Comment by electrograv | Hacker News Reader

Comment by electrograv | Hacker News Reader

Parent Full thread

Without even getting into misconceptions and inaccuracies here, the following quotes illustrate "the moving goalposts of AI": The almost-metaphysical "mystique" that many attribute (perhaps subconsciously) to human intelligence vs machine intelligence:

  > AI techniques, which are good at learning what is "in the data" 
  but have not yet been shown to be capable of thinking outside the box

  > AIs would be highly vulnerable

  > exploitable quirks in the AI

  > established lack of robustness of neural nets

  > clever human players will be quick to exploit them

Of course, it is important to identify pitfalls, so as to address them. But it is sad that no matter how good AI gets, there will always be pessimistic reasoning downplaying its success. With every step forward, there are the predictions of impassible future failures -- "just around the corner"!

Due to this mystical "cleverness" attributed exclusively to humans, I'm not convinced a majority of humans will ever allow machines to be considered 'intelligent', no matter how far the technology advances. I'd love to be proven wrong, though.

Gaming skill is often divided into mechanical skill and decision making. 1v1 like this is dominated heavily by mechanical skill (frame-perfect last hits/denies, the absurd creep block, etc). Generally, it's the sort of thing you can "download skill" in (cheat at) because computers are very good at it.

Decision making is how quake pros can beat aimbotters, and starts to become very important in 5v5 DotA, particularly at a higher level. You'd see it in terms of drafting, ganks, map vision/control, map positioning, sneaking rosh, and other unexpected meta strats. Considering the real-time computation requirements, this is almost certainly a much harder problem.

So while it's entirely possible bot-team would win just through insane mechanics snowballing the game (ganking 0ms reaction times would be fun), or through good-enough decision making + insane mechanics, they could also be outplayed.

"ganking 0ms reaction times would be fun"

wonder, against this type of AI, if it's even possible to land something like meat hook or sacred arrow?

on the note of good decision making, maybe even a failed hook could be enough to exploit the bot into death. for instance, hero X chases bot back toward tower. hook comes in predicting the bots path. bot would have to either walk into the hook or walk back into the hero chasing it.

I will consider an A.I. truly intelligent when it passes the Turing test with flying colors. And as a native English speaker, too (not just barely passing by posing as a 13 year old Ukrainian boy with a poor grasp of English, which is what happened last time).

As far I know, no A.I. developed has ever met this bar. It seems that the nuances, inconsistencies, and total freedom of language make it a totally different kind of problem to solve than a simple game-playing algorithm.

Or maybe I'm wrong and getting an A.I. to pass the Turing test is just another kind of game-playing algorithm with different parameters. At any rate, we're quite a way off.

Again, you're moving the goalposts forward whenever AI surpasses them. This time you move the goalposts forward by requiring a proof of native English proficiency. Next time it will be something else; that goalpost will be achieved, and when it is, you will move the goalpost forward yet again.

Perhaps you'll require the Turing+++ test be conducted with audio rather than text. Then it will be video. Then a physical android. Then it will need to prove itself better than the human in an IQ test. Then it will need to pass various real world success tests, because IQ tests are too synthetic. But those won't be good enough either, so it will have to live 80+ years as a human in the real world without being discovered. But that's not enough, because maybe its life was too easy -- the real human experience is characterized by pain and suffering as well as joy, so we'll also need it to somehow prove its experience is more real. And on and on and on. It's always something else -- with every step forward, the success will be dismissed and the goalpost will move forward.

I should say this isn't intrinsically a bad thing, since it pushes us to advance the field of AI more and more. But IMO this is too often promoted as a pessimistic criticism of the AI field, rather than an encouraging push of optimism.

Why must this notion of binary "true intelligence" exist, anyway? Can't we see that intelligence is a spectrum (most likely with multiple dimensions of proficiency), and AI is continuously advancing upwards?

The Turing test is hardly a new measure - it's been around for ~70 years and hasn't been passed yet. But otherwise I agree with what you said - this is just my personal measure of when we'll be in society-shaking territory.

It has been passed, but the definition of passed changes to make the test tougher each time. I think that is what the other poster is talking about.

Has it been passed? Turings original paper assumes adults conversing intelligently. That is the spirit of the test. Not a clever hack of the rules, like posing as a child which doesn't speak english.

This is the $20000 long bet from 2002 between Kapor and Kurzweil:

http://longbets.org/1/#terms

> Ray Kurzweil maintains that a computer (i.e., a machine intelligence) will pass the Turing test by 2029. Mitchell Kapor believes this will not happen.

Level 4 autonomous vehicles with a limited operational domain will already be in society-shaking territory without showing general intelligence, they only have to fail safely on outside domain problems.

Most of those are quite logical though. If you want to say that a general AI has an intelligence equal to or exceeding human intelligence then it should perform reasonably similar to a human at all tasks and in all scenarios.

A human that can solve complex equations in a few seconds would probably be considered highly intelligent, but a machine can do the same in nanoseconds. Would you say a calculator is an intelligent AI?

Will you consider a set of AIs based on the same underlying architecture that can perform several different online jobs as well as their human counterparts truly intelligent, without any reservations regardless of the approach used [1]?

This would include the ability to communicate with their human hirers and collaborators in natural language.

[1] We can also think of the human brain as a massive structure of relatively simple neurons with methods of communications among them that can be translated to mathematical parameters.

Reply on news.ycombinator.com