Could a current top player beat the same DeepBlue that beat Kasparov?

John Nunn did some reasonably objective analysis, comparing the rate of relatively basic errors from Karlsbad 1911 vs Biel 1993. He did a two-step process, first identifying things that looked like blunders with a computer, then analyzing them by hand to check that they were legitimate mistakes and not the sort of "missed a long sequence of unlikely-looking only-moves that would have lead to mate in 17" situations that computers sometimes label as "blunders".

His conclusion:
"I had no particular preconceptions about what the results of this search would be. Like most contemporary grandmasters, I was familiar with all the standard textbook examples from the early part of the century, but I had never before undertaken a systematic examination of a large number of old games. I was quite surprised by the results. To summarize, the old players were much worse than I expected. The blunders thrown up by Fritz were so awful that I looked at a considerable number of complete games 'by hand', wondering if the Fritz results really reflected the general standard of play. They did.

[...]

In order to be more specific about Karlsbad, take one player: Hugo Süchting (1874-1916). At Karlsbad he scored 11.5/13.5 or 'minus 2', as they say these days - a perfectly respectable score. Having played over all his games at Karlsbad I think that I can confidently state that his playing strength was not greater than Elo 2100 (BCF 187) - and that was on a good day and with a following wind."

(Süchting was probably in the top 20-30 players in the world at that point, which these days would make you 2700+)

I think the interesting point he makes is that we tend to overrate the strength (and creativity) of historic players because we mostly encounter them through games that have been selected for their quality, and don't play through all their games one-by-one to see the times that they (eg) hung one-move forks in the endgame or whatever.

For more extensive details, there's a writeup by John Watson with some long excerpts near the bottom of this page: theweekinchess.com/john-watson-reviews/historical-and-biographical-works-installment-3

That link provided no basis to change my position. There is one thing I bring up often, and that is before computers back in 1900s, we used people to try out different lines. So, instead of a computer engine cranking out lines and comparing them, people would play these lines. At best, it was the collective results that could be seen as superior over an individual today without an engine. Given the tools today though, we can see the mistakes of the past players and any unchartered territory is due to the computer showing that both past players and modern players have yet to master.

@RamblinDave
There were millions of past games.
How many did Nunn analyse? And by 'hand'?
Handful of games can't dictate terms. He should have analysed games by top level players like Capablanca, Lasker, Alehkine, Fischer, Tal, Petrosian, Keres, Spassky
Nunn was actually endgame analyst who has books on those thesis. He is known only for that not for stupid comparison of how good were past players. And that too is wrong after seeing the site.

Nunn's peak ELO was 2630 - what's yours that you understand chess so much better than he does?

And he analysed 800 games from a tournament that included the world numbers 2, 3, 5, 6, 7, 8, 9, and 10 of the time. Sorry that that's not enough from you, where's your equivalent weight of contrary evidence?

@RamblinDave
Lol! Just 800. Plus whose games were analysed. This much can't do.
No hate to Nunn, he's my favourite endgame tactician and theorist but analysing games for comparing new players versus old players is not his cup of tea. Tell names of players whose games were analysed and not stupid rankings. I know who were real top level players over history and mental knowledge and self dependence and thinking ability has been on decline since then over time. This clearly shows that new players are weaker than older players.

Engines have had a certain style for a while. Players have developped with their presence, some since their first exposure to chess. In absence of measures toward how covering is an engine of chess space, there is the possibility that chess has been restricted to engine style, even for humans.

chess space includes set of all legal positions. (not the squares on the board) (just making sure).

Human analysis includes intuition. So probably not needing as much data as a crude headless data analysis algorithm.
One should include Nunn's own experience behind that. Which is not very measurable, I am afraid.

But yes, there might be the problem still of the covering that whatever number of games have been included as representative of chess forever. I seem obsessed by the notion of covering, am I, it is because it screams to me, while people are generally obsessed with self referring and floating ratings, that could drift anywhere at the population level, we would have no clue. So I insist.

NNue innovation in SF should be pondered in that light. Even if, also self-referring (training by old SF), it was necessary to compete with outside newcomer in the engine community pool. it shows that SF had a invisible style before. a bias. that needed some patching. Of course depth race being the reflex, is would try to patch that way with some "future" peek ahead for past non-informative nodes, but still within its legal tree default search input depth, those nodes getting a further peeks were neglected before, hence the bias, not covering those branches is a bias on the complement (in style legal subset).

The Deep Blue was weaker than modern engines, but still it would be very hard to beat today and top GMs would most likely draw with the machine.

For those interested in chess necromancy, that is somewhat objective comparison of playing strength of dead chess players or dead chess computers I'm providing links to the important papers of Professor Regan about Intrinsic Performance Ratings:

cse.buffalo.edu/~regan/papers/pdf/ReHa11c.pdf
cse.buffalo.edu/~regan/papers/pdf/Reg12IPRs.pdf

Unfortunately, those weren't updated for the newer engines, probably because similar methodology is nowadays being used for cheat detection.

@pointlesswindows said in #38:
> The Deep Blue was weaker than modern engines, but still it would be very hard to beat today and top GMs would most likely draw with the machine.
Yeah cause same Deep Blue defeated Kasparov in his prime so it's really big thing.