OXO, Checkers, Chess & now Go

Go is just a mechanical game, so massive computation will eventually win over people. Computers are better at computing logarithms, too.

Eventually I guess that computers will win over any narrowly defined problem. But that's hardly "intelligence."

Executing the algorithms coded by a team of scientists and programmers is still just mechanical, just a big adding machine.

As long as people don't lie on surveys, which they often do.

On Thu, 28 Jan 2016 08:20:06 -0800, John Larkin Gave us:

I think it was Saudi Arabia that just recently outlawed Chess play in their country (by humans, of course).

Must not like folks displaying intelligence.. They have a Trump like fear of being put in their place.

Your lack of understanding of the deep subtleties of Go is unsurprising but be assured there is no way on Earth that anything less than human quality fast pattern matching will allow a machine to play it at the level of an international master. The rules and moves are very simple but game play is extremely complex with decisions made early on making an important contribution to the long term outcome.

It is intelligence when the problem is as generic as the strategic area capture problem of Go and a learning algorithm with refinements has now cracked it. Brute force never will. Quantum computers might.

Checkers 64 square, 10^18 states, game tree 10^31 (solved)

Chess 64 squares, 10^47 states, game tree 10^123 (computer wins)

Go 361 squares, 10^170 states, game tree 10^360

So it is very considerably more difficult than chess and had been totally out of reach of AI research on a full size board until very recently. Much like with chess in the 1980's no-one expected a computer to be strong enough to take on a top level Go player for quite a while.

Except that if you had read the article instead of jumping to conclusions you would have realised that it learnt the game by playing against versions of itself and other now inferior Go programs.

I was meaning deep sky surveys and pulsar searches where the final test of whether it is worth pointing a big scope at something still requires human intervention. Harnessed recently on Stargazing Live from Jodrell Bank and amazingly they got one really interesting find live.

And the actual data from Jodrell of the pulsar they found

Humans are so much better at pattern matching in noisy data - which is why captcha can sort humans from bots (most of the time).

Computers just execute code. With identical inputs, outputs are the same every time. I don't call that "intelligent"; maybe you do.

A quantum computer (which is certainly what the human brain is) could be a little more creative than a bunch of flipflops.

And why do people keep building humanoid robots? Makes no sense.

On Thu, 28 Jan 2016 11:50:47 -0800, John Larkin Gave us:

The supercomputer that played Jeopardy made decisions which were NOT "the same every time".

That was interesting, thanks. I'm not very up on AI, but I wonder if the problem of recognizing hand written addresses on letters (for the post office) has been solved.. and if not perhaps this will help.

George H.

Maybe the program reads some sort of randomnizer. But that's just another input.

Any program that does different things with identical inputs is broken.

My bank ATM got a handwritten check right. My sister's penmanship is pretty good though - I don't know if it could grok mine.

On Thu, 28 Jan 2016 13:52:43 -0800, John Larkin Gave us:

You always make retarded guesses when you are sadly uninformed.

You are an abject idiot.

Hah, no it's learning.... John, no offense, but did you read the article?

George H.

Any program that maintains variables "learns."

It's still just a computer program. Artificial yes, intelligent no.

It could certainly learn how to do medical diagnosis, especially to spot obscure problems that the average MD seldom sees. Climate is chaotic, so pattern recognition is less useful there.

There's a procedure for inverting sparse matrices where a random number generator built into the procedure can be proved to give the fastest inversion.

Anything that always does the same thing might not be broken, but it isn't as fast as it might be.

John doesn't know about chaotic systems, so the phrase "strange attractor" wouldn't mean anything to him

That isn't true for any of the modern game playing programs since they can never run to completion and the number of plys they search depends on operating system overheads and the previous state of the cache.

One trick for an expert chess player analysing a position is to take the program down a promising (to a human) line and let it compute deeply for a good while and then back up to the point where you think the program isn't right. The cache now containing some very deep evaluations down the line of interest that will supplant shallower evaluations and encourage it to follow your preferred line.

Almost all programs contain a random choice of equal continuation lines in practice since the evaluation function can sometimes give the same score to several possible continuations.

In the case of this particular Go program since it learns as it goes the game it plays the next time will be different at some stage since the network coefficients were altered by outcome of the previous game. Strengthened if it won and weakened if it loses. If you tried to play the same game again it will probably oblige at least until it comes to a node where several continuations evaluate to the same score and then it will choose lines where it has won or one of the others at random if a previous similar state led to losses. It doesn't often lose any more.

There is no evidence at all that the human brain is anything other than a classical machine using nerve impulses and a vast network of interconnections. If it *were* a quantum computer then we would find mentally factoring compound numbers into primes trivial.

The model systems being built by the likes of IBM are getting close to modelling a cats brain and have some prototype silicon implementations of modest chunks of a brain functionality arising I think out of joint research with EPFL.

They do it for fun. Most of the industrial robots look more like chunky servo powered anglepoise lamps or have you never seen any?

Single cells do impressive computation, without nerve impulses or vast networks or interconnects. Human brains do billions of operations a second using millisecond logic outputs. When you can explain how a person can hit a baseball going 95 MPH, I'll believe it's done with classic logic.

On Fri, 29 Jan 2016 08:42:30 -0800, John Larkin Gave us:

Meadowlark Lemon could shoot a mean blind over the head to the rear hook shot swish.

I can shoot pool without even touching the table and folks ask me how I do it all the time. I am amazed at the capacity to hit the exact spot for a one two or three rail bank myself, but I do it with pretty good consistency, and as the rail count increases, so does the complexity and the need to "hit that exact spot on the ball, with the exact English ball spin required to perform the shot successfully".

We are talking within tenths of a degree at the tangency point and

be right too. Too much and the english doesn't get time to work. Too little and the english does too much. Both mean a missed shot.

I can also take two steel balls of about 1 1/8" and 1 1/2" and place the larger one in my left hand and start throwing the smaller one at the larger one and catching it as it bounces off. I can get to the point where the separation between them is over a foot away and the ball still bounces right back to my throwing hand.

I am pretty good at frisbee golf too.

Sadly, for me, I am eluded by the exact point with which to hit a golf ball consistently.

NumeroUno:

hitting a golf ball right isn't so hard, but doing and putting some power behind it is damn near impossible

-Lasse

On Fri, 29 Jan 2016 17:02:32 -0800 (PST), Lasse Langwadt Christensen Gave us:

Hitting one right or left is damned easy. That is my problem.. :-) (see what I did there?) (same thing you said, including the reason why)

John von Neumann, who wrote Game Theory and had some degree of understanding of computers, said, "Chess is not a game" because you don't have to predict what your opponent is going to do. Your best possible move is the same no matter what he's thinking. I think Go and checkers are the same, and I don't know OXO unless it's another name for tic-tac-toe.