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

[durangodawood]

...Almost everyone answering has responded with telos as the difference....

The real difference between AZ (or evolving life) and the tornado>747 is pre-existence of the algorithm (or DNA).

And yes. AZ does demonstrate that iterations of random action + feedback can result in complex and novel forms.

 

[In situ]

I don't see how this is similar, unless you think tornadoes are created, trained and programmed by intelligence.

See post #140.

 

[xianghua]

we know that nature cant evolve since there is no st epwise way to evolve nature. if we will start with a self replicating molecule we will not get a self replicating walking creature\robot after millions of years.

 

[In situ]

The real difference between AZ (or evolving life) and the tornado>747 is pre-existence of the algorithm (or DNA).

This seasm to be a composite answer of the type "the rules was given" and "AlphaZero is not analogues to evolution".

Please, see post #86 for a reply to "the rules was given" and post #136 for a reply to "AlphaZero is not analogues to evolution". To understand what my question is about please see post #140.

 

[durangodawood]

This seasm to be a composite answer of the type "the rules was given" and "AlphaZero is not analogues to evolution".

Please, see post #86 for a reply to "the rules was given" and post #136 for a reply to "AlphaZero is not analogues to evolution". To understand what my question is about please see post #140.

What about the 2nd half of my post.

Wasnt that part directed right at the your OP question - basically agreeing with you?

 

[In situ]

we know that nature cant evolve since there is no st epwise way to evolve nature. if we will start with a self replicating molecule we will not get a self replicating walking creature\robot after millions of years.

My reply to this is found in post #140. In short; you assume evolution cannot work since it is not theistic. However, anyone is free to assume it is theistic, and therefore you still haven't answer the question.

 

[In situ]

What about the 2nd half of my post.

Wasnt that part directed right at the your OP question - basically agreeing with you?

Can you please provide with a link or the number to that post so I can have a look at it again.

Edit: Never mind, I know what you refer to now...

 

[durangodawood]

Can you please provide with a link or the number to that post so I can have a look at it again.

#141, here's the whole thing:

The real difference between AZ (or evolving life) and the tornado>747 is pre-existence of the algorithm (or DNA).

And yes. AZ does demonstrate that iterations of random action + feedback can result in complex and novel forms.

 

[In situ]

What about the 2nd half of my post.

Wasnt that part directed right at the your OP question - basically agreeing with you?

The intention of you post was unclear to me. I understood the 2nd part as being the telos answer to my question. Hence the reference to post #140 which address that. If you agree, I am confused about what point you are trying to make besides agreeing?

 

[In situ]

And yes. AZ does demonstrate that iterations of random action + feedback can result in complex and novel forms.

It is more than that. AlphaZero is analogues to evolution as explained in post #136.

 

[In situ]

Didnt the algorithm tell AZ how to do exactly that?

Yes and no.

Yes, since it use Monte Carlo Tree Search (analogues to evolution), i.e. it was told to do it with random trials (analogues to mutations) and errors (analogues to selection) - see post #136.

No, since AlphaZero was not given prior knowledge on how to play chess, i.e. AlphaZero was not told how to act in the world - see post #86.

I could even go so far as to say AlphaZero was not even told the rules but discovered them as well, but to claim that I need to get a deeper understanding of the inner workings of AlphaZero which I do not have right now.

To my understanding, AlphaZero discovered how play chess by by observing what happens if it takes certain actions, i.e. AlphaZero observes the effects of the rules. If I am to speculate now, AlphaZero might not be able to observed the cause directly which is the the rules themselves. If so, then AlphaZero does not even have knowledge of the rules. In principle the rules, which are patterns in themselves, can still be discover by just observing the effect from its own actions (and this is actually what happen in some research AI-robots - and in babies which e.g. loves to explore the effect of gravity, i.e. drops stuff on the floor to every parents annoyance).

 

[durangodawood]

...In principle the rules, which are patterns in themselves, can still be discover by just observing the effect from its own actions (and this is actually what happen in some research AI-robots - and in babies which e.g. loves to explore the effect of gravity, i.e. drops stuff on the floor to every parents annoyance).

So AZ moves a knight in a straight line. Then what happens?

 

[In situ]

it use Monte Carlo Tree Search (analogues to evolution), i.e. it was told to do it with random trials (analogues to mutations) and errors (analogues to selection) - see post #136.

Rereading what I just wrote, it strikes me; this is actually a tentative answer to my own question; the difference is that AlphaZero was told to use evolution. However, my objection would then be that evolution is given a priori in nature, i.e. evolution is implied in the laws of physics, and we have only observed that evolution exists. Which is something a creationist in turn will repel with flat out denying facts. So back to square zero.

Is there a better objection to this tentative answer?

 

[In situ]

So AZ moves a knight in a straight line. Then what happens?

See post #136 for an answer. If something is unclear, please ask.

 

[In situ]

So AZ moves a knight in a straight line. Then what happens?

Lol, did not see you broke the rules at first. Good question. The answer is, it cannot, and will never, happen because such move violates the rules. And this is the reason why the rules must be given, because we want AlphaZero to teach itself to understand chess and not randomly moving around the pieces on the board.

Also see post #86.

 

[durangodawood]

See post #136 for an answer. If something is unclear, please ask.

In #136 you say This is why the rule must be give as minimum knowledge.

But I responded to your other post where you said: In principle the rules, which are patterns in themselves, can still be discover by just observing the effect from its own actions

We have a contradiction here.

So AZ moves a knight in a straight line. Then what happens?

 

[In situ]

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

I think I have to agree now, because I think I have came to an understanding on how a hard core creationist will respond to this. I have summarized this in post #153. In short, it falls back to denial.

 

[In situ]

So AZ moves a knight in a straight line. Then what happens?
We have a contradiction here.

This is a straw man.

In essence the chess board is the analogues to an universe. AlphaZero exists in this universe. This univers has a set of natural laws which cannot be broken under any circumstance. The laws are the rules of chess.

E.g. this means AlphaZero is not allowed to make an arbitrary number of moves in one turn but are restricted to one move each turn. This is a rule that need to be enforced, otherwise AlhpaZero will figure out that the best way to win is just to make sufficient enough move in the first turn to win, but how will AlphaZero even know it won if there is no rule telling what the winning condition is? I.e. we don't play chess any more but are dealing with chaos. To avoid chaos we need to impose order on chaos. Order is imposed by rules.

Any system which pretends to do something meaningful must enforce rules. But rules cannot be enforced unless they are known. This is why rules must be given as domain knowledge, but that does not imply AlhpaZero has to know the rules. It is sufficient for AlphaZero to just observe the consequences of it own actions, i.e. observe the effect of the rules in order to learn to play chess.

With no rules given, i.e. no order imposes, there is no chance that anything at all will, or can, be learned because design, i.e. learning, is virtually impossible to emerge from chaos. Design is unlikely to emerge from chaos in finite time, it needs order for that - unless you got infinite time because then anything which can happen will happen. But you don't have infinite time, therefore you need order.

Now you know why a deity is not needed to create a universe with intelligent biological life which in turn creates Boeing 747's and chess engines which in turn learn by themselves to play chess. You also now know why tornadoes does not assemble Boeing 747's in junk yards.

 

[durangodawood]

That is a straw man.

In essence the chess board is the analogues to an universe. AlphaZero exists in this universe. This univers has a set of natural laws which cannot be broken under any circumstance. The laws are the rules of chess.

E.g. this means AlphaZero is not allowed to make an arrbitrary number of moves in one turn but are restricted to one move each turn. This is a rule that need to be enforced, otherwise AlhpaZero will figure out that the best way to win is just to make sufficient enough move in the first turn to win, but how will AlphaZero even know it won if there is no rule telling what the winning condition is? I.e. we don't not play chess any more but are dealing with chaos. To avoid chaos we need to impose order. Order is imposed by rules.

Any system which pretend to do something meaningful must enforce rules. But rules cannot be enforced unless they are known. This is why rules must be give as domain knowledge, but that does not imply AlhpaZero has to know the rules. It is sufficient for AlphaZero to just observe the consequences of it own actions, i.e. observe the effect of the rules in order to learn to play chess. But with no rules given, i.e. no order imposes, there is no chance that anything at all will, or can, be learned because design, i.e. learning, is virtually impossible to emerge from chaos. Design needs order to emerge.

I see what youre getting at. But it doesnt help at all when you say things like "This is why the rule must be give as minimum knowledge".... and then you go on to propose that its NOT necessary that rules are known.

 

[Astrophile]

You can at most achieve order, but not design. From order you can then create design. However, just because A implies B and B implies C does not mean A implies C, i.e. you cannot just assume the transitive law is true for every relation, that is you are not entitled to just skip one phase in how design can be achieved without actually showing this to be the case. And to my knowledge nobody has done that yet.



Feel free to disagree. That is the best way to straighten out once own thoughts.



You can claim that, but it does not make your claim true.



Beside the non sequitur, chaos and chaotic system is not the same thing. A chaotic system is simply put a nonlinear system. Nonlinear system posses order, i.e. the trajectory of the system. Here is an example of a chaotic trajectory:

Does the picture above look like the picture below to you?

A chaotic system is not the same as chaos, in fact chaos is not a system at all because it is random - which is the complete opposite to a system.




I suspect you are not clear on the definitions of things and therefore you make incorrect conclusion. I suspect the reason a tornado emerge is because of the order which exists in a weather system, i.e. it is the order in chaotic system which gives rise to design not the randomness in the chaotic regions.

And yes, if you think cloud looks designed, then we definitely mean different things with design.

Thank-you for your informative post. I think that I am going wrong here and that I had better withdraw from this discussion to think things through privately