AlphaGo Defeats World's Best Human Player

AI Go Program Performs Better Than Expected

Anything that doesn't fit anywhere else below.

Moderators: Calilasseia, ADParker

Re: AlphaGo Defeats World's Best Human Player

#21  Postby tuco » Mar 24, 2016 1:42 am

So far those are sci-fi scenarios. Apocalypse seems to be popular theme. However, its hard for me to imagine that by the time AI will be so strong .. to surpass humans in every respect, there will not be kind of safeguards. What to say? Read Asimov or something. But AI will disable even this! Ok.
tuco
 
Posts: 13755

Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#22  Postby Hardcoreathiest » Mar 24, 2016 11:55 pm

tuco wrote:So far those are sci-fi scenarios. Apocalypse seems to be popular theme. However, its hard for me to imagine that by the time AI will be so strong .. to surpass humans in every respect, there will not be kind of safeguards. What to say? Read Asimov or something. But AI will disable even this! Ok.



No, no please I must ask you not to assert this without good cause. Ai will be benevolent and one of the best things for happening to us !

Please I ask you to illustrate this by finding many high qualified people who are proving the doom-mongers wrong.

Think of the doom-mongers as having terrible bias, like the military advisors who spread scare stories about russia, yet do no pracital work ! While the people who build the Ai are the actual engineers doing real logic and high testing all the way.
Hardcoreathiest
 
Name: Joe Alan
Posts: 60

Country: ireland
Ireland (ie)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#23  Postby tuco » Mar 03, 2017 4:32 pm

AI shoves all in: DeepStack, Libratus poker bots battle Texas Hold 'em pros heads up

AI poker intuition
A game between two players betting any number of chips produces 10160 possible situations – a number too large for a computer to handle. To skirt around the problem, DeepStack “squeezes” it down to 1014 abstract situations that are learned by playing against itself.

Like DeepMind’s AlphaGo, DeepStack picks the best move to take by drawing on a bank of possible moves by calculating what types of scenarios are more likely, something the researchers compare to intuition: “A gut feeling of the value of holding any possible private cards in any possible poker situation.”

The programme’s “intuition” has to be trained using two neural networks. One learns to estimate the counterfactual – or “what-if” values after the first three public cards are dealt, and the other neural network recalculates the values after the fourth public card is dealt.

Simplifying the number of situations means the decision tree computed by DeepStack is effectively pruned, and it’s easier to approximate the Nash equilibrium – a solution in game theory which states that no player has an incentive to change his or her strategy – continuously, after each round.

Since it doesn’t have an overarching strategy decided before the game, it doesn’t need to keep tabs on all 1014 abstract situations – it can solve the decision tree in under five seconds.

“The DeepStack algorithm is composed of three ingredients: a sound local strategy computation for the current public state, depth-limited lookahead using a learned value function over arbitrary poker situations, and a restricted set of lookahead actions,” the paper said.


https://www.theregister.co.uk/2017/01/14/deepstack/

DeepStack: Expert-Level Artificial Intelligence in No-Limit Poker - https://arxiv.org/abs/1701.01724
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#24  Postby VazScep » Mar 07, 2017 11:42 am

I did my Masters back in 2007, and the top student (out of 200) in the year did his thesis on Go using genetic algorithms. He was an avid Go-player (not sure which Dan), and lived for three months in China to play. By the end of his thesis, he thought Go was basically hopeless, and changed to a different area of AI for his PhD.

So this stuff surprised me. Deep-learning people seem to be everywhere now. The new guy sat opposite me is one of them, and his old supervisor was a big deepmind guy at Oxford who has now "sold out" to work for Google. Our first chat in the pub was around the future of AI, and he ranted pretty extensively over scare-stories from people like Stephen Hawking and Sam Harris, insisting that these guys don't know what the fuck they're on about and that all this deep learning stuff is still basically just numerical optimisation (though I can't say I understand the maths myself).

On the other hand, another of my mates, who did his PhD at Imperial, on a topic which he believed was in the direction of General AI, submitted the first version of his thesis with a dismal conclusion that all such research should be abandoned lest we unleash hell. We have pretty much the same argument every time we meet up, with me saying that I don't buy the pessimism for a second. My concerns over AI are more pedestrian: I worry that people are going to start relying on recommender systems and other automated systems as authorities and forget to do sanity checks, ignoring that they'll often just be finding really shitty optima, over-fitting and otherwise solving for the completely wrong problem.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4505

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#25  Postby tuco » Mar 07, 2017 6:55 pm

Cool story bro ;)

Its inevitable. Imagine people not relying on calculators, doing own checks. Perhaps competing AI will be able to their own checks? I'd be interested in scenarios you and mate were debating. Scenarios where no sanity check was done and large number (?) of people followed shitty optima recommended by AI.
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#26  Postby VazScep » Mar 07, 2017 9:54 pm

Our go-to scenario is --- perhaps naive in itself --- throwing AI at tax-code systems.

My pessimistic mate paints a much bleaker scenario. He thinks that a general AI will always be some sort of reinforcement learner, and that as soon as it gets smart enough, it's first job will be to take control of its reward mechanism, which he thinks inevitably means deploying Machiavellian plots against its human controllers and ultimately getting us out of the picture. I think that idea is hopelessly naive.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4505

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#27  Postby tuco » Mar 07, 2017 10:13 pm

Well, I do not think? there is such a thing as optima for tax-code system as such system is part of/influences other complex systems. IMO its purely political matter, hence my belief that for all practical purposes economy is in fact political economy (tax haven vs countries with high taxes .. both work somehow). Not even sure how to approach such scenario tbh, where to start, what to define first, etc. Naive is trying to predict stuff decades from now tho ;)

I drive using Google maps. I input destination and let Google plan my route. Sometimes I do (sanity/rational) check but sometimes I just drive because .. it does not matter all that much (from my experience) and I dunno what kind of information Google works with, namely traffic. Am I, in this scenario, one of those you had in mind? At the same time, am I naive not to realize potential dangers? I judge by my own standards and by those standards, scept chick lol, I am not worried. Could be ignorance but nobody explained to me why should I feel otherwise.
tuco
 
Posts: 13755

Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#28  Postby VazScep » Mar 07, 2017 10:50 pm

tuco wrote:Well, I do not think? there is such a thing as optima for tax-code system as such system is part of/influences other complex systems. IMO its purely political matter, hence my belief that for all practical purposes economy is in fact political economy (tax haven vs countries with high taxes .. both work somehow). Not even sure how to approach such scenario tbh, where to start, what to define first, etc. Naive is trying to predict stuff decades from now tho ;)
The tax code thing might not be naive. I only say that because I don't really know much about how tax returns work. The idea is this: AIs, right now, are really impressive in largely artificial environments. Everything about the universe of Go is fully formalisable. It's a self-contained artificial universe. I guess that could be the case with tax laws. The stockmarket: no. That's not self-contained.

The stuff about an AI trying to take control of its reward function, on the other hand, is based on a whole truck load of presumptions.

I drive using Google maps. I input destination and let Google plan my route. Sometimes I do (sanity/rational) check but sometimes I just drive because .. it does not matter all that much (from my experience) and I dunno what kind of information Google works with, namely traffic. Am I, in this scenario, one of those you had in mind? At the same time, am I naive not to realize potential dangers? I judge by my own standards and by those standards, scept chick lol, I am not worried. Could be ignorance but nobody explained to me why should I feel otherwise.
So long as you're not doing this, you're good:

Here we go again. First, we discover recursion.
VazScep
 
Posts: 4505

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#29  Postby tuco » Mar 07, 2017 10:59 pm

Not like that but close ;) Its off topic but pre-Google maps time I sometimes (when I was lost) drove, especially in/close to cities, where others went. Reasoning .. they all cant go nowhere lol that's fucked up.
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#30  Postby VazScep » Mar 07, 2017 11:32 pm

Have you read Dirk Gently?
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4505

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#31  Postby VazScep » Mar 08, 2017 12:04 am

Perhaps it would save time if he went back to get his car, but then again it was only a short distance, and he had a tremendous propensity for getting lost when driving. This was largely because of his method of "Zen" navigation, which was simply to find any car that looked as if it knew where it was going and follow it. The results were more often surprising than successful, but he felt it was worth it for the sake of the few occasions when it was both.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4505

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#32  Postby tuco » Aug 27, 2017 8:04 pm

I will drop it here because its related and other reasons.

DeepMind and Blizzard open StarCraft II as an AI research environment

[snip]

StarCraft and StarCraft II are among the biggest and most successful games of all time, with players competing in tournaments for more than 20 years. The original game is also already used by AI and ML researchers, who compete annually in the AIIDE bot competition. Part of StarCraft’s longevity is down to the rich, multi-layered gameplay, which also makes it an ideal environment for AI research.

For example, while the objective of the game is to beat the opponent, the player must also carry out and balance a number of sub-goals, such as gathering resources or building structures. In addition, a game can take from a few minutes to one hour to complete, meaning actions taken early in the game may not pay-off for a long time. Finally, the map is only partially observed, meaning agents must use a combination of memory and planning to succeed.

The game also has other qualities that appeal to researchers, such as the large pool of avid players that compete online every day. This ensures that there is a large quantity of replay data to learn from - as well as a large quantity of extremely talented opponents for AI agents.

Even StarCraft’s action space presents a challenge with a choice of more than 300 basic actions that can be taken. Contrast this with Atari games, which only have about 10 (e.g. up, down, left, right etc). On top of this, actions in StarCraft are hierarchical, can be modified and augmented, with many of them requiring a point on the screen. Even assuming a small screen size of 84x84 there are roughly 100 million possible actions available.


[snip]

https://deepmind.com/blog/deepmind-and- ... vironment/

----

While being immensely ignorant of the field, let me dare to guess that taking the actions will not be an issue.
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#33  Postby GrahamH » Aug 27, 2017 9:28 pm

Is it wise to exercise AI on war-games? If something of AlphaGo is carried forward in to the next generation of Deep Mind could there be a deep-seated tendency to compete and defeat humans? Deep Mind object recognition already starts with a pre-trained dataset. The system is so hard to comprehend they developed Google Dreams as a way to get some insights into how it sees. If it gets to some sort of general intelligence that makes plans maybe something of StarCraft campaigns could be useful, but would GoogleDreams of strategy feature killing humans?

I look forward to hearing about more positive scenarios for developing DeepMind.
Why do you think that?
GrahamH
 
Posts: 17229

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#34  Postby tuco » Aug 27, 2017 10:33 pm

That seems like a question for philosophy. Unless convinced otherwise, I will go with lets say AI scientists who consider such environment suitable for stated reasons. Ultimately, the point of such exercise is imo for AI to reach set goal(s). If we take the war element out, we are left with, well, managing economy.
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#35  Postby GrahamH » Aug 27, 2017 10:56 pm

tuco wrote:That seems like a question for philosophy. Unless convinced otherwise, I will go with lets say AI scientists who consider such environment suitable for stated reasons. Ultimately, the point of such exercise is imo for AI to reach set goal(s). If we take the war element out, we are left with, well, managing economy.


I doubt the scientists give that much thought to such ethical or philosophical issues of unintended consequences

Maybe Elon Musk and Stephen hawking have a point

If AI is only good for war games and financial trading we would be better without it.

But relevant to the topic what constructive challenges can deepmind take on for the next headline event.
Why do you think that?
GrahamH
 
Posts: 17229

Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#36  Postby tuco » Aug 27, 2017 11:20 pm

That is why its question for philosophy.

Economy as in: managing resources, not necessarily and only predicting market dynamics. To reiterate, the way I understand is that AI is being used on checkers, chess, Go or SC2 because the games provide suitable environment for research and problem solving is base for further research and applications.

Constructive challenges .. dunno, human like behavior or is it too vague or even utopia and nonsense? Seems to me also like a philosophical question at this point.
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#37  Postby Matthew Shute » Sep 05, 2017 12:19 pm

tuco wrote:I will drop it here because its related and other reasons.

DeepMind and Blizzard open StarCraft II as an AI research environment


A bot has been performing rather well at Dota 2, also.



http://money.cnn.com/2017/08/12/technol ... index.html
https://www.theverge.com/2017/8/14/1614 ... -elon-musk

Their next goal for the Dota 2 bot is to have it perform against the best in a messier 5v5 match.
"Change will preserve us. It is the lifeblood of the Isles. It will move mountains! It will mount movements!" - Sheogorath
User avatar
Matthew Shute
 
Name: Matthew Shute
Posts: 3477
Age: 39

Antarctica (aq)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#38  Postby tuco » Sep 05, 2017 12:58 pm

(I am) trying to copy the bot .. lol? However, unlike the bot you make, inevitable, mistakes mate. gg
tuco
 
Posts: 13755

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#39  Postby VazScep » Sep 05, 2017 4:00 pm

GrahamH wrote:
tuco wrote:That seems like a question for philosophy. Unless convinced otherwise, I will go with lets say AI scientists who consider such environment suitable for stated reasons. Ultimately, the point of such exercise is imo for AI to reach set goal(s). If we take the war element out, we are left with, well, managing economy.


I doubt the scientists give that much thought to such ethical or philosophical issues of unintended consequences

Maybe Elon Musk and Stephen hawking have a point

If AI is only good for war games and financial trading we would be better without it.
This sort of AI is good for any sort of problem that amounts to numerical optimisation, such as the logistics of food distribution or town planning, standard problems for which we generally get a good bang for the buck with machine learning.

It happens that Chess and Go were classic AI problems, and so there was a lot of kudos in beating all humans at them. Computer games are a trivial source of AI problems, since solving games is always an AI problem. Forget the war aspect. A non-cooperative game (which is the vast majority of games) can always be described in terms of warfare. This is a case of having an all two powerful metaphor.

Another reason games are targeted is because they provide a closed environment for a learning machine: all the information is there, and the game is formally specified.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4505

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#40  Postby GrahamH » Sep 05, 2017 5:06 pm

VazScep wrote:
GrahamH wrote:
tuco wrote:That seems like a question for philosophy. Unless convinced otherwise, I will go with lets say AI scientists who consider such environment suitable for stated reasons. Ultimately, the point of such exercise is imo for AI to reach set goal(s). If we take the war element out, we are left with, well, managing economy.


I doubt the scientists give that much thought to such ethical or philosophical issues of unintended consequences

Maybe Elon Musk and Stephen hawking have a point

If AI is only good for war games and financial trading we would be better without it.
This sort of AI is good for any sort of problem that amounts to numerical optimisation, such as the logistics of food distribution or town planning, standard problems for which we generally get a good bang for the buck with machine learning.

It happens that Chess and Go were classic AI problems, and so there was a lot of kudos in beating all humans at them. Computer games are a trivial source of AI problems, since solving games is always an AI problem. Forget the war aspect. A non-cooperative game (which is the vast majority of games) can always be described in terms of warfare. This is a case of having an all two powerful metaphor.

Another reason games are targeted is because they provide a closed environment for a learning machine: all the information is there, and the game is formally specified.


I guess it would be more comforting for the AI to be trained on something other than zero-sum games because if we find ourselves in a zero-sum game with advanced AI we could well be screwed.
I'm not concerned with the war aspect as such, more that future AI may carry forward aspect of playing to win into new interactions.

Someone once speculated that setting a goal for AI of wining as many games of chess as possible could have dire unforeseen consequences if the AI took over as much computing resource as possible in order to meet that goal. Don't set a goal of eliminating human suffering in case an AI finds and implements the obvious (final) solution!
Why do you think that?
GrahamH
 
Posts: 17229

Print view this post

PreviousNext

Return to General Science & Technology

Who is online

Users viewing this topic: No registered users and 1 guest