Hacker News Reader

Comment by jerf | Hacker News Reader

Amusing, but as a serious argument it critically fails in that "Strong AF" is measured by whether it fools a real bird into thinking it's a bird, and once you are "birdy" you can't be "more real than real", but in the real world everybody understands that passing a Turing Test is the beginning of AI, not the end. The Turing Test is a relatively unambiguous measurement of how successful the AI is, and it will progress from there.

Yes, it's true that the Turing Test does force a bit of possibly-difficult humanity onto the resulting AI, but all that probably boils down to is some constant factor improvement on human intelligence (on some hypothetical measure where we can quantify intelligence along a single dimension), with the resulting slightly-superhuman-AI just "faking" being human. That would still win, and I have a hard time imagining some sort of weird result where we have radically superhuman AIs that just can't seem to convincingly fake being human.

View on HN

I have a hard time imagining some sort of weird result where we have radically superhuman AIs that just can't seem to convincingly fake being human.

I have a hard time imagining it too, but I see no good reason why that couldn't happen (which was the whole point of the article I think - our current technology can easily make many different things that can fly, but we are still incapable of making something that really flies like a bird).

Maybe it will be easier to build extremely intelligent, consistent, rational AI than to get the nuanced emotion and irrationality that regular people have.

Thing is, nuanced emotion and irrationality is often actually rational - not in the sense that the human brain thought logically through the situation and decided rationally to act irrationally, but in the game theoretic sense of producing a superior outcome.

For example, if you get angry you can lose control and do things you'd regret later (seemingly irrational). That forms a powerful argument to avoid angering someone, because you know that they might react badly even if it's not in their best interests. Thus, evolution can engineer in a defense against e.g. cheaters of various kinds, even when punishing any one cheater would be counter-productive for the one doing the punishing.

(It strikes me now that there's an argument in here about banking, bailouts, the angry public and avoiding collapse of the economy, but I'm not going to pursue it.)

which was the whole point of the article I think - our current technology can easily make many different things that can fly, but we are still incapable of making something that really flies like a bird

I read it differently. The aim is to make something that can fly. But we, from our position as birds, don't accept flying as anything other than bird-like flying - with wings and feathers.

Leaving the analogy, I think the article is saying that humans generally don't accept anything as artificial intelligence unless it exactly mimics a human, or is indistinguishable from a human. Thus the attempts to work out exactly how the brain (or wing) works and to replicate it.

In reality, AI research has progressed considerably - visual recognition, autonomous navigation, question answering etc. But these are only 'gliding', and are only leading towards 'other flight-like phenomena' such as bat and insect flight. We don't accept them as parts of an artificial intelligence because we think (rightly or wrongly) that there is something more to intelligence than just a set of special algorithms and feed back loops.

I thought it was a critique of people focusing on trying to create AI by modeling human intelligence. But I think it's a dumb article because I am reasonably sure that Avian aerodynamics undoubtedly inspired early aeronautics.

If his criticism is just of the Turing Test instead, then I think he's a simpering ninny.

I've long thought of the Turing Test not as an argument against the impossibility of AI, but rather an argument against the impossibility of a machine that thinks. Certainly, as an actual practical test of AI, it seems pretty poor to me - people doing real-world Turing Tests seem to me to be missing the point.

There are a lot of people, especially religious people, who try to argue against thinking machines by saying that the ghost, the soul, the existentialist bit, is missing. The Turing Test is an argument against that; that if you can't tell the difference between the machine and a human when the "interface" is something as simple as a terminal rather than the obvious differences between machine and flesh, then you don't have good grounds to still say that machines cannot think - because it exposes you to the accusation that you're just pre-judging based on appearances.

In fact, I think the problem is in considering the turing test to be a definitive test for AI. Its certainly a definitive test of some subset of AI (which I am unsure how to define exactly), but to really consider it a definitive test of all AI is silly. I think this is the main point of the article. Sure, if a machine that acted exactly like a bird was invented and flew, that would be a flying machine, but thats not the only way to build a flying machine. Thats "bird like flying" in the same way that the turing test is a test for "human like intelligence."

I prefer EWD's take on the question of AI:

>The question of whether computers can think is like the question of whether submarines can swim.

Reply on news.ycombinator.com