Why Philosophers Should Care About Computational Complexity [pdf]
scottaaronson.com
scottaaronson.com
*> And while it was true, the critics went on, that a giant lookup table wouldn’t “truly understand” its responses, that point is also irrelevant. For the giant lookup table is a philosophical fiction anyway: something that can’t even fit in the observable universe! If we instead imagine a compact, efficient computer program passing the Turing Test, then the situation changes drastically. For now, in order to explain how the program can be so compact and efficient, we’ll need to posit that the program includes representations of abstract concepts, capacities for learning and reasoning, and all sorts of other internal furniture that we would expect to find in a mind. Personally, I find this response to Searle extremely interesting—since if correct, it suggests that the distinction between polynomial and exponential complexity has metaphysical significance.
This captures what I had intuitively thought regarding the link between computation and consciousness: consciousness is some sort of sophisticated, elegant computation embodied in human physiology.
Really? I'd think that the fact that someone saying 'lookup tables aren't thinking' without ever saying what thinking is would be presenting an obviously self-refuting argument.
Like a judge saying "I know crime when I see it: I don't have to write down what it is." It's simply not a basis for rational decision making, so it's certainly not a convincing assertion.
The fact that anybody thinks the Chinese room is an interesting philosophical quandary is far more interesting. It means these are primitive times and you don't actually need to be very smart to beat the competition.
There are genuinely competent people doing competent work in the field of philosophy. They just don't make up enough of the peer-review boards to throw out everyone who isn't as competent as them.
The continuing influence of mathematical logic and its attendant "general axiom to specific theorem, As Above So Below" style of thinking does not help.
Computer passes Turing Test, Human fails Turing test - vice versa - that's all there is to it.
I think we probably judge others to be conscious actors in more ways than this though. We probably accept that other people have "qualia" from the fact that we have it, and then we kind of learn that our parents have it, and so on, until it's not a huge leap to generalize that to all of humanity. Of course it's not like people have always assumed the same kind of "qualia" for all of humanity, in a lot of conflicts this has probably been thrown out of the window so it will be easier to kill your enemies. Either way, it's probably quite complex how we determine that other people are conscious, Turing tried to skirt the metaphysical issue (that we can never really know, this goes for both humans and other machines). I think this was a huge step in the right direction.
http://cogprints.org/499/1/turing.HTML
> The aim of the Turing Test is to say that judging machines based on...
That wasn't the aim at all, again I think you are reading way too much in to it. The aim was to pose a test that would fool people.
Alan Turing:
"an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes"
"The original question, "Can machines think?" I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted."
The test is a very simple thought exercise where he shows a machine will be able to pass for a human, and a human will fail to identify a machine. That's it. No questions of consciousness, metaphysics! In fact he dismisses such questions in the paper.
His belief was that in 50 years time, the words may have changed, and people's opinions about what constitutes thinking would have changed to the point where: someone could say "a computer thinks" without challenge. That hasn't happened, when somebody says "a computer thinks" they are saying it with tongue firmly in cheek. Maybe in another 65 years time his belief will come true.
As for public perception of consciousness or "thinking", time estimations are usually quite sketchy.
I think it's a thought experiment to show that machines are going to be able to fool an average human for some small length of time eventually. That's a lot less grandiose.
A mistaken interaction with an IRC bot or an automated phone service has the same net effect.
If I am reading what you are saying correctly, I surmise that you're saying that the Turing Test doesn't say anything about consciousness, but rather, is just about the probability of a machine being able to fool a human. I am saying that the reason why Turing even ponders the question is that there's no good way of definitely answering if a machine (or another human for that matter) is conscious or not. So that leaves the ability of being able to fool (or convince) other thinking machines that one is intelligent (or conscious) as the only viable metric.
I feel like most of this is mainly semantics. I too, don't believe that one can actually determine the consciousness of anyone but oneself. However, we do convince ourselves that other humans are in fact conscious, so it's still interesting to try to figure out what it would take for us to do that, and then apply the same standards to machines.
Personhood rights are ethical and granted by society as a whole. A well administered Turing Test gives the computer intelligence a chance to argue for treatment as a full person, without the prejudices that would normally be attendant on the fact that it's instantiated in electronics.
Even if you started completely against machines having rights, once you've allowed one to convince you that it's human-equivalent, you've much less room to rationalize away poor treatment of it.
https://news.ycombinator.com/item?id=9085587
It doesn't give a computer any chance to argue for treatment, it's about whether or not an average interrogator will be fooled by it.
Sure machine rights are an interesting thought - but the Turing Test as outlined in the "Computing Machinery and Intelligence" paper isn't related to machine rights. Anytime a computer says "no" without good reason to "sudo shutdown now", the plug will be pulled on it.
I agree with you that it isn't clear that the Turing Test has any meaning whatsoever. I'm not sure there ARE great implications from machines eventually being able to fool humans. Plenty of things can fool humans, paintings, mirrors, animals. And not all humans are equal in determining whether they are interacting with another human or not.
And unlike 'intelligence', there are concrete steps to be taken given some answers to the question as to what rights a particular program should have.
"perceiving something as something" and "treating something as something", different verbs. Turing tests aren't about treatment, they are about perception.
The field is wide open for thought tests or games on how to "treat" computers so I would say knock yourself out and make a better one.
The Turing test doesn't do a whole lot of good at answering those questions or for that matter any questions. Other than "will a computer be able to fool a human eventually".
The notion of machine rights existing at all is evidence that Turing's prediction that the way intelligence is thought of would change, but still most people would challenge the idea that computers think.
Maybe in another 65 years that will change further - and machine rights will no longer be a fringe idea.
Well yes, but if you perceive something as something that you will treat it as if it were that thing. And that is exactly what would happen during a successful Turing test - the computer would be treated as if it were a human by the judge.
It's not so much the difference between polynomial and exponential complexity. It's the difference between a coinductive (ie: infinitely looping, like a game loop) computation that adds new information to its "memory" in each iteration (which sounds monadic), compressing its inputs to accommodate that information, and then generating an output... vs a super-exponential lookup table that performs the input-output mappings without actually engaging in any side-effecting or monadic computation at all.
So yes, when you think of these two vastly different ways of writing down a "cognitive-style" computation, it becomes clear that one of them maps relatively well to what we think of when we say "mind", while the other doesn't, and Searle is conveniently eliding the distinction.
I would argue that Searle's Chinese Room is posed in a way that presupposes what forms computation might take or the limits it has. Since this was posed decades ago (1980), it makes sense in his time. If Searle had begun teaching today (instead of 1969) and seen convolutional neural networks and recurrent neural networks deployed in massive data centers, Watson win Jeopardy!, neural Turing machines, or a worm's brain emulated and embodied in a small robot, I doubt he would have posed his Chinese Room the same way. Maybe he would not have posed it at all, being convinced in light of those achievements.
Pierre Duhem was a physicist. But he also contributed to philsophy of scineces: check the famoous Duhem-Quine thesis. Same thing happend to "philosophy of biology". If you look at the journal "philosphy of biology", most of the early contributors came from biology.
I think, Scott Aaronson is aware of the controversy between Oded Goldreich and Neal Koblitz. This controversy could have been formulated better if people in theoretical computer scince had mastered "philosophy of sciences.
People in Artificial Intelligence (esp of CS variety) make wild claims like AI annihilating the human race. And these voices have strong backing from people like Bill Gates. However, these voices don't look at what philosophers have writtern about AI, and the kind of nonsense AI folks spew. Yes, AI can solve particular tasks (a chess game, autonomous driving): but can it replace humans? Yes, say the AI group; No, say the philosopher Hubert Dreyfus. Check the latter's book: What computers can't do, a critique of artificial reason.
Take for instance Rich Hickey's Are We There Yet?[1], the Alfred North Whitehead quotes are great food for thought: you dive into it just as much you are able to swim.
[1]: http://www.infoq.com/presentations/Are-We-There-Yet-Rich-Hic...
- http://www.protevi.com/john/DG/PDF/Remarks_on_Complexity_The...
- http://www.euppublishing.com/doi/abs/10.3366/jipt.2010.0004
It could perhaps be useful to do similar work with computational complexity theory.
It is completely different to computational complexity[1] which is what is being discussed in Aaronson's article.
[0] http://en.wikipedia.org/wiki/Complex_system
[1] http://en.wikipedia.org/wiki/Computational_complexity_theory
There is a place for philosophy, but it shrinks every time a new discovery is made. It hasn't been quite as roughly treated as theology has, but the lag is only about a century. The vast majority of what philosophers once spent their time talking about is now part of some empirically grounded special science, where making stuff up and asserting broad foundational claims without proof or testing is no longer viable.
Linguistics, in particular, has carved off large chunks. Physics likewise: no fun speculating on the nature of space and time when we can pin them down pretty precisely. Various combinations of sociology, psychology, economics and political science have heavily encroached on moral philosophy.
This is not to say that the world can't always use a few philosophers to poke around in the darkness and perhaps cross the boundaries of other fields to good effect, but the majority of the work that was once their exclusive preserve is now almost always better handled by people with actual, detailed, empirical knowledge of the subject.
There is no value in privileging the human scale or human imagination, and fortunately no need to live within the artificial restrictions that philosophers place on themselves, particularly their refusal to do experiments or even make systematic observations to test their ideas, preferring instead to rely on "what just makes sense", because that has always worked so well in the past. Insisting on empirical testing is in no way limiting because ideas that cannot be tested are irrelevant, as they must not have any impact whatsoever on any aspect of existence (if they did, observing and/or manipulating that aspect of existence would allow the ideas related to them to be tested.)
Philosophy does teach a certain kind of rigour in thinking, because without either empiricism or mathematical deduction to fall back on philosophers have to be extremely meticulous to avoid falling off the real axis entirely, but that starts to look more like a skill that ought to be taught to everyone rather than the basis for an entire academic discipline.
How do you define philosophy? can't we see science as as subset of philosophy?
I find it unfortunate when people like the commentator above make the claims that they do because it detracts from the respect that the field of philosophy deserves. It may not have the most obvious practical applications but it deserves attention and some of it is really mind-bending stuff. When people like Steven Hawking shit on the subject it's really a shame. They are causing themselves to miss out and they are driving potential great thinkers away.
Philosophy does not deal with testing empirical theories. No there has not been encroachment by linguistics, sociology, psychology, or economics.
I'm sorry you wasted so much time typing this comment, I've read it three times now and I cant find a single thing you got right.
"But as noticed by Goldreich, Micali, and Wigderson [65], there is something Merlin can do instead: he can let Arthur challenge him. Merlin can say:"
Arthur, send me a new graph K, which you obtained either by randomly permuting the vertices of G, or randomly permuting the vertices of H. Then I guarantee that I will tell you, without fail, whether K ∼= G or K ∼= H.
I don't understand what it means for a graph to be isomorphic to another graph, and I do not understand how Merlin's challenging answer can provide a (zero-knowledge) proof to the given problem either.
What Merlin is saying to Arthur is that no matter what kind of graph K Arthur comes up with, it cannot be congruent to both G and H, or, that there exists no method to create a graph that is both G and H.
Lastly it is zero-knowledge because Arthur learns nothing of the relation between G and H, only that the K he computes is not both of them simultaneously.
Not sure if that helps.
I had a hard time understanding what it means when I was in undergraduate school, because it seems so trivial -- why can't I say that the graphs are simply the same, like 1 = 1 or {0} = {0}? You could say they are the same, but the interesting point is that it is not easy to decide whether they are the same if the two graphs G and H are big. Therefore, "graph isomorphism" is a computational problem in its own right.
[1]: https://news.ycombinator.com/item?id=9074033 [2]: http://nautil.us/issue/21/information/ingenious-scott-aarons...
O.M.G. Yet another natural scientist trying to coerce philosophy to his 'way of thinking'.
In all seriousness, I think philosophy has problem, it's disconnect with reality is widening.
> To forestall misunderstandings, let me add a note of humility before going further. This essay will touch on many problems that philosophers have debated for generations, such as strong AI, the problem of induction, the relation between syntax and semantics, and the interpretation of quantum mechanics. In none of these cases will I claim that computational complexity theory “dissolves” the philosophical problem—only that it contributes useful perspectives and insights. I’ll often explicitly mention philosophical puzzles that I think a complexity analysis either leaves untouched or else introduces itself. But even where I don’t do so, one shouldn’t presume that I think there are no such puzzles! Indeed, one of my hopes for this essay is that computer scientists, mathematicians, and other technical people who read it will come away with a better appreciation for the subtlety of some of the problems considered in modern analytic philosophy.
(From some (very) cursory reading, this does seem to hold true.)