Google's Deepmind AlphaZero chess AI is apparently learning itself human intuition and creativity

[Master Disaster]

For those that don't know Deepmind is a UK based AI company that we're purchased by Alphabet a few years ago. They've been running a neural network computer system called AlphaZero which they started as a blank slate system and allowed it to teach itself chess.

 

Famously AlphaZero beat the world's best chess computer 100 to zero back in 2014 after teaching itself the game in only 4 hours and today Deepmind have said the system is showing "a subtle sense of intuition similar to that of a human player" and the AI has developed its own unique playstyle suggesting that it's now learning, improving and developing its game in a similar way to a human.

 

In recent games AlphaZero went up against Stockfish 1000 times, it lost only 6 of these matches, 155 were won and the rest were drawn.

Quote

DeepMind’s artificial intelligence programme AlphaZero is now showing signs of human-like intuition and creativity, in what developers have hailed as ‘turning point’ in history.

 

The computer system amazed the world last year when it mastered the game of chess from scratch within just four hours, despite not being programmed how to win.

 

But now, after a year of testing and analysis by chess grandmasters, the machine has developed a new style of play unlike anything ever seen before, suggesting the programme is now improvising like a human.

 

Unlike the world’s best chess machine - Stockfish - which calculates millions of possible outcomes as it plays, AlphaZero learns from its past successes and failures, making its moves based on, a ‘nebulous sense that it is all going to work out in the long run,’ according to experts at DeepMind.

 

When AlphaZero was pitted against Stockfish in 1,000 games, it lost just six, winning convincingly 155 times, and drawing the remaining bouts.

The AI has taught itself to sacrifice pieces in order to achieve a better tactical position over it's opponent, something its never been shown how to do. The scientists behind the AI are claiming its showing signs of intuition and creativity, traits only normally present in humans and even chess grandmasters are impressed by what they're seeing. It's being claimed as a possible turning point in the journey of AI development.

Quote

Yet it was the way that it played that has amazed developers. While chess computers predominately like to hold on to their pieces, AlphaZero readily sacrificed its soldiers for a better position in the skirmish.

 

Speaking to The Telegraph, Prof David Silver, who leads the reinforcement learning research group at DeepMind said: “It’s got a very subtle sense of intuition which helps it balance out all the different factors.

 

“It’s got a neural network with millions of different tunable parameters, each learning its own rules of what is good in chess, and when you put them all together you have something that expresses, in quite a brain-like way, our human ability to glance at a position and say ‘ah ha this is the right thing to do'.

 

“My personal belief is that we’ve seen something of turning point where we’re starting to understand that many abilities, like intuition and creativity, that we previously thought were in the domain only of the human mind are actually accessible to machine intelligence as well. And I think that’s a really exciting moment in history.”

 

AlphaZero started as a ‘tabula rasa’ or blank slate system, programmed with only the basic rules of chess and learned to win by playing millions of games against itself in a process of trial and error known as reinforcement learning.

 

It is the same way the human brain learns, adjusting tactics based on a previous win or loss, which allows it to searching just 60 thousand positions per second, compared to the roughly 60 million of Stockfish.

 

Within just a few hours the programme had independently discovered and played common human openings and strategies before moving on to develop its own ideas, such as quickly swarming around the opponent’s king and placing far less value on individual pieces.

 

The new style of play has been analysed Chess Grandmaster Matthew Sadler and Women’s International Master Natasha Regan, who say it unlike any traditional chess engine.

 

”It’s like discovering the secret notebooks of some great player from the past,” said Sadler.

 

Regan added: “It was fascinating to see how AlphaZero's analysis differed from that of top chess engines and even top Grandmaster play. AlphaZero could be a powerful teaching tool for the whole community.

https://www.telegraph.co.uk/science/2018/12/06/deepminds-alphazero-now-showing-human-like-intuition-creativity/amp/

 

I've long said Google will be involved in the creation of the AI that kills us all. I don't find this to be a good thing. What happens when it starts to develop anger and looks to seek revenge?

 

I hope they know how to handle these machines they're creating.

[corrado33]

I for one welcome our robot overlords.

 

First chess, next... THE WORLD. 

[Bananaman1]

11 minutes ago, Master Disaster said:

What happens when it starts to develop anger and looks to seek revenge?

Lol thats not how that works, this isn't a sci-fi movie.  Even if it did work like that we're still a long ways from any sort of AGI

[Trik'Stari]

Has any of these people thought that maybe we should slow down a bit?

 

Automation stands very likely to put a third of the workforce of the developed world, out of a job. Less than 1/5th-1/15 of those (in my mind) is likely to find a new job repairing the automation machines. And how long before those people are replaced by machines?

 

I wouldn't mind, if we were seeing huge breakthroughs in "putting my brain in a jar on top of a robot so I can live forever".

[laminutederire]

Google being optimistic as usual

[DocSwag]

1 hour ago, Trik'Stari said:

Automation stands very likely to put a third of the workforce of the developed world, out of a job. Less than 1/5th-1/15 of those (in my mind) is likely to find a new job repairing the automation machines. And how long before those people are replaced by machines?

 

I think the trick would be to compensate for the lost jobs with something else (maybe a UBI? Not sure if it's the right solution though). Because if automation is saving the company money and increasing production, there should be excess capital being generated which could be redistributed to the people who's jobs it replaced.

 

That being said, this would require extreme changes to how the government would regulate companies and I'm sure many people would be heavily against such a thing. As you mentioned, slowing down could result in a more gradual change making it so more people would likely be on board with the idea.

[williamcll]

1 million dollar chess machine?

[Master Disaster]

7 hours ago, Bananaman1 said:

Lol thats not how that works, this isn't a sci-fi movie.  Even if it did work like that we're still a long ways from any sort of AGI

I'm sure only a few months ago they would have said the same thing about intuition and creativity. The problem with breaking new ground is that it's new ground and by definition, unknown territory.

[Master Disaster]

7 hours ago, laminutederire said:

Google being optimistic as usual

Actually the biggest praise is coming from the chess community.

[Guest]

Want to see the openAI vs Google's AI SC2 

[ScratchCat]

[RejZoR]

Chess is literally just a mathematical thing. If you can predict all movements through calculations in a fraction of a second on every turn, you don't need intuition. The thing why none of this worked before is because computers were just reacting to what human was doing. They lacked the initial response how to lay down the strategy, ie how to be the leading one and not reacting one. They just used a predefined set of rules when expected to make the first move and from ther on it was reactive again.

[Master Disaster]

14 minutes ago, RejZoR said:

Chess is literally just a mathematical thing. If you can predict all movements through calculations in a fraction of a second on every turn, you don't need intuition. The thing why none of this worked before is because computers were just reacting to what human was doing. They lacked the initial response how to lay down the strategy, ie how to be the leading one and not reacting one. They just used a predefined set of rules when expected to make the first move and from ther on it was reactive again.

You didn't read the article did you? This AI taught itself the game from scratch and in 4 hours was good enough to beat the world's best "traditional" chess computer.

 

Now it's taught itself human characteristics such as sacrificing pieces to gain the advantage. This isn't really about the chess, it's more about using chess as a means to see how far an AI can go if allowed.

 

This thing has no predefined rules nor has it ever been given ANY information by anybody on the subject. Everything it knows it has learned itself by trial and error.

[laminutederire]

2 hours ago, Master Disaster said:

Actually the biggest praise is coming from the chess community.

It's not the chess community talking bout "intuition" etc.

[Master Disaster]

19 minutes ago, laminutederire said:

It's not the chess community talking bout "intuition" etc.

And how else would you describe its behaviour?

[laminutederire]

3 minutes ago, Master Disaster said:

And how else would you describe its behaviour?

Well not intuitive. Intuition is based on considering a few of really good outcomes from the get go which are determined by simple rules and limited experience. Here it only does that because it precomputed heavily everything first, so it just fakes intuition.

They have pretty good results but they are overhyping it too much

[Master Disaster]

3 minutes ago, laminutederire said:

Well not intuitive. Intuition is based on considering a few of really good outcomes from the get go which are determined by simple rules and limited experience. Here it only does that because it precomputed heavily everything first, so it just fakes intuition.

They have pretty good results but they are overhyping it too much

Nah, intuition means the ability to understand something without conscious reasoning...

 

Quote

the ability to understand something instinctively, without the need for conscious reasoning.

...so actually yes, this AI is showing intuition.

 

Considering the fact everything it knows is self taught there's also a strong argument to be made for its thinking process to be closer to a humans than a computer.

 

The human brain does exactly the same as a computer, it searches prior knowledge it has on a subject and decided what would be the best way to deal with it. The difference between a brain and a computer is a computer must have the scenarios programmed into it where as a brain learns by trying and failing.

 

This machine has had ZERO programming so all the situations its drawing on while playing are from previous experience only. It hasn't been shown the best possible outcomes of a game, it hasn't been shown the worst outcomes, heck it hasn't even been shown which pieces can move where.

 

What this machine is doing is genuinely a big deal, the fact it's learning itself to operate in a similar way to how a human would operate is staggering.

[RejZoR]

9 minutes ago, laminutederire said:

Well not intuitive. Intuition is based on considering a few of really good outcomes from the get go which are determined by simple rules and limited experience. Here it only does that because it precomputed heavily everything first, so it just fakes intuition.

They have pretty good results but they are overhyping it too much

This is why I always cringe hard when they smack "Ai" on literally anything that has 3 IF statements in the code... Just because something makes an outcome automatically, that doesn't make it an "Ai". It's just a set of algorithms. Even if we expand things to "self learning" it's basically just self learning an algorithm, but it still needs to understand the context of the operation which has to be predefined so it learns movement and win condition and not shades of the figures under different light depending on position instead... You could technically call it "Ai" but in the end it's still a rather brute force approact to finding a solution, because our brain doesn't work like that.

 

Btw, that zero programming is a bull, because it's impossible for software to just pop itself into existence and be self aware of the task at hand. Someone had to specifically code a program that it knows it'll be doing chess and then learn parameters within this set. Because it also has to give a correct output. It has to know it's chess, otherwise it would just start drawing dick pics or something...

[Master Disaster]

Just now, RejZoR said:

This is why I always cringe hard when they smack "Ai" on literally anything that has 3 IF statements in the code... Just because something makes an outcome automatically, that doesn't make it an "Ai". It's just a set of algorithms. Even if we expand things to "self learning" it's basically just self learning an algorithm, but it still needs to understand the context of the operation which has to be predefined so it learns movement and win condition and not shades of the figures under different light depending on position instead... You could technically call it "Ai" but in the end it's still a rather brute force approact to finding a solution, because our brain doesn't work like that.

 

Btw, that zero programming is a bull, because it's impossible for software to just pop itself into existence and be self aware of the task at hand. Someone had to specifically code a program that it knows it'll be doing chess and then learn parameters within this set. Because it also has to give a correct output. It has to know it's chess, otherwise it would just start drawing dick pics or something...

Intuition does not involve conscious reasoning.

 

Nobody said zero programming either, obviously the thing was given logic. It has no programming on the game of chess, the sole reason for its existence.

[laminutederire]

14 minutes ago, Master Disaster said:

Nah, intuition means the ability to understand something without conscious reasoning...

 

...so actually yes, this AI is showing intuition.

 

Considering the fact everything it knows is self taught there's also a strong argument to be made for its thinking process to be closer to a humans than a computer.

 

The human brain does exactly the same as a computer, it searches prior knowledge it has on a subject and decided what would be the best way to deal with it. The difference between a brain and a computer is a computer must have the scenarios programmed into it where as a brain learns by trying and failing.

 

This machine has had ZERO programming so all the situations its drawing on while playing are from previous experience only. It hasn't been shown the best possible outcomes of a game, it hasn't been shown the worst outcomes, heck it hasn't even been shown which pieces can move where.

 

What this machine is doing is genuinely a big deal, the fact it's learning itself to operate in a similar way to how a human would operate is staggering.

Well define me conscious reasoning then. Because that algorithm is kinda doing everything consciously on could argue.

 

I'm just criticizing the way those scientists always overhype what they are doing all the time. If you'd believe them we'd have self conscious Ai since years and so on...

[Master Disaster]

Conscious reasoning would be defined as being aware of implications of the choice you're making (at least thats how I see it anyway).

 

Reasoning means the ability to look at a situation and choose an appropriate response and consciousness means being aware of something.

 

The AI is certainly able to make a choice but it's certainly not aware of the implications of that choice.

 

Being intuitive means making the choice without using consciousness to think if it's the right choice or not or effectively knowing what to do without thinking about it.

[JediFragger]

KILL IT WITH FIIIIIIIIIIIIIIIIIIIRRRRRRREEEEEEEEEEEE!!!!!!!!!!!!!!!!!!!!!!

 

[LAwLz]

11 hours ago, Trik'Stari said:

Has any of these people thought that maybe we should slow down a bit?

 

Automation stands very likely to put a third of the workforce of the developed world, out of a job. Less than 1/5th-1/15 of those (in my mind) is likely to find a new job repairing the automation machines. And how long before those people are replaced by machines?

Spotted the Luddite.

 

People said the same thing when "spinning Jenny" was first developed.

"If we no longer need people spinning threads 10 hours a day those people will be out of jobs, therefore this machine is bad!".

 

Machines taking over jobs has historically meant that people can spend more time on creative or more skill intense jobs. I think the problem is that people view available work as a finite amount and as machines start doing some tasks, we won't have enough work to keep people busy.

If we look at history, we can see a clear trend that as machines have taken over jobs, peoples' time have been freed up so that they can spend more time on doing creative or otherwise higher skilled tasks.

Someone who worked 10 hours a day spinning threads can now spend 8 hours a day coming up with designs for clothes to make with the thread for example.

 

The cost of production has also dramatically dropped as industrialization has taken place, which has lead to higher salaries and shorter work days (which again, means more time for fun and creative stuff).

[Shally]

I don't get why it's doing this, maybe I read wrong but it's sooooooooo much further ahead in chess than everything else so why would it bother to learn human creativity? Is it a side project for the dev team? 

[Deus Voltage]

Just now, Shally said:

I don't get why it's doing this, maybe I read wrong but it's sooooooooo much further ahead in chess than everything else so why would it bother to learn human creativity? Is it a side project for the dev team? 

From the little research that I've done, people seem to agree that it has a bit of distance to go before being called "the best chess AI (since it hasn't defeated Stockfish 10 yet). Still fascinating though.