We are doomed - AlphaGo Zero, learning only from basic rules

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Vinvin
Posts: 4020
Joined: Thu Mar 09, 2006 8:40 am

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Vinvin » Thu Oct 19, 2017 10:00 am

Amazingly, when you click on the link on "nature" in the sentence "Details of the programme were published in the journal Nature." here : http://www.telegraph.co.uk/science/2017 ... 000-years/
you can read the whole article !

But direct link in the forum doesn't work.

duncan
Posts: 6850
Joined: Mon Jul 07, 2008 8:50 pm

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by duncan » Thu Oct 19, 2017 10:08 am

Leo wrote:
duncan wrote:
Leo wrote:Has AI done something other than play games? Something that would help humanity. Maybe cheat death for some of the worlds billionaires.
http://www.bbc.co.uk/news/technology-41651839

US scientists are using artificial intelligence to predict whether breast lesions identified from a biopsy will turn out to cancerous.

The machine learning system has been tested on 335 high-risk lesions, and correctly diagnosed 97% as malignant.

It reduced the number of unnecessary surgeries by more than 30%, the scientists said.
I am glad to hear that. Those are the kind of things that matter, not what machine beat what human in some game, and then brag about how great they are.
that is the AlphaGo Zero's team ultimate goal as well.

https://deepmind.com/blog/alphago-zero- ... g-scratch/

While it is still early days, AlphaGo Zero constitutes a critical step towards this goal. If similar techniques can be applied to other structured problems, such as protein folding, reducing energy consumption or searching for revolutionary new materials, the resulting breakthroughs have the potential to positively impact society.

Cardoso
Posts: 275
Joined: Thu Mar 16, 2006 6:39 pm

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Cardoso » Thu Oct 19, 2017 10:40 am

For that one I think they will take milleniums, or simply never.
They can't cure migranes or diabetes, much less stop aging.
And I think some people expect too much from science.
My mother had a severe skin disease on her feet called hyperkeratosis, with profound cracks in the sckin wich hurted badly, she was treated with the best doctor in the field in the country (portugal), with very agressive medications, and none of the several treatements worked. Desperate my mother tryed some plant called "malvas" in portuguese, after 2 weeks she was much better, after 6 weeks she was completed cured and the problem went completely away.
I think the human body is too complex for science.
Even a single cell is tremendously complex.

Vinvin
Posts: 4020
Joined: Thu Mar 09, 2006 8:40 am

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Vinvin » Thu Oct 19, 2017 11:01 am

I hope for such an experience for chess : starting a very deep learning with only basic rules, piece coordinate and piece interaction.
Even no "piece value" concept hardcoded.

User avatar
reflectionofpower
Posts: 1447
Joined: Fri Mar 01, 2013 4:28 pm
Location: USA

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by reflectionofpower » Thu Oct 19, 2017 11:12 am

It would be interesting for chess to be done this way BUT how will it benefit the average chess user if we do not have the "juice" to run the program?

The programs of today on an average machine can beat the best of humanity.
"hodie mihi, cras tibi"

Lonnie

Vinvin
Posts: 4020
Joined: Thu Mar 09, 2006 8:40 am

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Vinvin » Thu Oct 19, 2017 11:34 am

reflectionofpower wrote:It would be interesting for chess to be done this way BUT how will it benefit the average chess user if we do not have the "juice" to run the program?
When the neural network is trained, it can be exported in a file to be use on every computer.
reflectionofpower wrote:The programs of today on an average machine can beat the best of humanity.
Yes, but sometimes top engines do very poorly on some positions. The idea is to get an engine strong in EVERY chess positions !

User avatar
Leto
Posts: 2024
Joined: Thu May 04, 2006 1:40 am
Location: Dune

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Leto » Thu Oct 19, 2017 12:05 pm

So when is Google's Deepmind team going to take on Stockfish, Houdini, and Komodo in chess?

Vinvin
Posts: 4020
Joined: Thu Mar 09, 2006 8:40 am

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Vinvin » Thu Oct 19, 2017 12:51 pm

Leto wrote:So when is Google's Deepmind team going to take on Stockfish, Houdini, and Komodo in chess?
Yeah ! Let's beat this puny engines 100-0 ! :twisted:

brtzsnr
Posts: 426
Joined: Fri Jan 16, 2015 3:02 pm
Contact:

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by brtzsnr » Thu Oct 19, 2017 12:53 pm

duncan wrote:While it is still early days, AlphaGo Zero constitutes a critical step towards this goal. If similar techniques can be applied to other structured problems, such as protein folding, reducing energy consumption or searching for revolutionary new materials, the resulting breakthroughs have the potential to positively impact society.
This. Developing AIs for games allows one to develop general AI techniques without the ethical and cost implications of testing on real objects, or, worse, animals and humans.

Vinvin
Posts: 4020
Joined: Thu Mar 09, 2006 8:40 am

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Vinvin » Thu Oct 19, 2017 12:54 pm

Vinvin wrote:Amazingly, when you click on the link on "nature" in the sentence "Details of the programme were published in the journal Nature." here : http://www.telegraph.co.uk/science/2017 ... 000-years/
you can read the whole article !

But direct link in the forum doesn't work.
DeepMind team wrote:Conclusion

Our results comprehensively demonstrate that a pure reinforcement
learning approach is fully feasible, even in the most challenging of
domains: it is possible to train to superhuman level, without human
examples or guidance, given no knowledge of the domain beyond basic
rules. Furthermore, a pure reinforcement learning approach requires
just a few more hours to train, and achieves much better asymptotic
performance, compared to training on human expert data. Using this
approach, AlphaGo Zero defeated the strongest previous versions of
AlphaGo, which were trained from human data using handcrafted fea-
tures, by a large margin.
Humankind has accumulated Go knowledge from millions of games
played over thousands of years, collectively distilled into patterns, prov-
erbs and books. In the space of a few days, starting tabula rasa, AlphaGo
Zero was able to rediscover much of this Go knowledge, as well as novel
strategies that provide new insights into the oldest of games
The experience of strong human games was a weak part of the learning ;-)

Post Reply