Hacker News Reader

Comment by Jevon23 | Hacker News Reader

There would be no point in keeping the human players around. You may as well just watch the computers play each other (which some people already do, occasionally). We’ve long since passed the point where humans can improve on the moves suggested by computers. The optimal strategy would be to just play the computer’s recommended move every time.

View on HN

The reason we watch sporting events is to watch a fallible human attempt to control their body or mind enough to best other humans attempting to do the same.

No one will watch a competition between two self-driving identical Mazdas racing down the road.

>No one will watch a competition between two self-driving identical Mazdas racing down the road.

That depends entirely on how it is presented. Over 13 million people watched these inanimate marbles[1] compete in a variety of races.

[1] - https://www.youtube.com/watch?v=O2AbcCbyJ2U

Marble races are fantastic, but not as good as the hotwheels races (3d botmaker diecast racing or something like that, the announcers are pretty funny)

~13m + 1, dammit!

This is satire, right?

Well if they're actually identical this seems like a stupid race.

But if the software is different, sure, why not? The Robot soccer leagues had a league where the platform is literally identical, all teams buy some Nao robots from Aldebaran Robotics, they have a coloured indicator so you know which team is which - but the matches certainly aren't like watching a mirror, if one team has programmed a vastly improved dribbling strategy or the other team's ball tracking isn't very good, it's pretty obvious who wins that.

Humans can no longer usefully assist chess computers. So all a human is doing is attempting to obey simple movement instructions, which isn't even a sport.

I think lots of folks would happily watch a race between two different manufacture's self driving cars.

http://selfracingcars.com/

What kind of team work between a human and a machine would result in interesting chess? I hope it would reduce the need to study openings.

Something simple to begin with: let computers play the odd moves and humans the even moves, without any communication. This way, the computer would need to play in a way that would make sense to their team mate.

This is the format in a lot of tournaments where a pro is paired with an amateur. For example, in pairs snooker or pool, there are two ways to arrange pairs: Alternate innings and alternate strokes.

With alternate strokes, the better player will not only execute their shot, they will arrange to leave a shot that their teammate can easily make.

---

But all that being said, if such a game becomes popular, cheating will still be a thing as the team will always benefit from having the strongest possible human player.

There's good reason to watch strong chess computers play each other! They can teach you some interesting techniques at the higher level (or so I've been told, I'm a lowly 1200 :P)

This is not true. There are chess leagues that allow unlimited computer assistance (most notably the ICCF, which organizes the World Correspondence Chess Championship), and you don't get anywhere close to the very top level by just playing the computer's recommended move every time.

Computers beat humans, but human+computer is still stronger than a computer alone.

There's been a handful of tournaments, but in the last one I can find[0] from 2017 an engine alone already beat all teams. Engines have only gotten stronger since then, so you'd have to use particularly bad engines or big limitations for humans to have useful input.

0. https://en.wikipedia.org/wiki/Advanced_chess#Online_tourname...

I would have thought that, since I mentioned the ICCF as the most notable organization in computer-assisted chess, you would have looked into it before trying to contradict my claims. Evidently you did not.

The ICCF organizes matches, maintains rating lists, grants titles including grandmaster, and organizes a world championship for computer-assisted chess games. Check them out [1] and stop making claims about a style of chess that you are obviously not familiar with.

[1] https://www.iccf.com/

If you want to reference an instance of player+computer games or tournaments where players+engines beat the strongest engines link to that rather than ICCF's homepage (who do allow computer assistence). Nothing on their homepage remotely proves your claim, and a quick search doesn't show anywhere they claim so, which is to be expected since you are simply wrong.

All that really happens in modern ICCF championships after the opening is babysitting engines and taking their top move(s). The human part is mainly picking and running the engine(s).

> Nothing on their homepage remotely proves your claim, and a quick search doesn't show anywhere they claim so

Because it's obvious to the community the website serves. ICCF having banner on their homepage saying "human + computer > computer" would make about as much sense as USA Track & Field having a banner on their homepage saying "Did you know running on your feet is faster than running on your hands?"

> which is to be expected since you are simply wrong.

This is a shocking level of arrogance for someone whose knowledge of the computer-assisted chess scene was limited to a Wikipedia stub a few hours ago.

There is a whole field of "anti-computer strategy" in computer-assisted chess. Humans study the weaknesses of engines to figure out how to exploit them in their opponents' play (and avoid them in their own play). Here [1] is an article discussing the general idea. Here [2] is a post by an ICCF player discussing why human+computer beats computer and giving an example of one of his own games. Here [3] is a a traditional chess GM discussing what the human player brings to the board in a computer-assisted chess game and warning against blindly trusting the computer. Here [4] is an ICCF world champion talking about how the human player steers the engine by feeding it ideas and avoiding positions where engines are known to have problems. Here [5] is another ICCF world champion talking about how his approach is to come up with ideas himself and then see whether the computer refutes them, rather than looking at what the computer suggests in a position.

Is that enough for you, or are you going to stubbornly continue to argue from your position of ignorance?

[1] https://iccfworldzone.com/cc-anti-computer-and-centaur-chess...

[2] https://www.reddit.com/r/chess/comments/4av9fe/correspondenc...

[3] https://www.iccf.com/Message.aspx?message=225

[4] https://en.chessbase.com/post/better-than-an-engine-leonardo...

[5] https://www.iccf.com/message?message=833

The first, and only recent (though without references and poorly written) link specifically says:

>Anti-computer playing is now irrelevant against engines as they can beat any human by the tactical strenght of standard engines or by the deeper evaluations of new neural network ones.

The second is a reddit comment which only implies they can maybe improved on a low-depth old version (lichess') version of Stockfish. Particularly unconvincing as an engine could also make progress even if it thinks it cant as more moves happen and it can see further, and as they tried multiple things, and then went back effectively just having their engine explore more paths deeper.

The third one is from 2008, when yes you could improve on strong engines consistently.

The fourth is from 2016, just before the 2017 breakpoint and possibly the last time you could improve on them.

The fifth one is from 2015. ditto.

There is a reason that there aren't links of say a team winning against the TCEC winner or anything solid proving your point.

There was no "breakpoint" in 2017. Engines still don't understand conceptual insights that humans spot instantly: positions where it's impossible for either side to make progress even though one side has a large material advantage, fortresses, etc. AlphaZero was claimed to understand fortresses based on some (possibly cherry picked) games but they wouldn't give anyone unrestricted access to it and lc0 has not replicated that understanding. It's still easy to come up with positions that stump TCEC winners, where an experienced human will instantly understand that the position is a dead draw but the engine thinks one side has a large advantage. This kind of weakness is exactly what human players exploit.

In the meantime, I'm still waiting for solid proof that running on your feet is faster than running on your hands. Why isn't this information anywhere on USA Track & Field?????

>In the meantime, I'm still waiting for solid proof that running on your feet is faster than running on your hands.

World record for 20m running on hands is 4.78s[0]. You can find plenty of <3s 0-20m from foot races[1]. That's how easy it is to prove claims with references when the claims are correct.

0. https://eu.indeonline.com/story/news/2021/10/05/zion-clark-f...

1. https://speedendurance.com/2012/08/09/usain-bolt-20-meter-sp...

Mmmmm, sorry, that was before the 2022 breakpoint so I can't accept it as evidence.

I thought it was interesting thatcomputer vs computer chess is largely just ties. In order to make them more interesting tournaments usually start the computers with some pieces already moved to force them into unusual positions

TCEC, probably the most prominent long-form chess engine tournament, does exactly that. It starts them in a pre-played opening position, usually ~10 moves or so into the game. There are still a lot of ties, but fewer than if you let them play from neutral.

Each engine plays both sides of the position (white and black) so it's always a bit of a cool upset when an engine wins both the white and black side of the same starting position.

New chess variant: Cyber. Each player has a budget and can buy a limited number of moves from the engine throughout the game :D

Wouldn't that depend on if there is a time limit? It seems under time constraints, a human grandmaster aided by a computer would do better than the computer alone, since the human can also use heuristics.

AI uses heuristics just fine. You might have missed the decades of post-Deep Blue development. Check out DeepMind.

Hm, ok what if we introduce time limits and also only allow the computer to use as much energy as the human brain does?

The results might surprise you:

An new iPhone can run the latest chess engines at a level that should beat all but the absolute best players (https://www.hiarcs.com/iphone-chess.html).

Apple doesn't publish power stats for their A-series chips that I could find, and I couldn't find a direct comparison of power draw running a chess engine, but Anandtec's numbers running synthetic benchmarks indicate between 3-5 watts when under single thread load: https://www.anandtech.com/show/16983/the-apple-a15-soc-perfo....

Estimates for human brain energy consumption are vague, but generally accepted somewhere between 12-25 watts, with 20 being the most used average guess (https://www.quora.com/How-much-energy-does-a-human-brain-con..., https://www.scientificamerican.com/article/thinking-hard-cal..., https://www.munichre.com/topics-online/en/digitalisation/int...)

Given these numbers, not infeasible to assume that a computer restricted to the energy usage of the human brain would win every game when not playing Magnus?

I think stockfish with limited depth to compute every move in under .1 second would still be incredibly stronger than a human to the point where their involvement would produce worse moves.

You'd probably just optimize the computer for heuristics then, and would be faster than the human.

Reply on news.ycombinator.com