I wouldn't have expected that. Is it just a relative lack of interest in building an AI which can dominate Scrabble?
It's a partial-information game, but the search space can't be as big as Go, and an AI has an advantage over human players in that the entire valid string family can be encoded into a trie or some other efficient data structure, it's never going to forget a word or think it can play one that isn't valid.
My intuition is that AI should be able to crush the best human players at this point in time, but I'm open to being corrected on that if there's some aspect of the game which I'm not modeling correctly.
Having written what I believe is the best bot out there, since that article (BestBot on https://woogles.io), it beats many top players around 55% of the time roughly, but still has so many fundamental issues that we have not solved yet. I believe if we matched it against the top player or two in the world that it would be roughly even. I would not put money on it beating Nigel Richards over a 100-game series. We have a while to go until we build a truly superhuman bot; this is not the case for Chess/Go.
As to what you're possibly missing here - it's that Scrabble is more than just about finding the top-scoring word. There's a lot of board shape considerations, inferences, volatility, etc that current engines, including mine, don't yet take into account.
That strikes me as odd too -- but it might be because searching a dictionary is such an obvious computer advantage that it's not interesting to optimize. There are only 10 articles on arxiv.org that mention Scrabble vs 100s on Chess
Just finding the highest scoring word won’t make you all that good. If you played the highest scoring play available to you each turn you wouldn’t be that strong of a player. Maybe around top 200-500 or so in the US I would guess? And it’d be a super exploitable strategy by a decent player.
The reason is that you need to apply some rules, like when to trade vs. making a play, balancing consonants and vowels for future plays, what parts of the board are too dangerous to make certain plays in, etc. It’s because of the distribution of unplayed tiles, the high-scoring spots on the board, and the 50-point bonus for using all of your tiles.
Because of that, generally, you’ll do better by building towards a 50-point bonus play every 3 or 4 turns than by maxing your score on each turn.
I’d be curious about letting a human player play with the assistance of the best bot available and seeing how much better that would make them. I guess part of the issue though is that in a 13 play game maybe 3 plays are meaningfully difficult. So it’d take awhile to see if the human is improving on the bot or not.
It's a partial-information game, but the search space can't be as big as Go, and an AI has an advantage over human players in that the entire valid string family can be encoded into a trie or some other efficient data structure, it's never going to forget a word or think it can play one that isn't valid.
My intuition is that AI should be able to crush the best human players at this point in time, but I'm open to being corrected on that if there's some aspect of the game which I'm not modeling correctly.