DeepMind founder motivates children

by ChessBase
11/15/2017 – DeepMind is a neuroscience-inspired AI company which develops general-purpose learning algorithms and uses them to help tackle some of the world’s most pressing challenges. The company's groundbreaking work includes the development of AlphaGo. This Deep Learning program defeated Go world champion Lee Sedol in 2016 — a breakthrough experts proclaimed to have arrived a decade ahead of its time. Demis Hassabis, Co-Founder and CEO of DeepMind, visited UCL Academy in Camden, UK, and took students on a journey through his career from chess prodigy to scientist and business leader and into the complex world of artificial intelligence.

ChessBase 17 - Mega package - Edition 2024 ChessBase 17 - Mega package - Edition 2024

It is the program of choice for anyone who loves the game and wants to know more about it. Start your personal success story with ChessBase and enjoy the game even more.

More...

Inspiring insights into A.I.

Press release

London: Students from Camden, Hampstead and surrounding areas were this week captivated by a presentation from one of the world’s most influential business leaders.

Demis Hassabis, Co-Founder and CEO of DeepMind, took time out of his busy schedule to visit UCL Academy, Camden. Here he took students on a journey through his career from chess prodigy to scientist and business leader and into the complex world of artificial intelligence. DeepMind, a world leader in A.I., was co-founded by Hassabis in 2010. The company was acquired by Google in 2014.

Hassabis, hailed by Time Magazine as one of the world’s top 100 most influential people, talked to UCL Academy students, and those from five other local schools, about the innovative work that DeepMind is undertaking in a fast-evolving field. He encouraged them to develop their passions and gave advice on developing core skills that are relevant in today’s rapidly changing world.

The event was part of a wider partnership between DeepMind and Chess in Schools and Communities, a charity which has introduced chess into over 800 schools around the country and delivers the game as part of a school’s curriculum. Hassabis, who himself was a child chess prodigy, also explained how chess and games like Go can develop the mind whilst giving students important life skills that can help them in their careers.

A select group of UCL Academy’s GSCE and A-Level computer science students were lucky enough to then benefit from a more intimate networking session with Hassabis, impressing him with the in-depth nature of their questions. Among the topics of conversation were DeepMind’s mission to ‘solve intelligence’, its most recent AlphaGo Zero paper, the growth of computing power and which A-Levels were the right ones to consider for a career in A.I. Students also learned of DeepMind’s ethical principles and the importance of building A.I. that benefits everybody.

Demis Hassabis, a former child prodigy in chess, reached master standard at the age of 13, with a rating of 2300. At the time he was the second highest rated player in the world Under-14. He and Malcolm Pein (right), Chief Executive of Chess in Schools and Communities, finished the afternoon by undertaking a simultaneous chess display.

Demis played six games at once against students who, in a unique format...

... used the Fritz computer programme to help with their strategy.

The final score was of the simul was 3-3. After Demis left Malcolm Pein took over six good positions he had achieved from the opening. But the children started to get the hang of Fritz... and the games became tough. Malcolm won three lost one and was cruising in the other two.

Malcolm: "It should have been 4-2 until I very diplomatically and of course involuntarily, allowed the head teacher a mate in one!" Can you spot it in the above picture (Black to play)?

Yes, the headmaster, Robin Street, got it and scored a win against the IM.

Demis Hassabis, Co-Founder & CEO of DeepMind, commented:

Chess has been immensely formative for me in helping to develop important skills such as problem solving, planning, visualisation, performing under pressure and the transfer of learning from one domain to another. At DeepMind we highly value these meta-skills and they are fundamental in our approach to business and to research. Indeed my journey into A.I. started with chess, because it challenged me to think about how we think.

I am therefore very proud to support the concept of learning chess in schools, particularly the work of Chess in Schools and Communities. Furthermore, I would encourage children to explore widely, find their passion and focus on learning how to learn.

Robin Street, Co-Principal of UCL Academy, said:

The focus of our curriculum is one of connectivity, helping and challenging the students to understand the relevance and links of subjects to each other. This systemised approach to collaborative group learning promotes an ethos of inter-disciplinary thinking, allowing the students to develop an appetite for a broader perspective on their studies.

Coupled with this, we aim to offer our students the opportunity to encounter ‘excellence’. This event has presented a unique opportunity for our students to interact with one of the world’s leading thinkers. This aspiration of meeting people who have achieved things that are extraordinary will help set the bar higher than it might otherwise be.

Melanie Dennig, Head of Computer Science at UCL Academy, noted:

The students have gained an incredible insight into the future from a person who is in the process of creating it. Our future will be built with A.I. in mind so anything that provides us with a better idea of what that world will look like is hugely beneficial, especially to the next generation, who will be affected most. Meanwhile we have demonstrated that a human-element stays central in this development process and shown, through effective use of collaboration, that the skills the students are learning are meaningful in the wider context of their lives.

Malcolm Pein, Chief Executive of Chess in Schools and Communities, added:

Demis delivered an engaging overview of A.I. and a vision of how technology may shape the world over the coming decades. His grounding in chess makes him a wonderful advocate for our work. We are grateful to him for giving his valuable time and for the support of DeepMind in our project to deliver chess to more schools in Camden and Islington.

About Chess in Schools and Communities

Chess in Schools and Communities (CSC) is a UK charity whose mission is to improve children’s educational outcomes and social development by introducing them to the game of chess. Founded in 2009, CSC now teaches in over 300 schools and supports 500 more nationwide. CSC also organises a world-class tournament, the London Chess Classic, and Yes2Chess, an international tournament for schools. For more information visit: Chessinschools.co.uk.

About DeepMind

DeepMind is a neuroscience-inspired AI company which develops general-purpose learning algorithms and uses them to help tackle some of the world’s most pressing challenges. Since its founding in London in 2010, DeepMind has published over 100 peer-reviewed papers, three of them in the scientific journal Nature – an unprecedented achievement for a computer science lab. DeepMind’s groundbreaking work includes the development of deep reinforcement learning, combining the domains of deep learning and reinforcement learning. This technique underpinned AlphaGo, a computer program that defeated Go world champion Lee Sedol in 2016 — a breakthrough experts proclaimed to have arrived a decade ahead of its time.

In 2014, DeepMind was acquired by Google, in their largest ever European acquisition, and is now part of the Alphabet group.


Reports about chess: tournaments, championships, portraits, interviews, World Championships, product launches and more.

Discuss

Rules for reader comments

 
 

Not registered yet? Register

e-mars e-mars 11/16/2017 07:14
@theflyingllama True, that's what - more or less - they did with Go. But Go has two types of "pieces" only (black and white stones) and one type only of move. The problem with Go was the size of the variation tree, just because the board is huge and one move might completely change its configuration. Chess, despite of having a smaller tree (compared to that of Go), has a much more complex set of rules: different pieces with different moves, exceptions (en-passant, castling, promotions, ...). This is why at the moment of writing AI hasn't been of any help for chess. I don't think our generation will see a full-AI playing chess.

@truthadjustr There's already been an experiment in that direction using e.g. TD learning algorithm (see https://arxiv.org/abs/cs/9901002). At a certain point in the past every chess engine we know tried to be "smarter" with some sort of learning algorithm (even only applied to opening books). These attempts miserably failed: the results weren't worth the effort. So they need to drastically change the approach, and no one is sure whether this is applicable for chess.
Current algorithms are good enough to build an invincible (against humans) chess engine so I don't think there will be any investment in chess AI, not now.
Even building custom chess-oriented hardware eventually failed (Deep Blue and Hydra dismissed).
fons fons 11/16/2017 04:13
@ rubinsteinak: "is there something inherent in this heuristic approach that makes it inefficient or problematic for chess"

Based on my limited understanding I think if alpha go did chess the result would be a lot less spectacular than it was for go.
Chess is very tactical so brute force goes a very long way, as the top programs of today show. These neural networks are better suited to figure out general ideas, long term plans etc and that doesn't work so well in chess because there's always some tactic that ruins everything.
Then again in go there's also a lot of calculation so I'm not sure about this.

@ Mr TambourineMan

That doesn't work or people would have already done it. Well of course there were go playing programs before alpha go but they were not better than the best human go players. I don't know how much they differ from chess programs but as far as I know brute force does not work very well in go, the variation tree is just too big.
kriegmaus kriegmaus 11/16/2017 05:31
https://en.wikipedia.org/wiki/Talk:Demis_Hassabis

Accuracy of #2 U14 Claim[edit]
The claim that Hassabis was #2 U14 seems questionable. Refer to the relevant FIDE rating lists on olimpbase: [1], [2]. As well, see his rating profile on olimpbase: [3]. He was among the leading U14 players in 1990, no question, but not #2; more like #5 in Jan 1990, when excluding Borislav Benev (whose DOB came to light in time for the Jul 1990 list). A claim that he was #2 U14 in England at that time, behind Jonathan Parker, would be more accurate. Also note that he seems to have no presence on FIDE rating lists pre-dating 1990, while Judit Polgar jumped from 2365 to 2555 between the Jul 1988 and Jan 1989 lists. Hassabis would have been U12 when she was 2365.

65.92.40.237 (talk) 04:41, 18 February 2016 (UTC)

Jump up ^ http://www.olimpbase.org/Elo/Elo199001e.html
Jump up ^ http://www.olimpbase.org/Elo/Elo199007e.html
Jump up ^ http://www.olimpbase.org/Elo/player/Hassabis,%20Demis.html
vanhelsing vanhelsing 11/16/2017 02:04
Very nice article. We need more like this.
Mr TambourineMan Mr TambourineMan 11/15/2017 10:48
People here suggest AI chess engines. Forget it. Instead the Chessbase Company should do this:

Convert the Chessbase and Fritz programs to work even for the Game of Go.
Should not be that difficult. But you must do it now, because soon it's get too late.
truthadjustr truthadjustr 11/15/2017 09:03
The proof of AI in a chess algorithm is for the chess software to play from the start at a level of 1500 playing strength. With no external reconfiguration and setting tweaking, it must gradually and gradually increase its playing strength using its past experience data alone.
rubinsteinak rubinsteinak 11/15/2017 06:38
@theflyingllama I totally agree. A chess teaching program, which used actual words and concepts, not just variations and +/- pawn-value evaluations, would not only be really cool, it would make the company who produces such a tool a lot of money.
rubinsteinak rubinsteinak 11/15/2017 06:18
Back when AlphaGo beat Lee Sedol, there was an article or two on Chessbase, and in the comments I asked a question that was never answered, so I'll ask it again: Would the programming (i.e., MC tree search method) used to create AlphaGo be applicable to chess? I have never seen Hassabis address this question. Given his background in chess, I'm sure he considered or has even created an "AlphaChess" engine, but is there something inherent in this heuristic approach that makes it inefficient or problematic for chess? Seemingly, if it could play chess very strongly while not using brute force, that would be quite an achievement, and a breakthrough in chess AI. Yet...we never hear any discussion around this topic from DeepMind.
theflyingllama theflyingllama 11/15/2017 02:20
@e-mars, actually AI would be great for chess since it would allow for true pattern recognition which is not done by current software (minimax algorithm). We can imagine that the AI will analyse billions of played and virtual chess games to extract empirically derived scoring functions, and imitate player's style, similar to work done on using deep learning to imitate painters styles. The software will also be able to teach us chess concepts - derived from actual data - in plain English (such as the importance of the bishop pair). Only today, can chess software be considered "stupid fast machines".
e-mars e-mars 11/15/2017 01:10
@vishyvishy Not sure what you are referring to but as far as I know Deep Mind are not interested in chess engines and AI has nothing (or just little) to do with chess: chess engines are just a bunch of very efficient, fast algorithms to explore the graph of variations. That might sound reductive but in essence chess engines are just freaking fast graph algorithms with nothing to share with AI. They're "stupid" fast machines.
In fact, it is not clear the role of Fritz in the article shown during the simul, apart from marketing purposes.
In fact AI would be an unnecessary burden to chess.
vishyvishy vishyvishy 11/15/2017 11:47
I wish to See AI in Active action in Chess ... When will deep mind play chess with todays players and enginers? Anybodys knows anything about that??
1