 100 people to describe their perfect partner, you'd get 100 completely different answers. If our species would converge to any one of those descriptions, those who have a different idea of perfection would find themselves in a objectively worse world. Converging to some kind of average wouldn't be any better either.
Just take a look at the standards of beauty even over the last, say, 40 years. There's no constant, purely objective measurement of beauty, so to me it seems impossible to create perfection.
It's actually easy. For each allele, you simply take the majority allele.
Yes, this will lose some new beneficial mutations, but since most mutations are deleterious, taking majority allele is guaranteed to be an improvement in whole. Huge improvement.
I don't follow your reasoning. It might work for some genes, but a very large part has no meaningful significance at all. If we were to simply take the majority allele, that would eliminate all blonde people, all redheads, everyone who has any form of biological talent for anything whatsoever; you'd even get rid of white people altogether: after all, they are the minority. Taking the majority would mean getting rid of all variation. To me, that is not at all what I would call a huge improvement.
Plus there's going to be interaction between the genes. After all, that's exactly what evolution is trying to evaluate. If it was possible to simply know the answer, don't you think it would have found that in 3 billion years runtime ?
Duplicate some hollywood star's genes ? Maybe even a scientist's ? Some combination ? Perhaps (have fun being the first one to try it though). "Perfect" genes/humans ? No way. It'll take decades before we can create smarter humans. "Duplicate" your beloved pet ? Sure (domesticated animals are pretty close to clones already, but whatever).
However, it'd be much easier to create the perfect dishwasher/street wiper/soldier. Obedient, maybe even dependent on a drug and unable to reproduce by themselves (ala Monsanto plants). And the question isn't whether we can stop the US military from doing that. The question is whether we can stop China.
Just out of curiosity – how much does CRISPR allow us to modify adult humans? I've read a lot about designer babies and such, but what can the current population expect to get out of it?
Blonde hair and blue eyes? It seems to me that traits not currently found in nature would be most difficult to produce, ergo most rare and expensive, therefore most prestigious among a certain crowd.
How about purple hair and orange eyes? Natural skin pigmentation that mimics tribal tattoos? A tapidum lucidum? Ears with elf-points? Sharp vampire-style canine teeth? People are going to be designing their babies, which means certain traits will go in and out of fashion, just to keep people paying more money for "better" babies.
It is likely that certain traits will be subsidized as a matter of national security and/or pride, like those identified for intelligence, athletic ability, or resembling the current dictator-for-life. The rest will be a la carte, and maybe paid in cash under the table.
Forward-looking countries will subsidize packages to meet national goals, such as uniform body size, lower total mass, higher bone density retention, and lack of freefall-sickness for a space program, or strength and endurance for soldiers, or fine motor control and analytic ability for surgeons. Frivolous cosmetic traits would be add-ons for the rich and/or powerful. As they trickle down to the lower classes, they become less fashionable, and rich people pay to have them removed. Star-bellied Sneetches pay to be unstarred, and Plain-bellied Sneetches pay to get stars upon thars. Sylvester McMonkey McBean takes all the cash and gets away clean.
The reason PC and Mobile hasn't converged is because of the lack of good inputs that aren't mouse and keyboard. As long as there's no good way to input text on mobile, there will never be a convergence of those two verticals.
The Librem 5 from puri.sm is nice in that respect. The phone runs Linux and will have a docking station to attach monitor, keyboard, mouse, then use as laptop.
Self driving cars. The amount of time saved for humanity will be absolutely staggering and free up people to spend more time accomplishing more things.
I think this is the biggest part, and less the "sleep in your car while you commute for 5 hours each way!" which sounds fucking terrible.
Imagine the amount of space and time saved by having 1 enormous lot on the outskirt of town, like an airport. You walk around your dense urban area, call your car, 10 minutes it's there, it drops you off, and parks itself. No circling the block, no terrible big box concrete hellscapes, just Park Slope without the frustration.
Agreed that a lot of time will be freed up, but I don't think the majority of people will use this time productively. Most people will just end up watching Netflix in their cars instead of focusing on the road.
The real revolution for self driving cars will be what it does to deaths in the 18-35 demographic. A quick Google search turns up this (https://www.verywell.com/top-causes-of-death-for-ages-15-24-...) article that confirms that for the 15-24 demographic automobile accidents are the leading cause of death. Think of how many productive person-years we'll gain if self driving cars are even just 10% safer than driving yourself (and by all measures I've seen, they're way safer than that)!
Edit: To understand the real impact, I've always enjoyed this hypothetical future conversation...
Future Kid: You mean to say that people used to operate machinery weighing several tons? Traveling at high speeds? With a minimum of training?
Note: Below are just my opinions. I don't think downvoting is justified. I'm completely right and downvoters are wrong, but anyway this thread asks for what we think.
----------
My opinion: (100.000000% guaranteed to be right in this case.)
It's obvious that general artificial intelligence will be a reality and surpass human cognition in every way, because our brains are a couple of pounds of meat running on less than a hundred watts, using biochemical pathways: no light-speed switching, about a six orders of magnitude slower. Meanwhile, our datacenters run in the gigahertz domain. (There are, however, about a hundred billion neurons in the human brain.)
Some people think a computer will never be able to do general cognition. They're idiots. It is like saying that nothing but a sack of meat will ever fly (i.e. that only a bird can, no human-made engine can) -- at a time when we have engines that are 1,000,000 times more powerful than birds' muscles. (I mean at the time that Lord Kelvin famously stated that heavier-than-air flight is impossible.)
Any task of human cognition that the sack of meat in a skull can do, a digital algorithm will be able to do, because the former is just slow, slow analog meat.
Granted the emotional part is a point of difficulty but higher mental functions seem pretty unrelated to emotions.
AI is like when people were saying heavier than air flight is impossible. Obviously it's possible. And it will rock and change the world. (Anyone who doesn't see it this way is stupid and wrong. They're just on the wrong side of history. They have zero vision for things that are completely obvious. It's not even debatable. It's a sack of meat in a skull doing slow analog operations 24/7 for 3 years before it can even read. The bar is not as high as people think.)
I don't think it's a long-term view due to the numbers involved. For example, everyday developers might have access to 64-bit laptop computer with 4 cores @ 4 ghz and 16 GB of RAM (literally millions of people use such a machine, zero exaggeration).
When working on their own desktops, AI researches might use an 6-core or 8-core desktop with 64 GB of RAM or, rarely, 128 GB of RAM.
Access to 2-32 terabytes of RAM are not in the realm of what most AI researchers get to run their algorithms on on a daily basis.
Google's AlphaGo ran on[1] 1202 CPU's, 176 GPU's, and I didn't immediately find the amount of memory but on the order of magnitude you would think it is from 1 terabyte (if it's 1 GB of RAM per CPU) up to (reasonably) about 128 terabytes, with my guess being 4 TB-16 TB. However I didn't find this figure.
This is hardware that exists today, but AI researchers would have to wait a few years to have normal everyday access to this kind of thing in their normal everyday devices. Right now only datacenters match the processing ability of the human mind. Although they certainly exist in one cluster, they're far from the type of thing that is in everyone's hands yet.
On the other hand, once it is set up these clusters can be far faster than the human brain, which runs at really slow analog speeds. (Look up the propagation speed of neural signals.) The human brain is just slightly ahead of the curve right now without any specialized hardware.
There are 7 billion extant examples of intelligent human brains. There are probably less than 10,000 clusters AI researchers can play around with for months, which have 16-128 TB of RAM and aren't used for anything more important, just whatever they feel like running on it. (Versus easily more than 15 million computers with just 16 GB of RAM, though that may not be enough.)
Agree. I just saw doyoutrustthiscomputer.org (still on YouTube) where it was stated by some guy that Google is building a 100 million server strong datacenter for DeepMind, which has access to all their data.
On the example of AlphaGo.. in 1997 a scientist and Go fan stated that it would take at least 100 yrs or longer for a computer to beat humans at Go. 20 yrs later the champion was beaten..
First of all, the current state of AI and machine learning seems to be abysmal. The major breakthroughs (such as backpropagation) were made decades ago. The only reason that AI seems to be improving, is that a couple of companies are throwing an immense amount of money, data, and computing power into it. Even then, the results are relatively poor at best: systems like Google Now and Siri can only respond to a limited set of hardcoded queries and every kind of "suggestion" system seems to be actually getting worse in the last few years.
Secondly, Moore's law is as good as dead. If an algorithm requires a terabyte of ram, it is all but useless for most practical purposes. I'm currently looking for a replacement of a 5-year old laptop and the newest model is at most 20% faster, with the same amount of memory! Heck, last time I got a desktop it had an AMD Phenom II X6 1055T: for the same amount of money, you'd now get an i5 8500 or a Ryzen 5 1600x. The performance gain? Well judging by the Passmark scores, it would be a jump from ~5000 to ~13000. So that's roughly 2.5 times as fast in eight years, with a jump from 45nm to 14nm. Memory progression is about the same. Considering that the progression is slowing down, it'll take quite a few decades before the performance required for current AI is generally available.
Thirdly, the stuff which is impersonating AI isn't actually intelligent. Every current solution focuses on what basically boils down to a very complex set of linear regressions and similar methods. There is no adaptability, no sense of context, not even the slightest form of creativity. Everything which we consider to be intelligence is lacking. This isn't really a big surprise, because we simply don't know how it works in living creatures. Sure, we know how the neurons work, but the high-level understanding is lacking, so how could we even begin to imitate it or even improve on it? Sure, we could try simulating an actual brain, but a Blue Gene/L can only simulate half of a mouse's brain at one-tenth of the speed - and that's assuming the simulation is actually correctly modelling the biology.
So no, I don't think we're on the brink of an AI revolution. We might have something with the intelligence of a service dog in a decade or three, but even that would probably be a stretch. Unless a few very major breakthroughs occur, AI is going to be a bit of a disappointment.
I understand your frustration with the results. But please bear in mind that until the wright brothers flew, the results were frustrating.
What you need to look at is the raw power of internal combustion engines, and the weight they add. Versus the raw power of muscles of birds.
You need to look at the raw number of neurons firing in a human brain. This is the analog device that is our benchmark.
You need to look at the speed each neuron, of which there are a hundred billion, fires at: 200 hertz maximum, and more like 0.1 hertz on average.
you need to look at the speed of propagation: 8 orders of magnitude slower than the interconnects in data centers.
You are looking at a bird in flight, and you are looking at an internal combustion engine that outputs 1,000,000 times more power than its wings, and you are saying, "But our results are abysmal."
We will, and very soon, match that output. And far, far exceed that output. For sure we will. Because it's just analog meat. And our computers are faster and wider.
The poor state of algorithmic research is due to researchers having small access to the algorithms in question.
We have all of the machinery for "flight" (intelligence).
Personally I also think that researchers are impatient. Humans takes years to bootstrap to where they can understand the world as well as a 3-year-old. Specifically, humans take three years to understand or communicate as well as a 3-year-old. 36 months.
Parity with human brains isn't the goal. The goal is achieving 144 months (12 years) of brain development closely assisted and supervised by parents and teachers, especially within the first few years, within 9 hours of unsupervised learning.
But it will be done. You have the facts there in black and white.
I would say firstly, we don't have to know how to achieve it exactly, since once we have the hardware we can just experiment. The human source code is about seven hundred megabytes (fairly precisely, DNA is almost precisely 2 bits per base pair) so somewhere in that is the encoding for human intelligence, since humans inherit it. So we know that there is some < 700 MB program. Now 700 MB is a ton of source code, but it's not an unimaginable amount. We know some of the compiled results, as we can see how human brains are built actually. So there's lots of indications that we will be able to get similar results once we have the hardware for it - which we do.
Secondly I would point to many of the specific advances reached using machine learning as excellent indications that we are quite near the kind of algorithmic breakthroughs that are necessary. Since the topology of these algorithms is in many cases closely inspired by human neural topology, and many algorithms achieve many similar results, it seems quite likely that we would be able to reproduce some of these effects experimentally.
More than "likely", in many domains the computers have done just that.
> Any task of human cognition that the sack of meat in a skull can do, a digital algorithm will be able to do, because the former is just slow, slow analog meat.
It's not slow at all. And you are obsessed with meat.
Ah.. brings me to another invention: artificial meat, but as tasty and indiscernable as the real thing, without all the food waste required to raise cattle.
It's exceedingly slow. Neurons fire at not 4 gigahertz, 400 megahertz, 40 megahertz, 4 megahertz, 400 kilohertz, 40 kilohertz, 4 kilohertz, or 0.4 kilohertz (400 hertz.) They fire at < 1 hertz to about 200 hertz. [1]
That is literally 8 orders of magnitude slower than processers. 8 orders of magnitude is the size of difference between a grain of sand and 1.2 short tons.
In the time that your neuron moves 1 grain of sand, a computer can move 1.2 short tons worth of sand 1 grain at a time (talking about just cache here) - because it's switching at 4 GHz instead of 10 hertz.
That is enough to make a lot of back and forth trips. Now, you have to make a lot of these back and forth trips because the brain is extremely parallel. But the difference between these extremely slow biochemical firing speeds and gigahertz light speed enables this.
The speed of propagation in the brain is about 100 meters/second [2]. The speed of propagation between two fiber optic end-points is about 299 792 458 meters/second. That gives you 7 orders of magnitude to play with. Or, in other words, it can let you take seven orders of magnitude more trips across a data center (or go that much farther) before you miss a real-time deadline for when you need to address a neuron by.
In the future it's pretty clear that computers are not going to think as fast as the human brain when doing similar calculations using a similar topology, but, for example, 10,000 times faster than real-time.
This really shouldn't surprise anyone growing up in technology. Brains are slow, mechanical devices. They're not digital and they don't do signal processing at the speed of light.
It's taken us some time to catch up because a hundred billion neurons is still a lot of neurons, and algorithm development using similar topologies has been slow.
No, brains work on the neuron level. We know how many neurons there are (about 100b).
The reason I mention the bit level is because if we knew for a fact that the "code" that ran the structure of the brain fit in, say, 64 KB (if we knew this because the whole human genome fit in 64 KB) we would have long-since tried to emulate it.
Instead it is "guaranteed" (in a sense) to be a subset of about 700 MB of bits. That's the human source code.
It just gives you a sense of the scale of complexity that might be involved there. (700 MB is quite a lot of source code, don't get me wrong.)
my sibling post lists some of the numbers involved, but you should especially look the number of neurons, their connectedness, firing speed, and above all realize how incredibly slow this analog biochemical signal propagation is. You can look at how many parts make it up. (About a hundred billion.) Finally you can know for sure that the algorithm can be encoded in well under 700 MB, since that is literally the entire blue print of the human genome (bit-perfect, at 2 bits per base pair), which (obviously) encodes human intelligence as just a small part of it. (Since it includes everything else too, such as your entire rest-of-your-body.)
Look at the number of "mechanical" (biochemical) parts making up the human brain, their blueprint, the size they're contained in.
Finally look at the results that analogous machine learning algorithms are making with (albeit much simpler) neural net topology models.
The idea that nobody will ever match it is like looking at a mechanical watch and saying no digital model of it could ever come up with similar emulated results. What.
When you look at the size of the datasets involved you will see that it is exactly the sort of thing we can expect to be emulating now.
I'm at a point in life where technology doesn't concern me so much as human behavior and where we're collectively headed as a unique species in the cosmos.
Regarding convenience, comfort and entertainment, we are good enough — A person can get all sorts of luxury and services without ever leaving one's home — but technology that encourages us to treat each other better, feel better about ourselves, and removes barriers to physical and social mobility, is what really excites me anymore.
Smarter conversational AI. More applications just being plugins connected to your cloud AI instance. It's a revolution on user interfaces whose scope we're not even close to understanding.
The biggest part: it'll make people better at asking sophisticated questions and understanding how important that is.
I think it'll be as big a shift change as the Internet was.
I would love to ask specific questions instead of translating everything into multiple queries of "search engine speak".
Every time we find an edge case (or a noob question) that lands us on stack exchange, we could just ask an AI directly to look up the relevant documentation or scour through thousands of bug reports and workarounds.
Stage two describing the solution you want generated code for, or stage three stating the problem, or stage 4 finding problems and solving them.
I think wolframalpha is already pretty close to that. And, I must say, a lot more trustworthy than even stackexchange. I mean, if it gives an answer, you can generally trust it to simply be true.
I'm not sure 3D-printing houses will do much for their final prices. Actually putting materials on top of each other is not an overwhelming fraction of the total cost.
According to State Farm, completely rebuilding my apartment is would be 1/7th of what I paid for it, and that's with NYC union labor.
There's a reason that crackhouses in Vancouver cost $1 million, and it's not because they're made of gold. You need to build density to reduce the share of land cost per dwelling, anything else is a gimmick.
120 comments
[ 107 ms ] story [ 3859 ms ] threadhttps://web.archive.org/web/20171029180947/http://www.cleari...
Also most languages built on top of llvm.
I'm working (like, on the side) on building a fully photonic computer. The progress is very slow, however some things fall into place quite a bit. Most photonic computing designs still use bits, mine would use signals which makes the computer much more powerful.
It's still not as accurate as we need it to be, but that won't stop biologists from modifying plants for improved performance as crops, bugs for less disease spreading, or other mammals to increase the pace of biological discovery. In time (and in China first, most likely), we will see the use of this genome-editing technology on humans flourish.
Thinking specifically of humans, it will allow us to eradicate inherited genetic diseases like sickle-cell anemia, to target cancer cells very specifically, and will hopefully revolutionize treatment in many sorts of diseases. Now, when it comes to enhancing humans and their germ-line (eggs/sperm), which allows enhancement on down the generations. this is the most controversial point. It will probably be done on the black market if we try to regulate it, and it would probably be immoral to disallow people from giving their children all the advantages possible.
But where does that stop? Do we allow a bunch of 6'5" mega-geniuses with blonde hair, blue eyes, and all the right genotypes to take over society? Will that occur? Who's to say. Either way, utopian or dystopian or somewhere in between, I think genome editing is the most influential class of technologies on the horizon.
People are just having sex without thinking twice.
I'm married and in the appropriate age and i have thought about having kids a ton.
I'm not sure if i'm even allowed to create a new thinking human being only for the motivation of having a kid.
The crispr thing takes all our medical advantages just an additional level up.
You see people argue about if you should do a gen test for down syndrome (don't get me wrong here! A human being is a human being. But yes there is a consensus in our society that you become a human after a period of 3 month, at least in germany) and other complications but would never think about denying a doctor to use a incubator for a baby which would never ever be alive today without the medical advances.
There is an unbelievable huge hypocritical thinking in this. Yes tons of human beings would not be alive today without medicine but on the other hand the responsibility coming from all this knowledge doesn't matter?
I believe that in a moral/ethical right way you would need to do your BEST for your kid and this means making sure that by using crispr you do that.
I have no idea where this journey will bring us and yes there is tons of things to figure out but we will be able to overtake natures randomness for future live. And we have to.
There is nothing wrong with this - as you said, everyone wants the best for their child. The big question is, what exactly _is_ the best? From my point of view, it seems that if designer babies were to become easily available, the average parent would just look on Facebook as to what the latest "influencers" fancy this week - and that would make the world an awful lot worse.
We will. Quoting from https://westhunt.wordpress.com/2012/02/27/typos/
"What would a spelling-checked person, one with no genetic typos, be like? Since no such person has ever existed, we have to speculate. I figure that kind of guy would win the decathlon, steal your shirt and your girl -- and you still couldn't help liking him."
The most obvious being, what exactly is perfection? If you ask 100 people to describe their perfect partner, you'd get 100 completely different answers. If our species would converge to any one of those descriptions, those who have a different idea of perfection would find themselves in a objectively worse world. Converging to some kind of average wouldn't be any better either.
Just take a look at the standards of beauty even over the last, say, 40 years. There's no constant, purely objective measurement of beauty, so to me it seems impossible to create perfection.
Yes, this will lose some new beneficial mutations, but since most mutations are deleterious, taking majority allele is guaranteed to be an improvement in whole. Huge improvement.
Duplicate some hollywood star's genes ? Maybe even a scientist's ? Some combination ? Perhaps (have fun being the first one to try it though). "Perfect" genes/humans ? No way. It'll take decades before we can create smarter humans. "Duplicate" your beloved pet ? Sure (domesticated animals are pretty close to clones already, but whatever).
However, it'd be much easier to create the perfect dishwasher/street wiper/soldier. Obedient, maybe even dependent on a drug and unable to reproduce by themselves (ala Monsanto plants). And the question isn't whether we can stop the US military from doing that. The question is whether we can stop China.
I.e. get the "secret" formula for a unique hair color or body proportions, or whatever.
If everybody's "perfect" in the same way, there's no advantage to being perfect, and people will look for ways to differentiate.
(Of course all of this is Hollywood-level speculation, but still).
[0]: http://www.businessinsider.com/researchers-use-crispr-to-edi...
Better than a 6'2" idiot with a ginger toupee thats already taken over a society.
How about purple hair and orange eyes? Natural skin pigmentation that mimics tribal tattoos? A tapidum lucidum? Ears with elf-points? Sharp vampire-style canine teeth? People are going to be designing their babies, which means certain traits will go in and out of fashion, just to keep people paying more money for "better" babies.
It is likely that certain traits will be subsidized as a matter of national security and/or pride, like those identified for intelligence, athletic ability, or resembling the current dictator-for-life. The rest will be a la carte, and maybe paid in cash under the table.
Forward-looking countries will subsidize packages to meet national goals, such as uniform body size, lower total mass, higher bone density retention, and lack of freefall-sickness for a space program, or strength and endurance for soldiers, or fine motor control and analytic ability for surgeons. Frivolous cosmetic traits would be add-ons for the rich and/or powerful. As they trickle down to the lower classes, they become less fashionable, and rich people pay to have them removed. Star-bellied Sneetches pay to be unstarred, and Plain-bellied Sneetches pay to get stars upon thars. Sylvester McMonkey McBean takes all the cash and gets away clean.
Electric cars and the explosion of renewables are going to change _everything_, faster than people think.
Doing a quarter mile in a Model S is fucking AWESOME.
Imagine the amount of space and time saved by having 1 enormous lot on the outskirt of town, like an airport. You walk around your dense urban area, call your car, 10 minutes it's there, it drops you off, and parks itself. No circling the block, no terrible big box concrete hellscapes, just Park Slope without the frustration.
Cheap, too. Building parking in NYC costs ~$100k a spot https://www.nytimes.com/2007/07/12/us/12parking.html
I'm actually looking forward to that. It gives a whole new meaning to the phrase "dying in your sleep."
I highly recommend this Benedict Evans post on the second-order consequences of electric & autonomous vehicles: https://www.ben-evans.com/benedictevans/2017/3/20/cars-and-s...
Edit: To understand the real impact, I've always enjoyed this hypothetical future conversation...
Future Kid: You mean to say that people used to operate machinery weighing several tons? Traveling at high speeds? With a minimum of training?
Future Adult: Yup.
Future Kid: And nobody died?
Future Adult: Oh no! People died all the time!
----------
My opinion: (100.000000% guaranteed to be right in this case.)
It's obvious that general artificial intelligence will be a reality and surpass human cognition in every way, because our brains are a couple of pounds of meat running on less than a hundred watts, using biochemical pathways: no light-speed switching, about a six orders of magnitude slower. Meanwhile, our datacenters run in the gigahertz domain. (There are, however, about a hundred billion neurons in the human brain.)
Some people think a computer will never be able to do general cognition. They're idiots. It is like saying that nothing but a sack of meat will ever fly (i.e. that only a bird can, no human-made engine can) -- at a time when we have engines that are 1,000,000 times more powerful than birds' muscles. (I mean at the time that Lord Kelvin famously stated that heavier-than-air flight is impossible.)
Any task of human cognition that the sack of meat in a skull can do, a digital algorithm will be able to do, because the former is just slow, slow analog meat.
Granted the emotional part is a point of difficulty but higher mental functions seem pretty unrelated to emotions.
AI is like when people were saying heavier than air flight is impossible. Obviously it's possible. And it will rock and change the world. (Anyone who doesn't see it this way is stupid and wrong. They're just on the wrong side of history. They have zero vision for things that are completely obvious. It's not even debatable. It's a sack of meat in a skull doing slow analog operations 24/7 for 3 years before it can even read. The bar is not as high as people think.)
Claiming to be 100% right and calling people with different opinion idiots might have added a downvote or two (Personally I upvoted).
When working on their own desktops, AI researches might use an 6-core or 8-core desktop with 64 GB of RAM or, rarely, 128 GB of RAM.
Access to 2-32 terabytes of RAM are not in the realm of what most AI researchers get to run their algorithms on on a daily basis.
Google's AlphaGo ran on[1] 1202 CPU's, 176 GPU's, and I didn't immediately find the amount of memory but on the order of magnitude you would think it is from 1 terabyte (if it's 1 GB of RAM per CPU) up to (reasonably) about 128 terabytes, with my guess being 4 TB-16 TB. However I didn't find this figure.
This is hardware that exists today, but AI researchers would have to wait a few years to have normal everyday access to this kind of thing in their normal everyday devices. Right now only datacenters match the processing ability of the human mind. Although they certainly exist in one cluster, they're far from the type of thing that is in everyone's hands yet.
On the other hand, once it is set up these clusters can be far faster than the human brain, which runs at really slow analog speeds. (Look up the propagation speed of neural signals.) The human brain is just slightly ahead of the curve right now without any specialized hardware.
There are 7 billion extant examples of intelligent human brains. There are probably less than 10,000 clusters AI researchers can play around with for months, which have 16-128 TB of RAM and aren't used for anything more important, just whatever they feel like running on it. (Versus easily more than 15 million computers with just 16 GB of RAM, though that may not be enough.)
I don't think this is very far-off.
[1] http://uk.businessinsider.com/heres-how-much-computing-power...
On the example of AlphaGo.. in 1997 a scientist and Go fan stated that it would take at least 100 yrs or longer for a computer to beat humans at Go. 20 yrs later the champion was beaten..
First of all, the current state of AI and machine learning seems to be abysmal. The major breakthroughs (such as backpropagation) were made decades ago. The only reason that AI seems to be improving, is that a couple of companies are throwing an immense amount of money, data, and computing power into it. Even then, the results are relatively poor at best: systems like Google Now and Siri can only respond to a limited set of hardcoded queries and every kind of "suggestion" system seems to be actually getting worse in the last few years.
Secondly, Moore's law is as good as dead. If an algorithm requires a terabyte of ram, it is all but useless for most practical purposes. I'm currently looking for a replacement of a 5-year old laptop and the newest model is at most 20% faster, with the same amount of memory! Heck, last time I got a desktop it had an AMD Phenom II X6 1055T: for the same amount of money, you'd now get an i5 8500 or a Ryzen 5 1600x. The performance gain? Well judging by the Passmark scores, it would be a jump from ~5000 to ~13000. So that's roughly 2.5 times as fast in eight years, with a jump from 45nm to 14nm. Memory progression is about the same. Considering that the progression is slowing down, it'll take quite a few decades before the performance required for current AI is generally available.
Thirdly, the stuff which is impersonating AI isn't actually intelligent. Every current solution focuses on what basically boils down to a very complex set of linear regressions and similar methods. There is no adaptability, no sense of context, not even the slightest form of creativity. Everything which we consider to be intelligence is lacking. This isn't really a big surprise, because we simply don't know how it works in living creatures. Sure, we know how the neurons work, but the high-level understanding is lacking, so how could we even begin to imitate it or even improve on it? Sure, we could try simulating an actual brain, but a Blue Gene/L can only simulate half of a mouse's brain at one-tenth of the speed - and that's assuming the simulation is actually correctly modelling the biology.
So no, I don't think we're on the brink of an AI revolution. We might have something with the intelligence of a service dog in a decade or three, but even that would probably be a stretch. Unless a few very major breakthroughs occur, AI is going to be a bit of a disappointment.
What you need to look at is the raw power of internal combustion engines, and the weight they add. Versus the raw power of muscles of birds.
You need to look at the raw number of neurons firing in a human brain. This is the analog device that is our benchmark.
You need to look at the speed each neuron, of which there are a hundred billion, fires at: 200 hertz maximum, and more like 0.1 hertz on average.
you need to look at the speed of propagation: 8 orders of magnitude slower than the interconnects in data centers.
You are looking at a bird in flight, and you are looking at an internal combustion engine that outputs 1,000,000 times more power than its wings, and you are saying, "But our results are abysmal."
We will, and very soon, match that output. And far, far exceed that output. For sure we will. Because it's just analog meat. And our computers are faster and wider.
The poor state of algorithmic research is due to researchers having small access to the algorithms in question.
We have all of the machinery for "flight" (intelligence).
Personally I also think that researchers are impatient. Humans takes years to bootstrap to where they can understand the world as well as a 3-year-old. Specifically, humans take three years to understand or communicate as well as a 3-year-old. 36 months.
Parity with human brains isn't the goal. The goal is achieving 144 months (12 years) of brain development closely assisted and supervised by parents and teachers, especially within the first few years, within 9 hours of unsupervised learning.
But it will be done. You have the facts there in black and white.
I don't have to disbelieve the possibility of the existence of humans to doubt that flies could create a human.
I would say firstly, we don't have to know how to achieve it exactly, since once we have the hardware we can just experiment. The human source code is about seven hundred megabytes (fairly precisely, DNA is almost precisely 2 bits per base pair) so somewhere in that is the encoding for human intelligence, since humans inherit it. So we know that there is some < 700 MB program. Now 700 MB is a ton of source code, but it's not an unimaginable amount. We know some of the compiled results, as we can see how human brains are built actually. So there's lots of indications that we will be able to get similar results once we have the hardware for it - which we do.
Secondly I would point to many of the specific advances reached using machine learning as excellent indications that we are quite near the kind of algorithmic breakthroughs that are necessary. Since the topology of these algorithms is in many cases closely inspired by human neural topology, and many algorithms achieve many similar results, it seems quite likely that we would be able to reproduce some of these effects experimentally.
More than "likely", in many domains the computers have done just that.
Won't be a problem if we use CRISPR to make ourselves smarter first!
It's not slow at all. And you are obsessed with meat.
Currently still in the lab phase, I think.
That is literally 8 orders of magnitude slower than processers. 8 orders of magnitude is the size of difference between a grain of sand and 1.2 short tons.
In the time that your neuron moves 1 grain of sand, a computer can move 1.2 short tons worth of sand 1 grain at a time (talking about just cache here) - because it's switching at 4 GHz instead of 10 hertz.
That is enough to make a lot of back and forth trips. Now, you have to make a lot of these back and forth trips because the brain is extremely parallel. But the difference between these extremely slow biochemical firing speeds and gigahertz light speed enables this.
The speed of propagation in the brain is about 100 meters/second [2]. The speed of propagation between two fiber optic end-points is about 299 792 458 meters/second. That gives you 7 orders of magnitude to play with. Or, in other words, it can let you take seven orders of magnitude more trips across a data center (or go that much farther) before you miss a real-time deadline for when you need to address a neuron by.
In the future it's pretty clear that computers are not going to think as fast as the human brain when doing similar calculations using a similar topology, but, for example, 10,000 times faster than real-time.
This really shouldn't surprise anyone growing up in technology. Brains are slow, mechanical devices. They're not digital and they don't do signal processing at the speed of light.
It's taken us some time to catch up because a hundred billion neurons is still a lot of neurons, and algorithm development using similar topologies has been slow.
[1] https://aiimpacts.org/rate-of-neuron-firing/#Estimates_of_ra...
[2] https://hypertextbook.com/facts/2002/DavidParizh.shtml
All that aside, how does general cognition emerge from speed? Why hasn't anyone made a "slow" version yet?
The reason I mention the bit level is because if we knew for a fact that the "code" that ran the structure of the brain fit in, say, 64 KB (if we knew this because the whole human genome fit in 64 KB) we would have long-since tried to emulate it.
Instead it is "guaranteed" (in a sense) to be a subset of about 700 MB of bits. That's the human source code.
It just gives you a sense of the scale of complexity that might be involved there. (700 MB is quite a lot of source code, don't get me wrong.)
Look at the number of "mechanical" (biochemical) parts making up the human brain, their blueprint, the size they're contained in.
Finally look at the results that analogous machine learning algorithms are making with (albeit much simpler) neural net topology models.
The idea that nobody will ever match it is like looking at a mechanical watch and saying no digital model of it could ever come up with similar emulated results. What.
When you look at the size of the datasets involved you will see that it is exactly the sort of thing we can expect to be emulating now.
Regarding convenience, comfort and entertainment, we are good enough — A person can get all sorts of luxury and services without ever leaving one's home — but technology that encourages us to treat each other better, feel better about ourselves, and removes barriers to physical and social mobility, is what really excites me anymore.
Also E-ink HDMI screens with fast refresh rate.
Edit: These downvotes remind me of how the greeks drowned Hipassus for talking about the existence of irrational numbers lol.
The biggest part: it'll make people better at asking sophisticated questions and understanding how important that is.
I think it'll be as big a shift change as the Internet was.
I would love to ask specific questions instead of translating everything into multiple queries of "search engine speak".
Every time we find an edge case (or a noob question) that lands us on stack exchange, we could just ask an AI directly to look up the relevant documentation or scour through thousands of bug reports and workarounds.
Stage two describing the solution you want generated code for, or stage three stating the problem, or stage 4 finding problems and solving them.
And it never shows ads.
There's a reason that crackhouses in Vancouver cost $1 million, and it's not because they're made of gold. You need to build density to reduce the share of land cost per dwelling, anything else is a gimmick.
Or (meaningfully) make proximity less important.
I think the change will be similar in magnitude to the invention of the internet.
https://www.youtube.com/watch?v=YJg02ivYzSs