Cyprus-Forum.com

[Pyrpolizer]

Cypriot Demis Hassabis:

https://gizmodo.com/alphazero-annihilat ... 081928/amp

https://chess24.com/en/read/news/deepmi ... shes-chess

[Get Real!]

What’s the fuss all about?

Just as there are only 10 ways you can START a game of chess, and 12 ways you can make your second move, so too can every other outcome be recorded along the way although they will run in their hundreds and then in their thousands, etc as the game progresses.

In other words, you can SAVE all the possible arrangements to a hard disk and then when you play an opponent you can follow the best path based on what your opponent has done by retrieving that pattern from the hard disk you previously saved.

Essentially you are creating LOOKUP TABLES of every possible route so as to always select the best possible path for you.

The saved “patterns” directory could look something like this…

C:\Saved\Move001\ > Contains the 10 permutations (pattern files)
C:\Saved\Move002\ > Contains the 12 permutations (pattern files)

etc

The content of a typical pattern file can look like this… “5B,8C,3E”

So it’s not hard to write a program that will record every single eventuality of a chess game and save it to a hard disk so that you can then LOOKUP to defeat all opponents.

Throw in the word “artificial” and everyone thinks that the computer can think!

[Get Real!]

If you want to beat EVERY single opponent, there is no other way but to record every single eventuality and then retrieve the pattern as required. This will guarantee that your absolute worse outcome will always be a draw.

The truth is there’s no such thing as a computer “thinking” because computers can’t think but they can RECORD (save) and RETRIEVE (read) vast amounts of data (read chess playing patterns) very fast, which makes them appear “intelligent”.

The number of routes/paths that a chess game can take is LIMITED. The path permutations may run into thousands or even millions but they are still LIMITED, so it *is* possible to record all eventualities to a storage space.

It took this server 4 hours of pitting two opponents to save all the winning paths it could record in this timeframe and in this way “educated” itself how to play!

I’m pretty sure that’s how these guys did it because I know computers can’t think but they can retrieve LOOKUP TABLES very fast.

In the past, what chess games did (instead of recording millions of patterns) was to use “look-ahead” steps so as to minimize the chance of mistakes… in other words the game would simulate the next 2, 5, 10, or 20 steps to find the best possible move… that’s how you got the “difficulty level” in a game. You were basically adjusting the number of “look-ahead” steps.

I wrote a chess game ages ago and that’s how I did it. Back then we did not have 64bit, 128bit or 256bit servers so as to have access to MASSIVE binary trees that can aid in recording every single outcome so it was out of the question and we could only rely on limited “look-ahead” moves.

[Robin Hood]

As I have said before my knowledge of computers is limited but I do know how they work in principal. GR, you have put into words what I have come to believe about computers and the effect this so called Artificial Intelligence is going to have on our lives. It is all down to memory coupled to the speed at which that memory can be accessed and used to make a decision.

If you look at the ‘professions’, they are inevitably jobs that require memory as opposed to original thinking or skills of dexterity. They all require degrees of learning, which requires a good memory. So any job that depends on the human memory will eventually most likely be replaced by a computer.

Take a visit to the doctor; he/she asks you a series of questions; they take BP and temp and look at your tongue and into your eyes looking for abnormalities. They then relate that to what they have learned and give their diagnosis. If they miss vital signs or don’t ask all the right questions, then they get the diagnosis wrong.

A computer will not forget to ask pertinent questions and can do it millions of times faster than the human mind. But I doubt a computer will ever perform brain surgery without the skill set of a human operator.

I have found ‘intelligence’ based simply on academic ability and thus memory, is often counter productive. Indians I have worked with are inevitable well qualified, often with multiple degrees up tp PhD level and have phenominal memories but their ability to apply that knowledge is often very limited. They do not have the ability to work out what to do because that was taught separately and they have difficulty combining the two elements of knowledge and original thinking.

An interesting subject and we can see how this has progressed in our own lifetime. I am told that in the UK now banks don’ have people in them any more ..... you ‘talk’ through computers. A garage mechanic plugs your car into a computer to diagnose faults. The scope is endless ....... but a computer will never watch over a flock of goats as a shepherd would ..... unless they chip all the goats and herd them with drones!!!!

[Get Real!]

Every game is a sequence of something that looks like this...

“5B-6B,8C-3E,2E-1A,6D-3D”

In this case the odd pairs are Color1 and the even are Color2. Color1 moved 5B to 6B and then Color2 responded with 8C > 3E, etc.

By the end of the game I will have a long string of these characters which will serve as a UNIQUE SIGNATURE of this specific game path.

With a super fast computer I can simulate two virtual players playing for the next 24 hours and just keep saving these signatures in a database as they finish each game; which will only last for a few seconds p/game HEADLESS (no usage of the graphics card necessary… no need to look at them playing).

The end result is a huge file of say 5 billion signatures which I can later LOOKUP when I’m playing an opponent.

Presto… I’m a chess “genius” that will always win or draw!

They gave it away when the article said they had many wins and some draws but zero losses… and that means PERFECTION, so I knew there and then that they weren’t relying on complex playing logic but signatures.

Someone here has done some calculations on the total number of moves possible etc, but I'm pretty sure a good mathematician can tell you the EXACT maximum number of game paths possible...

https://www.chess.com/chessopedia/view/ ... -and-chess

[Pyrpolizer]

What a huge disappointment reading all this nonsense from GR!!

For a start I don't think he even understood what this new achievement is all about and why they now turned to chess to demonstrate it's power (after having been demonstrated on other equally highly tactical games like GO, Joseki etc.)

Chess is just one of the examples that a computer would need billions of trillions of years to examine all possibilities and construct a tree search before making the first move. Just a simple move in a middle game where white holds all his pieces and each one can on the average make 4 moves set on the Stockfish program powered with 64 CPUs doing 70 million moves per second would need 9 trillion years to complete a simple analysis at depth 7 using the standard TREE approach. I can now imagine what kind of silly moves GR's program was doing.

This is not about a Cypriot who's the CEO of Google Alphazero team, nor about chess though.
it's about a tremendous new achievement in neural nets and artificial intelligence that's going to change the world for ever.
Do some reading for Christ sake. The computer was just given the rules of chess nothing else. It learned to play at top level within just 4 hours by just TRAINING ITSELF without any further human input!!

What we are having here is a combination of the classic tree search with convolutional neural nets. The experiment with chess was just to prove the huge power of this new achievement.
Standard neural nets have been tried on chess in the past and they failed completely. Same with standard tree searches like GR's program .

Here' are some links on You tube.More should appear on your browser.
https://www.youtube.com/watch?v=WXHFqTvfFSw
https://www.youtube.com/watch?v=tXlM99xPQC8&t=26s

[Get Real!]

Well of course you; having such an extensive programming and computer science background, are far more capable of appreciating what these guys are feeding the public!

[Get Real!]

Pyro, it’s called ARTIFICIAL intelligence because it’s pseudo intelligence (imaginary gimmick) and not real, so stop getting excited over metaphors and buzz words.

Without countless lookup tables of data, a computer can never appear “intelligent” because there’s no such thing as an algorithm that can LEARN despite their claims. (read LIES)

Show me a piece of source code that demonstrates learning and I’ll believe you… but I KNOW for a fact that it doesn’t exist!

It is definitely not possible with the current transistor-based technology we have at our disposal. Maybe in the future if something else is invented that doesn’t have the crippling limitations of today’s computers…

As an example, is the Wikipedia intelligent? It has the whole world’s knowledge stored away and by typing something it can bring up detailed information about it.

Well, the Wiki is not intelligent but a data retrieval system of information that was previously entered by humans... tons of manual human labour has made the Wiki seem “intelligent”.

I assure you there is no algorithm (source code) that can “learn” in any programming language… it’s a pipe dream.

[Get Real!]

Let’s take the sophisticated robots made by Honda, are they intelligent?

Honda has added to its robots motors for movement, sensors for moving about without bumping into things, speech recognition software, image recognition software, software for the structure of the Japanese language so that it may respond “intelligently” to some questions with structured sentences, countless lookup tables for a bunch of other miscellaneous requirements and so forth.

Now which part of all that demonstrates “learning”? NOTHING!

It is actually a very stupid machine that had to be GIVEN everything in order to do very limited things and thus appear “intelligent” to the public.

[Pyrpolizer]

What you are talking about has little or nothing to do with neural nets and artificial intelligence.
Don't mix up standard computer programming that always outputs the same thing for the same input.
(as e.g is the case of a robot).

Also don't mix up standard Monte Carlo tree searches which basically rquires a huge database for the algoritm to make a decission.

The DeepMind team had already tried all known methods of supervised training using human data as it can be seen here:
https://www.youtube.com/watch?v=XuzIqE2IshY

The breakthrough came with AlphaGoZero (as it's name implies it got absolutely no supervision and no human data).
And yes it learned by itself.

What's the meaning of "learning" to you?? Learn to be polite? No we don't need the computers to learn manners we need them to do specific tasks.
Here are some examples of tasks from IBM's website:
https://www.ibm.com/developerworks/libr ... index.html

Are you still unable to understand the importance of this new discovery?