DeepMind's AlphaZero plays chess like a tornado in the junkyard

[pitabread]

Evolutionary theory does not allow for a task to complete or a process of learning, or there being any sense of there being a predetermined goal (except in all cases by metaphor).

In a way it does, though. The "goal" of biological is the survival and reproductive viability of organisms. And the 'learning' part is simply the recursive process of biological evolution which is a form of trial and error with respect to viable biological forms.

 

[Tom 1]

In a way it does, though. The "goal" of biological is the survival and reproductive viability of organisms. And the 'learning' part is simply the recursive process of biological evolution which is a form of trial and error with respect to viable biological forms.

Haha yeah - ‘in a way’ - that’s the kind of work around/ metaphor that has me believing that sñr Dawkins has a subconscious belief in a volitional life force. Take away the metaphor and all you have is a description of a building being built minus the builder. That’s my view in any case.

 

[Tom 1]

In a way it does, though. The "goal" of biological is the survival and reproductive viability of organisms. And the 'learning' part is simply the recursive process of biological evolution which is a form of trial and error with respect to viable biological forms.

I’m not sure if pattern learning algorithms are strictly analogous to the random generation of mutation

 

[pitabread]

I’m not sure if pattern learning algorithms are strictly analogous to the random generation of mutation

I don't think they are strictly analogous either, but loosely speaking the concept is similar.

 

[FrumiousBandersnatch]

Googles AI engie DeepMine has been trained to play chess. They call the program AlphaZero. Only given the rules of chess it had to experiment by random trial and errors games with itself. After 4 hours traning it was matched against the best chess engine in the world; Stockfish. After 100 games, AlphaZero won 28 and lose zero time, with 72 draws

For those not initiated this is quite impressive. As quoted form Chess.com:

"This would be akin to a robot being given access to thousands of metal bits and parts, but no knowledge of a combustion engine, then it experiments numerous times with every combination possible until it builds a Ferrari."​

Which is similar to the junkyard tornado argument creationists like to use to "disprove"evolution. It is like DeepMind's AlphaZero learned to plays chess similar to a tornado in a junkyard would create a Boeing 747. I know some Creationist now will claim the AlphaZero been made by an intelligent designer. I grant that, but it misses the point; AlphaZero still had to figure out but itself, by random trial and errors, how to play chess on super human level.

I am curios what take creationist have on this; if randomness cannot create design, no matter what time is given, then what did cause AlphaZero to achieve a superhuman performance in chess in less than four hours time?

Indeed, but you haven't mentioned the really impressive part - AlphaZero can learn to play any game of this type from only the rules. It has also thrashed AlphaGo and AlphaGo Zero (which beat AlphaGo) - at Go, of course.

 

[Tom 1]

I don't think they are strictly analogous either, but loosely speaking the concept is similar.

I’m not sure how loose you can get and still have a connection between a pre-designed ability to learn in a specific fashion within clearly defined parameters, and a process that involves randomness leading to improbable outcomes leading to modified randomness (putting it a bit crudely I know) but it would be interesting to have that explained by someone who understands it better than I do.

 

[Tom 1]

I don't think they are strictly analogous either, but loosely speaking the concept is similar.

It’s late here, I’m tired, and it might sound like I’m being facetious but I’m not, I am genuinely interested in this. The analogy, I think, falls down in a few different ways. One of them is the matter of will. So far at least, AI has no will. It will accomplish (or not) what it has been designed to do, within set parameters, and then stop. Just stop. It won’t decide it wants to do something else, or rehash what it did and do it better just for the sake of it etc. If you strip out the metaphors, what is it that drives life - what is the explanation for why ‘life finds a way’, in the purely scientific worldview?

 

[Strathos]

Indeed, but you haven't mentioned the really impressive part - AlphaZero can learn to play any game of this type from only the rules. It has also thrashed AlphaGo and AlphaGo Zero (which beat AlphaGo) - at Go, of course.

Let's see it try football

 

[Speedwell]

No, but if I have understood what you are saying you are comparing a piece of software with the theory of random selection. The software is designed to complete a task; in that sense it is volitional, i.e it has a task to accomplish and a preprogrammed ability to complete that task. It does this by trial and error and by remembering and applying what it learns through this process. Evolutionary theory does not allow for a task to complete or a process of learning, or there being any sense of there being a predetermined goal (except in all cases by metaphor). What you are describing is a better argument for creationists than for those who believe in a blind process of non volitional random selection.

The "goal" of evolution is that creatures remain fit to survive in changing environments. It does this "by trial and error and by remembering and applying what it learns..."

I am not sure what you mean by "non-volitional random selection."

 

[Tom 1]

The "goal" of evolution is that creatures remain fit to survive in changing environments. It does this "by trial and error and by remembering and applying what it learns..."

I am not sure what you mean by "non-volitional random selection."

What is the volition and memory in the process of random selection?

 

[Astrophile]

Googles AI engie DeepMine has been trained to play chess. They call the program AlphaZero. Only given the rules of chess it had to experiment by random trial and errors games with itself. After 4 hours traning it was matched against the best chess engine in the world; Stockfish. After 100 games, AlphaZero won 28 and lose zero time, with 72 draws

For those not initiated this is quite impressive. As quoted form Chess.com:

"This would be akin to a robot being given access to thousands of metal bits and parts, but no knowledge of a combustion engine, then it experiments numerous times with every combination possible until it builds a Ferrari."​

Which is similar to the junkyard tornado argument creationists like to use to "disprove"evolution. It is like DeepMind's AlphaZero learned to plays chess similar to a tornado in a junkyard would create a Boeing 747. I know some Creationist now will claim the AlphaZero been made by an intelligent designer. I grant that, but it misses the point; AlphaZero still had to figure out but itself, by random trial and errors, how to play chess on super human level.

I am curios what take creationist have on this; if randomness cannot create design, no matter what time is given, then what did cause AlphaZero to achieve a superhuman performance in chess in less than four hours time?

Would this be analogous with God creating the first self-replicating molecules and leaving them to learn how to evolve by random trial and error, or would it be more like God creating a large number of organic molecules and leaving them to develop into life-forms that could evolve, also by trial and error?

 

[FrumiousBandersnatch]

Let's see it try football

Football isn't a game of that type.

 

[Speedwell]

What is the volition and memory in the process of random selection?

The "volition" is the constant production of randomly distributed somatic variation. The "memory" is the storage of the genetic record of reproductively successful variants in the gene pool.

 

[brinny]

Ummmmmm....SOMEbody had to "create" the robot and its "intelligence" from the gitgo.

Didn't they?

 

[Chesterton]

Anyhow, I strongly disagree. Can you please explain why you do not think AlphaZero evolved by random play and selection?

Because there's no such thing as randomness.

That is, if AlphaZero was given knowledge by design how to play chess then why was this design not already achieved by Stockfish - the top most ranked chess engine in the world until AlphaZero came around a frankly made Stockfish look like an amateur?

I think the wording is a bit confused. You note that Alpha was given something, and then ask why Stockfish didn't achieve something. Neither program achieved anything. Alpha simply was given instructions that Stockfish was not.

 

[FrumiousBandersnatch]

I’m not sure how loose you can get and still have a connection between a pre-designed ability to learn in a specific fashion within clearly defined parameters, and a process that involves randomness leading to improbable outcomes leading to modified randomness (putting it a bit crudely I know) but it would be interesting to have that explained by someone who understands it better than I do.

A possible interpretation is that evolution is itself a simple (although inefficient) learning algorithm that learns how to make creatures well-adapted to many environments, and in time, creatures who can adapt to many environments. The most successful of these eventually use learning to adapt environments to themselves.

 

[In situ]

Evolutionary theory does not allow for a task to complete or a process of learning or there being any sense of there being a predetermined goal

Yes it does.

The fact that evolution progress along the largest gradient in the local fitness landscape and the global landscape change dynamically and thus potentially change the position of the global maximum in the landscape does not mean it cannot find the global maximum, i.e. "a predetermined goal", because it can. If there does not exists "a predetermined goal", i.e. no detectable gradient in the fitness landscape, evolution degenerate to a random walk in the local neighbourhood. Then, but only then, you are right.

Evolution, like the Monte Carlo Tree Search used by AlphaZero, is two different search algorithms which finds a global solution by repeated local random searches. In fact without the random local searches none of them would work. The more random the local search is, the more precise, accurate and directed will the global search be.

 

[In situ]

AlphaZero didn't "create a design" it was given the rules of a game and according to the given rules of its program it 'learnt' to play the game.

Rules in chess tells what you can do. However, the rules does not tell you how to act or even win. It is the latter AlphaZero discovered by itself with random searches.

There are similar rules in nature that evolution must follow. They are called the Laws of Nature. To be able to do anything you need rules. Without rules, or assumptions, you degenerate to chaos. The rules of chess are the "law of nature" given to AlphaZero in order for it to be able to achieve something. In this case, play chess because you do not want it to plat tic-tac-toe when it is supposed to play chess.

 

[Kenny'sID]

I doubt you'll get any creationists here able to look past this point.

Exactly, and why should we? But nonetheless, I will.

For those not initiated this is quite impressive. As quoted form Chess.com:

"This would be akin to a robot being given access to thousands of metal bits and parts, but no knowledge of a combustion engine, then it experiments numerous times with every combination possible until it builds a Ferrari."
Which is similar to the junkyard tornado argument creationists like to use to "disprove"evolution. It is like DeepMind's AlphaZero learned to plays chess similar to a tornado in a junkyard would create a Boeing 747.

That is the most ridiculous part of the whole thing...because chess.com says that, it's true?

Give that robot those parts and talk to me afterwards.

 

[In situ]

Because there's no such thing as randomness.

The Monte Carlo Tree Search used by AlphaZero wont even work if it isn't randomized so I do not understand what you mean when you say there is no randomness.

I think the wording is a bit confused. You note that Alpha was given something, and then ask why Stockfish didn't achieve something. Neither program achieved anything. Alpha simply was given instructions that Stockfish was not.

Stockfish has been given the rule of chess and heuristic tree searches and strategics discovered by humans etc etc in order to be able to play chess, i.e. Stockfish has been designed by humans, over several decades, to play good chess.

On the other hand, AlphaZero has only been given the rules and then discovered, by itself, how to play chess, i.e. AlphaZero was not designed to play chess, yet AlphaZero can play chess and it achieved superhuman capability with randomness and no plan or goal whatsoever in just a few hours, i.e. just like evolution.

Btw, AlphaZero is a game changer. Before AlphaZero, humans had to teach chess engines how to play chess. However, AlphaZero was has not been taught by humans to play chess, instead it seems like AlphaZero might actually teach human how to play chess instead. I.e. AlphaZero is better at chess than any human - ever.