Tip of my hat to akkartik for pointing this article out to me.
Thank you. Borderline frightening.Likewise, when a computer program beats a grandmaster at chess, the two are not using even remotely similar algorithms. The grandmaster can explain why it seemed worth sacrificing the knight for strategic advantage and can write an exciting book on the subject. The program can only prove that the sacrifice does not force a checkmate, and cannot write a book because it has no clue even what the objective of a chess game is. Programming AGI is not the same sort of problem as programming Jeopardy or chess.
Some, such as the mathematician Roger Penrose, have suggested that the brain uses quantum computation, or even hyper-quantum computation relying on as-yet-unknown physics beyond quantum theory, and that this explains the failure to create AGI on existing computers.
I'm not entirely convinced that its even possible, in principle, to program a being that could have an imagination. The author's contention is that it is, but only if we can develop the philosophical framework to decide what it means to have a creative thought, and thus the programmer might have a guide. I don't necessarily think he's wrong, but I think its incomplete for two main reasons. Firstly, consciousness isn't a function of a brain, but of an organism; your kidney, liver and heart all fundamentally contribute to your imagination. A disembodied brain is essentially a useless lump of matter. And secondly, if it even makes sense to compare computers and brains, then a brain is much more like an analog computer (though some neuronal outputs have a binary component), and I wonder whether digital logic can ever replicate its subtleties. The quantum suggestion is just silly. Penrose is famous. If he weren't, that suggestion would have been laughed off years ago. Quantum coherence certainly cannot exist at the temperatures and with the complexity of a body.
He lost me when he used "universality" to say this: That's the basis of the buildup to his main argument (philosophical framework). It seems flimsy. Maybe because I barely have a working knowledge of AI sufficient enough to understand this article. Anyway, the reason I liked his point about chess so much was because every approach to building AI, to depicting AI in film or literature, always starts and ends by metaphorically copying the brain. I feel that this may be impossible, but that there are other avenues to creating artificial intelligence.And yet ‘originating things’, ‘following analysis’, and ‘anticipating analytical relations and truths’ are all behaviours of brains and, therefore, of the atoms of which brains are composed.
My opinion - what's holding us up is simply that we don't understand intelligence yet (as we wish to create it - i.e. "like us") . That's all. it's meaningless to say that we won't have enough computing power, until we know at least the general outline of the problem - and we just don't. As the article says - "Expecting to create an AGI without first understanding how it works is like expecting skyscrapers to fly if we build them tall enough."
I buy Davids premise of universality of computing but I fear there may be physical limits of binary computing i.e. the amount of matter in the universe and the time before its heat death. lets say I am modeling the interactions of a few cubic meters of hydrogen
5.4199274e+25 particles I need to know their location 3 variables orientation 3 more, speed, temp, etc so maybe only 10 things about each h2 pair.
Because this sort of thing is very sensitive to initial conditions and small errors in calculation lead to big differences later a "Double" (64-bits) just is not enough. Easy I will use 512-bits (Moore's law and all) . . . 5120 bits per pair * number of particles = 2.7750028e+29
so only about 2.8e+17 terabits or 3.5e+16 terabytes of storage needed for one
step.
Seagate promises 60 terabyte 3.5-inch hard drives within 10 years
5.8e+14 of these seagate hard drives should do the trick.
that is only 3.22233e10 miles of hard drives or 346.65 Astronomical Units
(the distance from the sun to pluto is 29) of course placing the hard drives in a straight line might be a mistake. so we will stack them in a 2.40822 mile cube. just for an instant of data we are at 58215044860:1 in digital storage.
volume wise. but hell with an earth sized hard drive we could describe 107963558147 cubic meters of hydrogen (a sphere with a diameter of 1.476 km) wait do we have to do some computing as well?
sorry about a arithmetic rant.
Definitely no apologies necessary. That's essentially what I meant below when I said numerical solutions can't really be described as correct. They can be useful approximations sometimes for some problems, given reasonable assumptions and well built difference equations. That said, there is certainly no such thing as an "analytic solution" to the brain. And given that numeric solutions depend wholly on good assumptions, there will never be a working brain model (maybe biochemically modeled, but certainly not 'conscious', in the sense we understand it). The Blue Brain Project has already wasted a billion dollars. Now the NIH (who is close to broke) is spending something like $50,000,000 to figure out the 'connectome', the set of all synaptic connections in an average brain ("a map of the mind," they like to call it--laughable, IMO). Maybe you can send your back of the napkin calculations to someone in power and help them realize the futility of their work.
has anyone read universality of computation? what are the axioms of the proof?
I think it could be worth a read.
Ken Arrow, John Nash and Vilfredo Pareto have wonderful math based on bullshit axioms. but I think I can buy that anything can be modeled as long as it does not turn into
"anything can be modeled = this model is correct."
Not just the solutions but the models themselves are incorrect i.e. they don't describe what they claim to describe.