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."
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
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.
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.
Neural networks are good at generating correlations, but that's about all that they're good for.
No... What a supervised neural net does, in full generality, is to tune a massively parameterized function to minimize some measure of it's output error during the training process. It's basically a back box with a million (or billion) or so knobs on it's side than can be tweaked to define what it does.
During training the net itself learns how to optimally tweak these knobs to make it's output for a given input as close as possible to a target output defined by the training data it was presented with. The nature of neural nets is that they can generalize to unseen inputs outside of the training set.
The main limitation of neural nets is that the function it is optimizing and error measure it is minimizing both need to be differentiable, since the way they learns is by gradient descent (following the error gradient to minimize the error).
The range of problems that neural nets can handle is very large, including things such as speech recognition, language translation, natural-langauge image description, etc. It's a very flexible architecture - there are even neural Turing machines.
No doubt there is too much AI hype at the moment, and too many people equating machine learning (ML) with AI, but the recent advances both in neural nets and reinforcement learning (the ML technology at the heart of AlphaGo) are quite profound.
It remains to be seen how far we get in the next 20 (or whatever) years, but already neural nets are making computers capable of super-human performance in many of the areas they have been applied. The combination of NN + reinforcement learning is significantly more general and powerful, powering additional super-human capabilities such as AlphaGo. Unlike the old chestnut of AI always being 20 years away, AlphaGo stunned researchers by beng capable of something *now* that was estimated to be at least 10 years away!
There's not going to be any one "aha" moment where computers achieve general human-level or beyond intelligence, but rather a whole series of whittling away of things that only humans can do, or do best, until eventually there's nothing left.
Perhaps one of the most profound benefits of neural nets over symbolic approaches is that they learn their own data representations for whatever they are tasked with, and these allow large chunks of functionality to be combined in simplistic lego-like fashion. For example, an image captioning neural net (capable of generating an english-language description of a photo) in it's simplest form is just an image classification net feeding into a language model net... no need to come up with complex data structures to represent image content or sentence syntax and semantics, then figure out how to map from one to the other!
This ability to combine neural nets in lego-like fashion means that advances can be used combinatorial fashion... when we have a bag of tricks similar to what evolution has equipped the human brain with, then the range of problems it can solve (i.e. intelligence level) should be similar. I'd guess that a half-dozen advances is maybe all it will take to get a general-purpose intelligence of some sort, considering that the brain itself only has a limited number of functional areas (cortex, cerebellum, hippocampus, thalamus, basil ganglia, etc).