Slashdot Mirror
Wired Founding Editor Now Challenges 'The Myth of A Superhuman AI' (backchannel.com)
Wired's founding executive editor Kevin Kelly wrote a 5,000-word takedown on "the myth of a superhuman AI," challenging dire warnings from Bill Gates, Stephen Hawking, and Elon Musk about the potential extinction of humanity at the hands of a superintelligent constructs. Slashdot reader mirandakatz calls it an "impeccably argued debunking of this pervasive myth." Kelly writes:
Buried in this scenario of a takeover of superhuman artificial intelligence are five assumptions which, when examined closely, are not based on any evidence...
1.) Artificial intelligence is already getting smarter than us, at an exponential rate.
2.) We'll make AIs into a general purpose intelligence, like our own.
3.) We can make human intelligence in silicon.
4.) Intelligence can be expanded without limit.
5.) Once we have exploding superintelligence it can solve most of our problems...
If the expectation of a superhuman AI takeover is built on five key assumptions that have no basis in evidence, then this idea is more akin to a religious belief -- a myth
Kelly proposes "five heresies" which he says have more evidence to support them -- including the prediction that emulating human intelligence "will be constrained by cost" -- and he likens artificial intelligence to the physical powers of machines. "[W]hile all machines as a class can beat the physical achievements of an individual human...there is no one machine that can beat an average human in everything he or she does."
Kelly proposes "five heresies" which he says have more evidence to support them -- including the prediction that emulating human intelligence "will be constrained by cost" -- and he likens artificial intelligence to the physical powers of machines. "[W]hile all machines as a class can beat the physical achievements of an individual human...there is no one machine that can beat an average human in everything he or she does."
Anything is possible in 10-20 years, just give me all your money!
Given that our knowledge of the computational complexity of a single neuron is growing steadily, I think it's safe to say your FPGA cell estimate for a neuron was significantly too low.
For example, scientists now know that one single neuron (of certain types) is an entire neural network all by itself. Dendrites with multiple localized spikes communicating with each other and with other cells. Ultimately performing non-linear computation prior to forwarding any signal to cell body.
there is no one machine that can beat an average human in everything he or she does
Neither can most humans. There is no such thing as an average human. Every individual human specializes, and increasingly so as they get older (or they do not improve). It is a pervasive strawman to require AIs to "beat" an average human when the same quality isn't used to judge humans.
Those who do not learn from commit history are doomed to regress it.
The first three assumptions in this article have already been met sufficiently well enough to debunk the Wired article. AlphaGo has displayed superhuman intelligence in the first three areas of assumptions. 1) AlphaGo exploded on the scene by beating world class Go players much faster and much earlier than expected. Exponentially is a loaded word. e^0.0000001 is an exponential growth rate. So let's not quibble about how exponential the growth rate is.
2) AlphaGo is a general purpose learning tool. Just listen to the lectures and articles penned by the DeepMind team.
3) Alphago has displayed human-like intelligence, as claimed by the Go professionals it has played. They have said that AlphaGo plays like a human player.
4) If you take the fourth assumption literally, AI's intelligence is going to expand infinitely. Talking about infinity in human terms is unreasonable. Yes, AI's intelligence will expand.
5) The fifth assumption can be argued many ways. Some problems are not solvable due to their paradoxical nature. Other problems are subjective and are uniquely unsolvable by some individuals, but not by all individuals. It is a matter of time before a general purpose AI program will solve subjective emotional problems. Whether all human beings accept the solutions is subjective and open to speculation.
The human population is composed of experts, with divisions of labor. It is not unreasonable for AI programs to have areas of expertise.
AlphaGo is not a "general purpose" learning mechanism. It won't ever write sonnets meaningful to humans, or be able to to dance, or even employ symbolic differentiation.
It is a really nice toolset, and it is able to solve a task which is difficult for humans, but so does Google or your high-school calculator when you calculate sin(1.2).
It won't ever go beyond the computational underpinnings of playing Go-like games: evaluating game positions and calculating game trees. It won't ever say 'forget it, I'd rather be drinking beer with my buddies', which is an intelligent thing to do for most of us with respect to playing Go.
There's nothing human-like about AlphaGo, except that it solves a problem relevant to humans; the calculator example comes in mind.
I'd be thrilled to know what kind of specific major human problems you'd consider AI-approachable, because I currently only see a bunch of more or less advanced mechanisms that are fine-tuned to solve very specific computationally well-defined problems, and most human problems are not computationally well defined.
If people like Bill Gates and Elon Musk are unrealistic in one direction, this person seems unrealistic in the other direction. He's basically betting against technological progress. And that's usually a losing bet, at least over long enough time periods.
1.) Artificial intelligence is already getting smarter than us, at an exponential rate.
Computers are already better than us at many tasks. That's been true for ages. And they're continuing to improve while we aren't. The set of tasks that computers are better at is constantly growing. I don't know of any fundamental limits to prevent them from eventually becoming better than us at the remaining tasks too. So it seems pretty likely they eventually will.
2.) We'll make AIs into a general purpose intelligence, like our own.
It's hard to even define what a "general purpose intelligence" means. But anything a human brain can do, computers will probably eventually be able to do it too.
3.) We can make human intelligence in silicon.
We can certainly make intelligence in silicon. We've already done it. Whether you consider it to be "human intelligence" or "inhuman intelligence" is kind of beside the point. If a computer can do something, whether it does it in the same way a human does is just an implementation detail.
4.) Intelligence can be expanded without limit.
I don't know of anyone who's claiming that. Where does he get this from? Anyway, the claim isn't that computers will advance without limit, only that they'll surpass humans.
5.) Once we have exploding superintelligence it can solve most of our problems...
Um, no. That's not at all what they're claiming. We certainly hope that it will solve many problems, but Gates, Musk, et al. are warning it could also create huge problems.
emulating human intelligence "will be constrained by cost"
Computers are cheap, and getting cheaper all the time. Humans are expensive and staying expensive. That's why automation has become such a big deal. Here again he seems to be betting against technological progress.
"I'm too busy to research this and form an educated opinion, but I do have time to tell everyone my uninformed opinion."
Unfortunately Sam Harris is bad at math. He claims "It's crucial to realize that the rate of progress doesn't matter, because any progress is enough to get us into the end zone. We don't need Moore's law to continue. We don't need exponential progress. We just need to keep going.". It seems he has never seen a monotonically increasing, yet asymptotically bounded function. However, that is exactly the kind of progress we are seeing in older technologies, e.g.: Airplanes stay at almost exactly same speed (because going past the sound barrier would use lots of energy) and get slightly more efficient each year, but will never get to the point where they can operate almost without any fuel or other large energy source, simply because the laws of physics don't allow that kind of progress.
But even if the possible progress is not bounded, it is still not guaranteed that we will get there. It can still take so long, that it never going to happen before human civilization is completely destroyed by some disaster. Or it could simply be stopped by economics as further improvements can easily get so expensive or tiny, that the likely benefits from pushing the research further can not offset the cost.
Harris also seems to think that general AI is ineviatable, because we want to make progress towards things such as things such as cureing cancer or Alzheimer. But it is not clear that such an achievement actually requires general superhuman intelligence. It likely requires superhuman intelligence, e.g.: the computers that simulate protein folding way better than any human could ever do, but not necessary general intelligence. Specialized artificial intelligence seems to be much easier to achieve and is at the same time likely almost as good as general intelligence for topics such as those. You don't need to develop an artificial general intelligence to cure cancer, if you already developed a specialized artificial intelligence that is able to find a cure.
Imagine what could happen when a huge neural net is applied.
The problem with huge neural nets is training them. The more possiblities a network has, the harder it becomes to train it. Large parts of the progress in the last few years were made by finding clever constraints on the network in order to make them easier to train.
Jan
Holy Shit! Donald Trump is a Slashdotter! Don't worry Donnie. They'll be working on bigger hands next!
Guns don't kill people; Physics kills people! - John Lithgow as Dick Solomon on Third Rock From The Sun
This kind of problem will be very visible in healthcare. Human doctors will self censor for fear of contradicting AI and taking the wrong choice, that ends up badly for the patient. Because AI is right most of the time, who will have the courage of saying otherwise? Saying truth to AI could cost a person their job. Many doctors stop giving honest feedback the moment they hear another doctor has given a diagnosis out of solidarity with their colleagues or fear of the consequences of making enemies.
Human (or just vision in general) is the best example. It accounts for 30% of the brain capacity. At one end, you have the human eye with a retina consisting of 100 million rods and cones. Then just in that space of a 5mm disc, there are seven layers of processing used to do contrast detection between colors and intensity along with edge detection. The optic nerve takes the compressed information from a thousand areas then passes it through to the brain into two paths; one to identify where objects are, the other to determine what the objects are and their orientation. Understanding what just a single region or layer of brain cells does leads to dozens of papers being published and advances in digital photography (image stabilization, motion correction).
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Indeed!
And one of the myricals in this is: if an object is about to hit your eyes or comes close by, the reflext to close the eyes and raise your hands etc. is triggered _before_ that information has even reached the brain/visual cortex.
The signal processing in the eye can bypass the visual cortex to trigger protective actions.
Cost free eBook I read (by iBook/Kobo/Amazon/ObookO/Gutenberg etc.): "The Green Odyssey" by Philip Jose Farmer.
1) AlphaGo exploded on the scene by beating world class Go players much faster and much earlier than expected. Exponentially is a loaded word. e^0.0000001 is an exponential growth rate. So let's not quibble about how exponential the growth rate is.
Particularly as we can't really measure intelligence. But "exponential" has a meaning, and it means a steady rate of doubling.
2) AlphaGo is a general purpose learning tool. Just listen to the lectures and articles penned by the DeepMind team.
No, it's a narrow AI. In the end, it's simply doing math. It's not "thinking" in any sense of the term. It's just able to hold many more probabilities in its memory than a human, and play them out much faster.
3) Alphago has displayed human-like intelligence, as claimed by the Go professionals it has played. They have said that AlphaGo plays like a human player.
That is what you call anthropomorphizing. The human players are simply projecting onto the machine.
4) If you take the fourth assumption literally, AI's intelligence is going to expand infinitely. Talking about infinity in human terms is unreasonable. Yes, AI's intelligence will expand.
"Infinite" simply means there's no limit. We don't know whether or not there's a limit to intelligence, but since the universe is finite, there would seem to be a limit to the things one could know. Infinite intelligence is our notion of God. What are the odds that infinite intelligence is also mythological?
-- sudon't
Air-ride Equipped
Given that our knowledge of the computational complexity of a single neuron is growing steadily, I think it's safe to say your FPGA cell estimate for a neuron was significantly too low. For example, scientists now know that one single neuron (of certain types) is an entire neural network all by itself. Dendrites with multiple localized spikes communicating with each other and with other cells. Ultimately performing non-linear computation prior to forwarding any signal to cell body.
Right you are. The absolute give-away (in addition to the ridiculous low-ball answer he provided) was "... that was pretty straightforward..." which shows the Dunning-Kruger Effect in full bloom. He had no idea now little he knows about the subject.
The example I like to use to illustrate how much smoke is being blown about this my tech types is the model organism Caenorhabditis elegans. This 1 mm long nematode has had every one of its 302 neurons in its nervous mapped out, including all connections to every other neuron, as well as the process of development from the initial fertilized egg - we have mapped out exactly how the nervous system develops (indeed every one of the 959 cells in its body have been similarly traced out).
Given this complete map of C. elegans nervous system we must have a spiffy computer of the little worm's "brain" able to replicate its behavior? Right?
Not even close. So far we cannot accurate model the behavior of even a single neuron in C. elegans. Even one single neuron represents computational complexity that we are still trying to understand.
Starships were meant to fly, Hands up and touch the sky - Nicky Minaj