Bot used in 19x19 Hex championship

It's a shame, but hexanna account is using KataGo bot to make the move choice.

I was suspicious towards the initial play vs me of that account, but then I noticed the games vs Daniel Sepczuk both finished therefore I could look into them.

I had to install the 19x19 version, as I only had the 13x13 one, but then I followed through the moves of both games between hexanna and Daniel Sepczuk. Majority of the hexanna moves are exact top suggestions by KataGo.

In the game that Daniel won at some moment the moves diverged from the bot suggestions and the evaluation turned into Daniels favour. I don't know if it was a strategy to avoid being discovered or what.

I'm fine with playing vs a bot, but please just make the explicit claim that the bot is playing. Especially don't give unequal chances to different players by winning to ones and losing to other.

I will continue making my moves vs hexanna without any help.

I knew the accusations would come at some point :)

My accuracy was about 50-75% in the first 20-30 moves in the two games against Daniel, according to KataHex. That is close to my average when I practice against the bot on my computer.

One of my "best" games where my accuracy was in the 90s was game 2366711 vs HappyHippo, though most of my moves were forced in that game. 7. m5 was part of my prep when I played that opening. I've also analyzed games of other players who've certainly studied with KataHex and they are also able to get winning middlegame positions against top players.

I have been studying a long time with KataHex (6-10 hours on weekends, 2-3 hours on weekdays, several hundred hours total). Eventually one develops a good intuition of bot moves. When I play against KataHex, I can open a4 or a6 and keep my advantage for 20-30 moves 30% of the time. Human players (including myself) don't play nearly as perfectly as KataHex, and humans tend to play corner moves instead of more challenging moves in the opponent's 4th to 6th rows.

It's relatively easier to play near-perfectly in the 19x19 opening (first 20-40 moves) because tactics don't matter. In any given opening position, there are usually at least 10 moves that are within 2-3% win rate of KataHex's top choice. You just have to have good intuition, and plan ahead - play moves that leave you more "obvious" options for the next few moves. I've attempted some articles on HexWiki, but it's hard to write down all the intuitive patterns one picks up with direct study.

Openings on 19x19

Strategic advice from KataHex

Let's say on every move there is on average 20 reasonable choices (sometimes it's 3, in opening maybe even 50). To play 20 moves in a row the top engine choice means that you "learned" 20^20 possible scenarios, a 26 digit number.

You can make your claims, we cannot prove each other right or wrong, and surely you can't fool me. :)

Did i play against human or bot?

Especially in games against me, hexanna seems too strong for a human. In games with other people, I see mistakes made, and it's obvious. Maybe I'm wrong, doesn't matter.

This isn't the first time I have seen hexanna suspected of cheating. But I still don't really see the evidence.

I have let KataGo/Hex analyse the 19x19 championship game which hexanna won, and the numbers don't look suspicious to me. Link to analysis results. I see a high match rate in the opening, but, for example, it's quite easy to learn that Kata likes k5+L6. The style seems simplistic, with lots of bridges, but I don't know if that indicates cheating.

Anyway, I hope legit players stay, and cheaters go away or stop cheating.

(I have failed to give access to the analysis at first, but the link should work now.)

In game 2392428, if play human not bot, i have no words. perfect play.

Actually I confused myself in that game, planned to play O8 in this position which would make some threats, but for a moment I forgot ;)

Anyway it was already white win, very precise endgame play by hexanna.

At the time of writing we are at move 76. You shouldn't say perfect play or white win, I wasn't 100% sure and telling me that gives me an advantage. It doesn't matter though (and some of you think I'm using a bot anyway) so I might as well talk about the game.

I felt good in the first 36 moves since it was all familiar patterns. Moves 11 and 33 seem like obvious mistakes (why allow 34. q3?). Move 24 is the best way to bridge from k5 when c2 is present, it seemed good even with c6 played. I was considering playing the standard 26. c5 earlier but saved it for when I had fewer options and my edge was intruded into. Move 30 felt correct, it's the 6-6 4-4 obtuse corner equivalent of 5-5 3-3 which is covered in example F in the 19x19 opening guide, 5-5 obtuse corner.

I felt pretty lost after that but tried to play good local moves. 40 is the same idea as move 4 in this joseki which I've played hundreds of times against KataHex. 48 leaves the option to play either m2 or n2 later, so I don't commit yet. 55 by black seems like a good move close to my edge, similar to this. 56. j12 was obvious to prevent black j12 followed by a sequence like l11 k13, which feels very strong for black. 58. j12 was probably my hardest move to play, but nothing else was appealing; m12 looked weak with my other stones, and I didn't want to play too close to my edge because I already had q2/q3/o3.

I don't understand 59 allowing the efficient bridge on move 60. I also don't understand the purpose of 65; it's a bad idea to play two adjacent stones like that even when trying to minimax. 66 captures g11/g12 and connects bottom so it seemed obvious. I thought it was too obvious and looked for a trap but didn't see any. I played 70 hastily without thinking much because it looked good with 72 (the black cluster of stones after 73 looks pretty inefficient), but I realized a couple minutes after that k7 was also a possibility.

I look at games at the top of the ladders and this 19x19 tournament and see a ton of obvious errors (like 1 every 5-10 moves), many of which aren't exploited by the other player. I'm much weaker at tactics or general reading than top players though. I'm having trouble with some games with 2200 rated players feel easier because all the joseki and patterns are familiar to me.

Anyway, I have lost some interest in hex recently, in part from these accusations, and I might resign some games.

I'm having trouble with some games with 2200 rated players feel easier because all the joseki and patterns are familiar to me.

Oops, that should read "I'm having trouble with some games with rated players where I'm sure I have the advantage because I've played more efficient moves so far, but it's really tactical and one mistake would cost me the game. Funny enough, some games with 2200 rated players feel easier because all the joseki and patterns are familiar to me."

The thing about using a bot is that is impossible to prove either way, not even in OTB tournaments, let alone online.

Typically players establish themselves on top level for a long time, in hex for years because the intensity of play is so low. In the last 20 years many players came close to the high level, like shalev, urmaul, simpledeep, but each of them later gave up trying. I'm happy for people to give hex a shot, but there wasn't so far a case that someone arrived and immediately played stronger than people who studied the game for 20+ years.

Some say is different now that one can study with the engine. But is it? I don't see in chess community people suddenly playing super strong after studying with stockfish. As a matter of fact, chess teachers discourage training with stockfish (I disagree to this though). So, maybe indeed hex is such a different game to chess that engine training is more applicable, maybe also the general level of hex is so low that it doesn't take much to defeat everybody. There are many maybes but surely is always suspicious if a beginner comes and wins against everybody, with many moves fitting engine suggestions.

Therefore, if I got it wrong, I'm sorry. If my intuition is wrong, that you opened the engine within the duration of the tournament, to look at engine patterns, I am sorry to have wrongly accused you.

Hexanna started playing a few years ago already and seems to have studied the game very seriously. You can tell even from the initial posts of Hexanna in this forum that they have been working hard to understand the opening. Hexanna has written an opening guide for Hex 19x19. This guide looks to be extensive -- even if others might have helped (it being a page on the hex wiki), it still presumes a solid basis of understanding to make that kind of contribution.

I don't have proof one way or another if people are being honest or not, but I hope Hexanna doesn't get discouraged by some few people who were shocked that someone could become a top player after playing more than 1000 games.

2392428 can someone produce similar analysis (ranking moves by katahex preference)?

I was surprised that in the analyzed game that Daniel lost, his moves had a better match overall with the bot.

Anyway, I have lost some interest in hex recently, in part from these accusations, and I might resign some games.

This is tragic. Losing interest just because Arek can't accept he has been outplayed by newcomer... :D

Anyway btw katahex is so strong that it casts doubts on the whole game. Personally I always had doubts about hex (it always seemed that there are too few balanced positions) but recently kata really confirmed all these doubts.

So maybe we should all stop playing... or maybe we find a way to prove kata wrong :)

I think I have already proved it wrong in one opening and I'm working on another now...

KataHex may be able to destroy me in Hex, but it can never have fun playing it like I do.

And if it ever does learn to have fun playing Hex, then we're having fun together, me and my new synthetic friend.

I can't lose.