IBM has a long history in AI — machine learning (“ML”), deep learning (“DL”), neural networks (‘neural nets’), and open-source large language models (“LLMs”); I just used the word “history”, but don’t worry, this article isn’t a history lesson.
AI is made by humans; it doesn’t occur naturally, which makes it ‘artificial’ (that’s its definition). AI can ‘acquire and apply knowledge and skills’ — thus exhibits ‘intelligence’ (by definition). And as is the case (currently) with all AI, humans provide — or provide the means to acquire, that knowledge.
Most IBM-ers know that in the ’90s IBM built AI (called Deep Blue) that beat the-then world chess champion Garry Kasparov; but few IBM-ers know more than that, so I dug a little deep-er…
My name’s Alex: I’m in Customer Success with IBM Canada — opinions are my own! If you’ve played chess (even if you aren’t any good at it) or you’re interested in AI (or both) then I think you’ll learn something in the next 4 minutes.
In 1996, a team of IBM computer scientists, and chess grand masters, built a computer (Algorithmic Intelligence) pre-programmed with 1,000s of games capable of calculating 100 million moves per second, and lost to Kasparov 4–2.
Deep Blue had a bug (it’s software so it’s no surprise) and in one game randomly selected a valid-but-poor choice of move. At world-class level (whether you’re human or computer) with all else being equal, one poor choice is all it takes to slowly-but-surely lose the game from that point onwards.
In 1997, re-programmed to counter Kasparov’s attacking style, calculating 200 million moves per second, Deep Blue beat Kasparov 3 games to 2, with 1 draw.
IBM were allowed to make changes to Deep Blue between games (just as a human could) but Kasparov suggested human players intervened during one of the games and demanded a re-match — the machine was being dismantled and IBM declined.
On aggregate, Kasparov won more games than Deep Blue (6½ to 5½) which he did with 20 watts of brain power, compared to 1,400w of machine power.
Kasparov wasn’t thinking of millions, thousands, or hundreds of moves per second. It might be fairer to say he was feeling his way through the games because there was psychology at play too — which reminds me of a quote by neuroscientist António Damásio who said: “We’re not thinking machines that feel, we’re feeling machines that think”
Kasparov didn’t know his attacking style wasn’t difficult for Deep Blue to play against. Deep Blue’s human-designed algorithm looked 6-to-8 moves ahead (sometimes 20) and literally crunched the numbers. It calculated the best move within a fraction of the time allotted for a game under tournament conditions.
Deep Blue’s chess rating is estimated at 2,853. Kasparov’s peak rating was 2,850. The peak rating of Magnus Carlsen, the highest rated (human) player ever, was 2,882 and his fans believe he’d have been able to beat Deep Blue with his ‘creative’ openings — moves he’s used to beat other world-class players; opening moves which are rarely made, so rarely studied — because they’re sub-optimal moves. However, Stockfish, the highest rated (chess) engine today, is 3,642 so I doubt his fans believe he’d be able to beat that!
My chess engine rating (my brain) is about 1,550, and over the last few weeks I’ve been watching Grand Master Simon Williams on chess.com. He’s been teaching / I’ve been learning: the London System (from white’s perspective) and King’s Indian Defence (from black’s perspective). He believes chess engines underestimate positional advantage, so last month after I beat Air Canada’s in-flight entertainment chess computer on its ‘hardest’ setting (it only has three settings), I opened the chess.com app (which uses the Komodo engine) and played ‘Wally’, rated at 1,800, and lost several games in a row (losing a minor piece and being unable to recover) before finally managing to win as white — with The London System, and win as black — with King’s Indian Defence. I’m inclined to agree with Simon: in both games I sacrificed a piece to gain a positional advantage and went on from there to win. Had Kasparov known this kind of anti-engine strategy, he might have won against Deep Blue in ’97 too.
An Algorithmic chess engine rating of 3,000+ doesn’t mean AI’s better than humans. It hasn’t stopped humans playing chess and it hasn’t displaced human coaches in schools and clubs. It means we’ve been able to write a software algorithm that performs better at this task than we do (in this case playing chess) — albeit using two orders of magnitude more power than we do (can do) for any task.
The human brain understands complex visual images more efficiently than AI, but AI is more effective (faster, more accurate, and more reliable) at tasks like identifying cancer; and Quantum Computers (when we overcome error rates associated with scaling) will be more efficient at tasks like protein folding than ‘classical’ computers like Deep Blue.
It may sound like I’m playing-down AI, and I may have made Algorithmic Intelligence (with my example of chess) sound mundane, but number-crunching is an incredibly powerful, incredibly useful tool, for solving certain problems. And if AI meets the needs of the many (not the wants of the few) I don’t see anything wrong with building artificial systems that perform tasks more effectively (and more efficiently with respect to time) than we do with our naturally limited capacity.
Oh, and a final word of caution: there are more positions in chess than atoms in the observable universe, and no (classical) computer can memorize every position or theorize every position through to the end of a game; so, if you’re thinking about building AI to play chess, perhaps start with Noughts and Crosses or Connect Four!
Source: www.meduim.com
