Will the Real AI Critics Please Stand Up?
I'm having great trouble finding any citeable work that argues that artificial intelligence is completely impossible. People throw kiwis at AI theory in its current state, or the philosophy of functionalism, but every single argument I can find stops short of outright denunciation.
For instance, Gerald Edelman, winner of the 1972 Nobel Prize in Medicine and coiner of the term "Neural Darwinism", argues that "AI" is impossible, expelling much hot air on the subject, but then it turns out that he believes, "It seems reasonably feasible that, in the future, once neuroscientists learn much more about consciousness and its mechanism, why not imitate it?", and remarks "We construct what we call brain-based devices, or BBDs, which I think will be increasingly useful in understanding how the brain works and modeling the brain. But it also may be the beginning of the design of truly intelligent machines." So that's not very anti-AI. Edelman was also quoted in John Horgan's recent anti-Singularity piece in IEEE Spectrum, the "Consciousness Conundrum", in support of the idea that AI is difficult. But if he thinks AI is so difficult, why is he spending time and money on brain-based devices, which are steps towards AI?
According to his Wikipedia article, Hurbert Dreyfuss, author of "What Computers Can't Do: the Limits of Artificial Intelligence", argues "that we cannot know (and never will) be able to understand our own behavior in the same way as we understand objects in, for example, physics or chemistry: that is, by considering ourselves as things whose behaviour can be predicted via 'objective', context free scientific laws." But then the article also states, "he doesn't believe that AI is fundamentally impossible; only that the current research program is fatally flawed. Instead he argues that to get a device (or devices) with human-like intelligence would require them to have a human-like being in the world, which would require them to have bodies more or less like ours, and social acculturation (i.e. a society) more or less like ours."
Very confusing, but I'm not done yet. Next comes famous physicist and Hawking-collaborator Roger Penrose and his poorly thought out theories on consciousness. Penrose argues that quantum decoherence in neural macrotubules is essential to our intelligence and consciousness. This was decisively refuted by our friend Max Tegmark in 2000, who calculated that the timescale of neuron firing and excitations in microtubules is slower than the decoherence time by a factor of at least 10,000,000,000. Still, although Penrose fusses about the alleged non-algorithmic nature of intelligence throughout his books on the topic, according to a review by Robin Hanson, "Penrose grants that we may be able to artificially construct conscious intelligence, and "such objects could succeed in actually superseding human beings." But he thinks "algorithmic computers are doomed to subservience." Another thinker who objects to the mainline AI philosophy and approach but doesn't actually believe that AI will never be possible if we aren't creative enough.
There's more stuff out there. Paul Churchland says, "Classical AI is unlikely to yield conscious machines; systems that mimic the brain might". More of the same. Copy a certain type of big-headed ape exactly, and intelligence will pop out, but if you try anything else, you'll fail. Even Searle, the king of AI criticism, acknowledges that "machines with internal causal powers equivalent to those of brains" could think. I'm not sure precisely what he means by this, but by bothering to say something besides humans, even Searle seems to believe that some form of Artificial Intelligence is possible.
Where are the people saying "AI will never happen" or "only human beings can think"? I can find hundreds of references made by laypeople on various forums, but they generally don't present coherent arguments, they just throw out their opinions.
If no philosopher, cognitive scientist, or computer scientist is willing to claim in public that true AI is impossible, then isn't this an important finding in and of itself? If it is, then I totally get the credit.
August 14th, 2008 - 21:43
You can find more people arguing that AI will never be conscious (absurdly, I would say), but philosophical zombie AI still has the massive practical effects (potential for extraordinarily rapid economic and technological development, risk of human extinction, etc).
August 14th, 2008 - 22:18
The fundament of “intelligence”, whatever that is (there are other vague terms such as self-aware, conscious, sapient) should be reducible to a core device we could label “an abstractor”, much as a devices that abstracts as much as a computer computers. Such a hypothetical abstractor would be capable of almost nothing but certain mathematical functions – it would need specialized subroutines to be able to function meaningfully. Obviously to act as a human a CAP “core abstractor processor” would need a sensory set of utilities, maybe a communicator suite and a human personality simulator. I can literally see these software sets being developed by a player like Microsoft, and by competitors. You’d have a rather autist nonhuman CAP, with human simulation software patched on. The result would be a thinking entity that can interface with humans. I don’t know if such a simulate would actually require a body (which would also need specialized software tools to operate – version 1.0 would be a bit C3PO clumsy I suppose) or, if the sim wasn’t “rendered” with a body – a good virtual reality environment.
How much data would a VR that adequately simulates all sensory features of a human body? Holy shit, that might be gigabits per second, for all limbs, senses, nerve endings, smell, sight, temperature, balance. Gigabits of very meticulously rendered data. And realise that synthizing such detailed data would be equivalent in complexity as processing it. If we would be capable (probably indirectly through complex tools) of creating a human-like android (as versatile as a schwarzie terminator, to take one example from fiction) I think synthesizing input (rendering an adequate VR with all sensory data) would be as tricky as creating a device that can meaningfully process it.
I can only hope we come there through cooperative corporate effort – lots of software tools that interact, sometimes imperfectly. We want as many people involved in this research as humanity can muster. Obviously that’ll be decades away at the least.
August 15th, 2008 - 01:56
Anyone can keep moving the goalposts, Michael. “true” is a weasel word.
August 15th, 2008 - 03:24
I think you’re right. At least I can’t think of any published research papers that argue that AI is strictly impossible. As you point out, some come close, but seemingly they always leave some window of possibility. When people tell me in person that they think AI is impossible, if I push them a bit what they usually mean is that it’s very unlikely to be achieved in the next hundred years. But 500 years from now… yeah, maybe.
August 15th, 2008 - 04:42
I define “true” as anything a human can do, including consciousness, “free will”, etc.
August 15th, 2008 - 04:54
Michael Bishop (2003). Dancing with pixies: Strong artificial intelligence and panpsychism
This paper argues against computationalism by showing it implies panpsychism. It’s hard to see how one could build a computer and _know_ that it had internal experiences if it depends on something other than what algorithm is running.
I think he would still grant that it’s probably possible to build a conscious (i.e. with qualia) AI if you made it out of the right stuff but that it’s impossible to know whether you’ve succeeded or not.
August 15th, 2008 - 05:13
http://www.soapboxxer.com/soapboxx/11420?page=6
25% of these people think AI is impossible.
August 15th, 2008 - 10:49
Michael, after perusing some of the comments on that site, it becomes clear that the label of “impossible” is too fuzzy for a lot of the folks.
It looks like some of the individuals stating said “impossibility” are stating that the “impossibility” stems from their view on design & development/approach. In some cases, it would appear it is a function of education.
On the other hand: Why even argue? When it comes it will come. Speculation on the subject is a waste of brain resources, imho. I would use all that wasted brainpower to do something to prepare for a future where if it comes, I am ready…And if it doesn’t I am still positioned to be in a position of advantage.
August 15th, 2008 - 14:22
Once one accepts determinism, then the possibility of true AI according to Michaels definition is trivial: Just simulate every molecule and all its forces (we are not talking about practicability here!), of a human being, and you have it!
OTOH, if you believe in free will and nondeterminism, then this becomes more of a spiritual argument (is the soul describable in physical terms? Where is the consciousness situated?).
However, this viewpoint is uncommon under scientists, as it assumes forces outside the physically describable world.
August 15th, 2008 - 22:36
Of course AI or super intelligence will arise, but what I see being neglected in these arguments is the fact that our intelligence/consciousness is based off a living system. The question is “Does intelligence necessitate life?”…If so, then we should be designing artificial living computer systems or hybrid life/machine computational systems.
August 16th, 2008 - 02:04
Penroses ideas of a quantum basis for consciousness may turn out to be wrong, but I wouldn’t describe them as “poorly thought out”.
Also Edelman is not exactly an AI refusenik and his “Darwin” automata have competed successfully in soccer matches against CMU robots. Edelman’s approach is more one of brain emulation rather than traditional symbolic AI which he rejects.
What these people are usually refuting is that AI is possible with currently popular techniques. Often if questioned in more detail they will admit that smarter than human machines might in principle be possible if only their preferred approach were followed (quantum, brain emulation, etc).
August 17th, 2008 - 07:46
The only real argument against the possibility for artificial intelligence is a semantic one: intelligence is the set of patterns including the evolutionary algorithm, the laws of physics, the meddling of some preferred deity, etc. A machine can uptake new capacities, but can never truly outstrip the capacities of it’s creator, who lives in an often symbiotic relationship with his techno-children.
The way we approach science, mathematics,etc. are highly influenced by linguistics. We’re asking the wrong questions. We make no progress because we create false dichotomies; that the anthropomorphic view is threatened by the ever-increasing prevalence of thinking machines. Let’s ask new questions.
August 18th, 2008 - 19:03
I like Paul. He just won the smartest post award.
August 19th, 2008 - 05:44
Yes, it’s an important finding and you get the credit.
August 19th, 2008 - 18:45
Suppose that we were interested in creating artificial rabbits, rather than AI.
The Artificial Rabbit proponent would say “I think we can make an artificial rabbit that’s just as alive as a natural rabbit – using MEMs and integrated circuits, and relying on the advance of Moore’s Law to get us sufficiently complex hardware in a few more years.”
Skeptics cry “Impossible!”, pointing out that rabbits are incredibly complex, and while we have some understanding of their physiology and behavior, it seems unlikely that any human or team of humans will ever understand the nature of rabbits sufficiently to build the true equivalent of a rabbit from scratch. “There might even be chemical processes going on in real rabbits, that can’t be practically emulated electro-mechanically.”
To which, AR proponents would say – “We really wouldn’t even need to fully understand a rabbit to make an artificial one. In principle, we could just copy a rabbit onto the electro-mechanical hardware, by replacing its cells one at a time with a functionally exact electromechanical equivalent. It’d keep functioning perfectly, until eventually we’d have a fully artificial rabbit, equivalent in every way to a live natural rabbit.”
Skeptics grumpily admit that such a thing might, in principle, be possible, as impractical as it sounds.
AR proponents: “See – all those people saying AR is impossible admit that we’re right!”
Asked by a skeptic “But haven’t you been claiming that you think an AR is only a few years away? And that an AR will quickly become vastly superior to rabbits – in fact superior to humans?”
AR proponents would respond “Yes – but we’ve got a plan. Instead of trying to build a whole rabbit from scratch, we’ll just build an artificial rabbit embryo – much simpler – and let it develop into an AR. Then we’ll just let it keep developing past that point, and it’ll quickly become a super-rabbit. We don’t have the artificial embryo figured out yet – but it can’t be that hard, since it’ll be much simpler than a full AI.”
:-)
August 19th, 2008 - 18:59
Oops – that last word should have been “AR”. Obviously…
But to be serious for a moment – if a seed AI is the equivalent of an embryo, we should consider that an embryo is essentialy a highly data-compressed, self-unpacking expression of a full-grown entity.
Programming a “seed” might be the equivalent of trying to develop a software application by coding it in its compressed form!
Or if the seed is the equivalent of DNA – seeming orders of magnitude simpler again – it’d be the equivalent of data compressed AND encrypted. DNA needs a complex external system (egg cell, rabbit womb) to decode it, before it can start to un-pack itself.
Not saying seed-AI is impossible – just saying that that common analogy seems to be radiating warning signs.
August 20th, 2008 - 02:00
When you tell the laypeople that computers will be smarter then people their response is “aren’t they already?” (they problem they have is with the implication of this statement.) And in a way they are right, computers dominate more and more in tasks that were used to be thought the domain of people only (not to mentions the ones that they were superior to us from the very beginning.) No self respecting scientist will claim that AI is impossible as this would contradicts materialism. And even theology does not role out other intelligent beings. The day when intelligent machines surpass our ability is close. And the day when everyone will realize that is even closer. Looking for a scientist that will predict that AI will not happen is like looking for a weatherman that will predict that it is not going to rain just before the big storm.
August 22nd, 2008 - 21:42
You ask me to stand up.
The reason it is impossible is because we have no capacity to design it. We can only evolve it, and that’ll take some time.
August 23rd, 2008 - 11:42
Any one who say that AI is imposible is saying two things:
a) human being have unphysical soul, which is nessesary for intellegence.
b) human does not evolve from apes by natural selection, but had reseived their intillegence from God.
If one agree with darvinian evoution, he should agree that we could create new intelligens by selection of shimps, though it could take many thousand years.
August 25th, 2008 - 17:50
Tom Craver, what I love about your wonderfully clever comment on AR, artificial rabbits, is that both AIers and skeptics like me can find support for their position in it. However, you seem finally to end up in the skeptic camp, when you mock the hand-wavy evolutionary/genetic algorithm approach of some AIers. My skepticism toward AI is based not on theoretical arguments like those of Penrose or (even less) the hopelessly muddled Edelman (although I’m happy to quote him for my insidious ends) but on the embarrassing track record of the field. Only adolescent sci-fi fanatics (and obviously there are plenty on this site) can believe based on progress in voice-image-etc recognition that software programs are approaching anything akin to what makes our minds tick. I’m not proud, just stating a fact.
August 25th, 2008 - 17:52
Oh, and Michael A., another very thoughtful stimulating post. The Singularistists don’t deserve you. Hope we cross tracks at the Singularity Summit.
August 26th, 2008 - 14:52
Tegmark’s refutation was not decisive:
http://www.ncbi.nlm.nih.gov/pubmed/12188753
That being said, do not take this post as an endorsement of Penrose-Hameroff.
September 3rd, 2008 - 16:25
Responding to the original post.
I think alot of people reject A.I. more for emotional reasons then anything then logical ones. I’ve heard you talk about the “paranoia†of the west towards new technology (and this definitely deserves to be elaborated on.) This paranoia exists for a reason.
I think many possible technologies represent an assault on the Western view of identity (which largely came from Christianity, and certain Pagan religions before that.) The reason the East doesn’t seem so worried might be because they come from a different philosophical tradition (largely Hinduism and Buddhism) in which individual identity is not as important.
The idea of Artificial Intelligence, Genetic Manipulation, Brain Manipulation, and Personality Manipulation (whatever form they take)is terrifying to the Western mind; which puts more weight in the idea of individuality.
September 8th, 2008 - 10:09
You cannot dispute the inevitable. AI is an apparent truth. Like evolution, which the creationists are still trying to argue about. When you look at the development of life, and then intelligent life, the only thing that could logically succeed us is AI. How else will we be able to link our minds up and become one global consciousness?
October 25th, 2008 - 14:36
On the first day of my introductory computer science class my professor, Tim Farage, introduced the concept of the singularity by asking if anyone was familiar with the concept. Kurzweil’s singularity was given by another student, then I talked about friendly AGI as a potential pathway. Next the professor stated their had been formal mathematical proofs AI that was human intelligent is impossible. He didn’t give anything specific but I can ask him after my test this tuesday. He has a blog.
http://www.timfarage.blogspot.com/
October 31st, 2008 - 09:12
By the time we get intelligent enough to recreate human intelligence, we will be far more advanced. I don’t think AI will ever catch up to us, it’ll always be two steps behind.
November 12th, 2008 - 03:49
It is not out-of-date information? Because I have other data on this theme. http://hotsalego.com/map.html
November 13th, 2008 - 02:16
Some have argued that Godel’s incompleteness theorem prevents Computational devices from reaching levels that intelligent beings can reach. But ultimately, the question has too many undefined terms. What is intelligence? What is life? What is consciousness? What is artificial intelligence? We don’t have formal definitions for any of these. If thoughts and consciousness are nothing more than the result of deterministic processes inside our brains, then AI should definitely be possible. This is to say that life forms don’t actually have the ability to make choices, but are bound to choose one way or another based on the processes of the brain.
September 2nd, 2010 - 05:14
Which blogs do you read regularly?
November 12th, 2011 - 01:58
I am glad for commenting to make you understand what a awesome encounter our princess encountered going through yuor web blog. She realized plenty of things, not to mention how it is like to have a marvelous giving character to make other individuals quite simply gain knowledge of certain advanced issues. You undoubtedly surpassed readers’ expected results. Many thanks for delivering these necessary, trustworthy, informative and also unique tips about this topic to Jane.
May 6th, 2012 - 12:19
This certainly clarified my problem, thank you!