This discussion reminds me of an anecdote. Our club had a new player who didn’t seem to know some of the basics, so after my game I volunteered to take a look at his game with him. (This was tournament play, so the moves had been written down.) In the game, he kept moving pieces to squares where they could be immediately captured and of course lost material quickly and repeatedly. It turns out he was using a heuristic of “Well, I want to get to their king, so I should move my pieces close to their king.”
Yes, moving pieces “close” to the king is a higher-order principle with some relevance to the game, but it’s very subordinate to so many other considerations that it is just instantly losing if you promote it to the principle evaluation concept.
Which part of my post is incorrect? I assume the number of games it would take?
I noticed you left out my last sentence about a handful of orders of magnitude, which means it could take 100,000 times longer than my offhand estimate for the genius to start winning 100% of the time, which comes out to around 20 million games? This would take many centuries, possibly millennia of they did nothing but play chess 24/7.
It was mainly the number of games, but also you didn’t mention the genius studying and analysing away from the board.
In competitve chess, the players analyse together after the game. This helps them both spot not just errors, but also when the game turns in one player’s favour - plus finding better alternatives for next time.
(For amusement, I remember in one international tournament watching the post-mortem between two Grandmasters. One was from Iceland, the other Hungary - and their common language was English. )
This is a specific prediction for a mere 200 games, plus the claim that the genius is not only now unbeatable, but also wins every game from now on.
(I may never have beaten a Grandmaster, but I’ve drawn with a few. )
If you wanted to make the number of games needed to be between 200 and 20 million, I think you could have said that directly!
“Win 100% of the time” is almost certainly impossible. Chess is probably a drawn game, with perfect play, and while current players (human or computer) aren’t perfect, draws are still by far the most common outcome of high-level play. Plus, there’s a significant edge for white (who moves first). I’m pretty sure that the top human players, when playing white, can still manage a draw fairly often against even the top computers.
Agreed. This is an ELO Win Probability Calculator. A match pitting a genius at an ELO of 3500 against a FIDE GM at 2500 has a 0.000458255 chance of being a draw. At 200 games this would 9.1651%. Not insignificant.
The genius has a 99.95% chance of winning a single match. If I am doing the math correctly, the genius would have a 91.18% chance of winning all 200 games. A little worse as white.
I picked 3500 based on some searching claiming that is what Stockfish Level 8 is capable of (in practice it is restricted a bit to make that lower). It’s an arbitrary choice meant as an illustration. At 4000 ELO, it is close to 100% and at 3000 ELO, it is close to 0%.
The top engines today have estimated ratings 800 to 900 points higher than the top human chess players. This is a rough estimate at best since there isn’t a pool of games along a rating ladder to calibrate the engines’ strengths to FIDE-equivalent ratings, but it’s the best estimates available all the same. Using these numbers and after 100 games, the expected W-D-L score for the engine is 100-0-0. It shouldn’t drop any points. After 1000 games, an elite player should have scraped together a couple of draws, but no win.
However, draws could be made harder (not that they need to be) using the engines’ “contempt” setting, which dials how willing the engine is to accept a draw when the position is objectively equal despite knowing that it has negligible risk of losing by playing on (and hoping to win).
That’s based on outcome probability estimation. In actual practice, humans never play exhibitions against top engines anymore unless significant odds are employed. This definitely includes materials odds (say, two pawns or even a full knight removed from the engine’s side at the start), but also some or all of: time odds, human always has white, and/or no engine thinking on human’s turn. And then it’s just a question of whether the odds were set aggressively enough for the particular pairing. No halfway official one of these matches has happened in several years, and engines are only stronger now than then (by 150 pts or so in the most recent case I’m thinking of).
I’m not sure what you’re agreeing with in that quote, unless I am misunderstanding something. According to your own calculator, if today’s top engine (est. ELO 3634) played Magnus Carlsen (ELO 2847), the draw probability is 0.0058, so one out of every 172 games with white going first. This does not meet my definition of a top human being, while playing white, managing a draw “fairly often.”
You’re correct, my ‘agreed’ wasn’t clear (or wasn’t quite correct). My brain was getting ahead of myself.
What I meant was that the GM playing white could get one or more draws over 200 matches fairly often. Their draw rate for a single game is only 0.0046% – definitely not fairly often.
Whilst it is interesting to hear how computers will do against Grandmasters, the OP stated a genius human (and we play chess differently from computers.)
So I stand by my assertion that the genius would not win every game.
The OP does not stipulate “genius”-they say “300 I.Q.”
I stipulate that this ludicrously high I.Q.ed entity could move their pieces trans-dimensionally and, while still following the written rules of the game, win within four moves…if they don’t find the game incredibly boring and refuse to play at all.
I would like to learn that secret.
It is true that the rules of chess do not forbid the use of ‘trans-dimensional’ moves. Nevertheless one would hope that either the arbiter would step in or that our genius has morals…
It may amuse you to learn that I was asked to comment on a very early computer chess program. I thought it was an excellent effort, except that they didn’t test for legal moves. So one game went:
Even if we posit that a 300 IQ would result in psi powers, transdimensional moves would still be against the rules. The rules lay out the dimensions of the chessboard within with all of the moves must be made. Moving a piece to Dimension Upsilon Beta 9-6 using your super-mind is no different, legally, to moving a piece to the top of the refrigerator.
If you’re going to win a game quickly using psi powers, the better route would be to mind-control your opponent and force them to Fool’s Mate.
No. I think it’s clear to me that the genius would regularly lose until they had enough games played and studied under their belt. Under a strict reading of the rules in the OP, the 300iq person stands no chance.
To continue on the “no way, no how” line, it might be instructive to see how much top level chess players rely on pattern recognition, which the hypothetical genius won’t have any of. Here’s a clip (picked randomly from the many like it out there; it’s not necessarily the most interesting or impressive one) of a top GM streaming himself playing the “Puzzle Rush” mode on chess-dot-com, wherein you are given a fixed time (here, 5 minutes) to solve as many chess puzzles as you can, where they start easy and get harder as you continue. The goal is to find the tactical sequence of moves that leads to a decisive advantage or checkmate.
You’ll notice that he sees the principle ideas in the position if not the exact sequence of moves more or less instantly. Even as the puzzles get harder and he has to calculate some variations to find the right one, it’s clear how he relies on “bite-sized” pattern recognition to facilitate his calculations.
GMs have spent a significant fraction of their life training their pattern recognition, both deliberately through tactics exercises and intrinsically through playing and analyzing games. The newbie genius has none of this, and no realizable raw speed of thought can serve as a substitute. Like speaking a language or flying a fighter jet or identifying local flora/fauna – lots of people do these things, and you can understand how such things work in a general way, but that isn’t relevant for how the human brain actually achieves these tasks. The brain requires substantial training over time (and for many tasks, ideally when you’re young) to actually do them, even if we can conceptualize the tasks as “straightforward”.
)
)