Giraffe Becomes an IM in 72 Hours !

The only unanswered question here is consciousness. Is this a property of a neural network? Is it an epi-phenomenon of information processing ( which has no relevance to the workings of the system ) or something that has an existence of its own right and makes a difference.

May be a neural network one day will think to itself: ok, I can't solve this damned chess by deep learning, but I build a machine, which will do it for me. Please give me a hammer and a screw driver...

P.S. By calculate, I meant the "brute force" technique of running through scenarios, however many ply ahead, btw.

Apparently not. It plays against itself and learns from it, developing some form of pattern recognition.

Nothing to do with intuition or "gut feeling", by the way.

A very informative article here. The comments also offer some insights.

Pattern recognition and gut feeling are the same. You have the gut feeling that a position is fishy because your neural network recognized it as an unfavorable pattern of chess pieces. You cannot calculate concrete forcing lines by which it loses, just you have this feeling that it is wrong.

Thank you for bringing this up OP, with the article link, and thanks kkl10, for the  follow up article.  

Pattern recognition is one thing, gut feeling is something else entirely. Gut feeling is a very high-level function, probably very complex. It's subjective by nature since it can only be experienced by sentient beings. Pattern recognition is a low-level function in comparison. It can be programmed into machines. One thing has nothing to do with the other.

Gut feelings can arise from many different things, not just from recognizing patterns or remembering things.

Impressive as they might be, these artificial neural networks are extremely archaic compared to biological brains. And we don't even know how the brain works exactly. ANN algorithms are merely an expression of how it is assumed that the brain works, and they're extremely limited in their capability. Just because some artificial neural network can "learn" something, it doesn't mean it has "gut feelings".

Top modern chess engines also use neural networks to learn, ever since the early versions of fritz.

The difference here, I guess, is that this one relies heavier on precise evaluation, and much less on calculation (if at all).

using software to model natural processes can be useful over traditional methods. I would think combining neural nets with genetic algorithms would allow computers to evolve. I.e. generate 10 different neural nets using a random genetic code. train each network with the same chess data. set up a round robin tournament and have each network play against the others. select the top two performing programs and breed them to produce the next generation. repeat as needed.

Michael Lai, the developer of this Giraffe, says that he has so much exploration w/ it that he'll have a Super GM much much sooner than later on. He's burning the seal oil as we speak........(forgive me....staying with the animal theme)

Watcha brings up a really important legal question about neural network "control" in #21. Could it be like "owning" a dependent (child) who never turns 18 ?....but can be adopted (sold or acquired) by another entity ?

Time will tell....fun !

....and I hope chess.com lines up an interview w/ developer Lai.

HINT HINT !

The prevailing mainstream thinking is that the brain is just a neural network. Alternative theories like that of Penrose's Orch-OR or the holonomic brain theory are trashed on a daily basis. ( For the holonomic brain theory there is at least some experimental evidence: information in the brain is stored more like a hologram than a stream of sequential bits as in computer memories. So if you cut out a relatively small part of the brain ( this has been done experimentally ), it is not true that part of the memories stored in that brain segment are lost as if they were localized to the part that was cut out, on the contrary, memories are not lost, the same way as if you cut out a small part of a hologram, the full image remains intact because it is stored in all over the hologram ( the quaility of the image may suffer, but no local part of the image disappears ). However if you cut out a large part, the integrity breaks down, memory is lost, the image of the hologram collapses. )

Alternative theories of the brain may ultimately be right. If there are relevant cognitive activities in the brain that cannot be reproduced by neural networks then to many this would come as a real surprise.

One thing is for certain: it is already evident that some coginitive activities of the brain can indeed be reproduced by neural networks with striking efficiency. If it turns out that techniques learned from the brain when applied systemtically can play a better chess than search algorithms, then this is a real breakthrough.

Emotion is very much a part of human decision making.  When Phineas Gage had an iron bar through his head which took out his amygdala (the emotion centre) he was no longer able to make decisions and spent hours and hours going through all possibilities to try to cognitively decide what the best decision was.  Humans use their gut - emotions, intuition - to decide which is the best option out of those available. 

If monte carlo methods applied to Go still don't surpass humans then humans either run monte carlo methods in their minds just more effectively than machines or you have to have a second look at Go with more advanced neural networks or any other technology that strives to imitate the information processing in the brain.

As far as searches are concerned, even if neural networks just provide a better evaluation function and they have to be combined with searches, but this combination surpasses search engines created by humans, this is still a breakthrough, though a somewhat lesser one.

(#37) Humans use their gut - emotions, intuition - to decide which is the best option out of those available. 

I'm not sold that gut feeling and emotion cross here. IOW's, is emotion required for gut feeling ?

The chinese say that Yin (feminine pink) is right brain. Yang is blue left brain. This is tradition....but it has merit.

I have hope that neural networks can get better into the right side....at least in my lifetime (I turn 40 on Halloween).