Is AlphaGo really a huge breakthrough in artificial intelligence, or does it really just show how AI is impossible?
No.
>>879366
To which question?
>>879313
A computer defeating a 9p (9-dan) is a really big deal.
A 9p defeating AlphaGo is a big deal for the 9p, but us mortals can't really understand how big a deal it is.
None of this has anything to do with whether or not AI is impossible. I'm an intelligence, but I'm only 5k.
>>879416
If that's the case then why is this impressive? Is it just bruteforcing like they do with chess?
>>879395
No.
>>879430
So AI isn't possible?
>>879313
It's good AI, but people generally think "general AI" when people say "AI".
It's good at the task it was designed for. Go is a pretty complex task, by the way.
>>879313
I can't even beat 9x9 board on easy difficulty. Does that mean I'm less than human?
>>879435
I just wonder what the point of it is other than bragging rights though. Can you apply the things learned from this to other forms of intelligence?
>>879313
Most people thought making an AI for Go was utterly impossible until now.
So many combinations that it is next to impossible to have the same game ever.
Anyways, the human player beat the AI once (1 : 4). This is more than a victory for humans- for now.
>>879423
No, it's not bruteforcing. The game is much too large to do that due to the possible number of moves being much greater (like, by a mind-bogglingly huge amount) than that of chess at most points during the game.
It's impressive because defeating a professional at go was considered impossible even 2 years ago. If I say more I'll start talking out of my ass.
>>879479
>It's impressive because defeating a professional at go was considered impossible even 2 years ago.
That sounds like bullshit but whatever
>>879313
>Is AlphaGo really a huge breakthrough in artificial intelligence, or does it really just show how AI is impossible?
tacogirl.jpeg
>>879488
Not that anon
https://theconversation.com/googles-go-victory-shows-ai-thinking-can-be-unpredictable-and-thats-a-concern-56209
>There is a real sense that this month’s human vs AI Go match marks a turning point. Go has long been held up as requiring levels of human intuition and pattern recognition that should be beyond the powers of number-crunching computers.
>>879416
>>879313
>>879423
>>879459
AlphaGo is a milestone, but not really an innovation.
Computer chess was saying we have computers that are more capable at computing than the human mind.
AlphaGo was saying, we have an algorithm that is good at pattern recognition and can mimic intuition and can learn and beat a human when hooked up to a chess super computer.
It has been known forever, go requires intuition and pattern recognition in addition to calculation, and simply can't be brute forced like chess. The first go program written in 1968 had pattern recognition. Later developments, with the internet, included machine learning, because it's simply not feasible to enter every likely pattern.
So none of this is new, except hooking it up to a super computer and having it learn against itself, since every prior iteration of go has been significantly worse at go than a human it makes more sense for a computer to learn from a human. Obviously when you get to a point where a computer rivals a good human, it;s more efficient for the machine to play against itself 24/7.
The only important thing here is AlphaGo proved that you can, a super computer can do more than brute force calculation, and can mimic things only humans used to be able to do, like intuition, on a level that rivals a real person. It's a milestone because of this, and a shift in focus to other things important for AI besides brute force calculating power.
>>879313
The essential problem with machine learning attempts to solving such problems is that we have no way to analyse and understand the "reasoning" of the machine yet.
All we can do is train the machine with a certain set of input data from which, based on certain rules, which are essentially educated guesses rather than analytically chosen, and hope that what comes out of it is capable of doing what we want it to do.
If it works, we have a machine that can do it, but we have no way to find out how it does it. If it does not work, then we have to start anew with different guidelines.
This is also why many mathematicians have refused to work with such methods when it comes to proving theorems, since the machine constructs proofs in a fashion that are occasionally unreasonably structured and completely unintuitive to human readers to the point where we can "verify" whether a proof is true or not (by using a machine to do so), but cannot understand how it was constructed.
>>879510
That's quite interesting. Especially the idea that a computer can find ways to do things that have never been considered by a human being thanks to their learning in a different way. I wonder what the implications of that may be.
>>879479
>It's impressive because defeating a professional at go was considered impossible even 2 years ago. If I say more I'll start talking out of my ass.
No one else had a dedicated super computer. Crazy Stone was considered the strongest AI, and it can run on your PC and only learns in development. AlphaGo with DeepMind basically brute forced the learning process.
>>879531
dis
A computer can play a board game. Wow.
