Scrabble GM vs. AI -- the Rematch! Game #23
ฝัง
- เผยแพร่เมื่อ 22 ก.ย. 2024
- The Scrabble AI BestBot got the best of me in my 100-game Human vs. AI Ultimate Scrabble Battle, but I'm not ready to cede to our AI overlords! Introducing... the GM vs. AI rematch!
This 100-game series, running every Monday and Wednesday at 5pm ET for 50 weeks, will feature 20-minute games against BestBot with post-game analysis. Hope you guys enjoy, and wish me luck!
BestBot is the upcoming ultimate Scrabble AI from Woogles.io, to be launched in 2024. For questions, please email woogles@woogles.io.
Want personalized help taking your game to the next level or a fun gift for a friend? Check out www.mackmeller...! for more info or email me at mackmeller@gmail.com!
The bots are just really much less afraid of the worst case scenarios than humans it seems. I think it's been a pretty consistent pattern that if they can create an opening that they can guaranteed hit, and the opponent may or may not be able to hit, they will push their luck and just gamble that the opponent doesn't have it, as was the case here with HANGS. It's really fascinating to me, obviously I trust the expertise of national champions, but BestBot is clearly doing a lot right - and this setup-heavy style is also famously favored by Nigel, so who the hell knows.
It's just really cool to me that in spite of all of our collective human AND machine-assisted wisdom, we just don't have definitive answers in spots like these. Scrabble is awesome
It is the strength of a computer program, as it does not take emotions into play. A human is willing to sacrifice value for peace of mind, and just defaulting to the fundamental principle of "play defensively when you have a lead". You kinda have to do this, as if you try to analyze every single situation in great detail, you will run out of time. To save time, you have to assign blanket rules to similar looking situations. You have to make shortcuts. Maybe if you had 25 hours to play a full game instead of 25 minutes, top players would be much more aggressive, but since they get relatively low clocks, it makes sense that more defensive play is common.
From a human perspective, Mack's evaluation of that setup is right, as probably the only human in the world who would exploit him in that spot is Nigel Richards, but from a theoretical optimal strategy, it's more debatable. In a real game situation I would not play like Bestbot did, because it just wouldn't come to me in the moment. After the fact though, I fully agree with what Bestbot did, as hindsight makes everyone a genius.
Yep well said! We may never really know :)
Tough game, but you are always fun to watch. I love your commentary while playing as well as your after game analysis. Good luck in the next game!
for real, although could you please put a spoiler flag next time? Your comment appeared on the bottom of my screen, and I didn’t watch the video yet man… 😅
Thanks so much, I appreciate it!
I'm getting so frustrated watching your luck in these games, so I can only imagine how you are feeling with the bot drawing so well. Thanks for playing on, Mack!
Well played Mack. Great videos. Keep to em coming.
You're obviously playing very well and are very unlucky with the draws, but I wonder if talking over the game, having to explain the moves, takes away some playing strength.
I'm strongly feeling that effect myself and have actually stopped my video series altogether. I found it highly stressful to record the games and talk while playing, and I'm not stressed while playing without that added difficulty. I recently tested this, recorded five games with the camera on an with commentary, and lost 0-5. I then played five games without recording and won 4-1. Obviously nowhere near a definitive conclusion, but my record against BestBot has generally been much better off camera, not just in those games.
So I really wonder how you would do in a best-of-100 series that you don't record. I like to sit down in the evening and play a game on my physical board against the engine. Not just very enjoyable, much more so than with the camera on, but also a great way of training. So far I'm up 11-10 in those games...
Also, I don't know if you saw his Facebook post, but Kenji Matsumoto is playing such a best-of-100 series right now. Last time I checked I think it was 5-7 or something. Will be interesting to see how well he does.
To add to this, there's a classic study in psychology showing that verbalizing judgments can override intuitive judgments and lead to worse decisions ("Thinking too much: introspection can reduce the quality of preferences and decisions"). Explaining out loud why you prefer one option leads to less accurate judgments than not doing so. It's admittedly a stretch to connect that study to Scrabble games (but there's an idea for a cool Scrabble-specific study!), but it _could_ be the case that talking over your moves overrides some of your intuition in a detrimental way - since high-level Scrabble is a _very_ intutive thing
It might make the commentary easier as well (not that there’s anything wrong with current commentary). You might record a quick "game recap" to talk through your thought process. That way, when you're playing, you only have to think about playing; when you're commentating, you only have to think about commentating.
@@Roezaea It would make for much less engaging content though 😃
@@AlexDingsIf I recall correctly, that is the case for non experts. But for experts, it wasn’t found to affect their judgement.
This is something Hikaru nakamura has talked about before, how when he first started streaming chess he played much worse. Maybe it gets easier the more you do it
Definitions of Interesting Words Played in Game 23:
GOMERALS (90 pts) [noun] - plural of "gomeral"; (dialectal, chiefly British) simpleton, fool [of unknown origin]
PANDIES (81 pts) [noun] - plural of "pandy"; (dialectal, British) a blow on the hand usually with a cane or stick [perhaps from Latin]
JUBE (58 pts) [noun] - a screen separating the chancel of a church from the nave and often surmounted by a cross or crucifix [French, from Latin]
YAUPS (36 pts) [intransitive verb] - third-person singular simple present tense of "yaup", variant spelling of "yawp"; to make a raucous noise, bawl, squawk
KUFI (41 pts) [noun] - a close-fitting brimless cylindrical or round hat [probably from Arabic, perhaps via Swahili]
AZONS (28 pts) [noun] - plural of "azon"; an aerial bomb that can be guided to the left or right by radio control [alteration of another English word]
CORVEES (94 pts) [noun] - plural of "corvee", variant spelling of "corvée"; unpaid labor (as on roads) for a day or longer period due from a vassal to his lord [from a Latin-derived French word]
KNAWEL (26 pts) [noun] - a low spreading Old World annual weed (Scleranthus annuus) with inconspicuous greenish flowers that is now widely distributed in North America [from German]
Gomerals and pandies!
On the AZON(S) turn, I kind of liked C2 OVA, setting up the N hook with only one N unseen, and giving an opportunity to draw the U to play J11 (L)UTZ to block the bottom if the bot didn't hit that turn.
That's a cool idea! Could even hit something like DONZEL for 80+ a turn or two down the road. I might've been a bit too impatient to shed the Z here, I don't deny whatsoever that AZONS is far from an inspiring play haha
I’d like to imagine that Mack’s gotten so used to commentating his games, he was doing it in his head during the SPC.
Hahaha maybe not quite *that* much, but it's definitely gotten more natural over the last year or so :)
HM was absolutely the correct play. There were 9 E's and 6 A's unseen and a draw of AE would give you a bingo that could only be blocked by playing _parallel_ to TRANSFER--VINTAGE(S) or AVE(R)TING. AVERTING would also be possible with a draw of AR as there was an E floater; ES would've given you VESTING(S), and that's just eights that played on existing floaters; no way of knowing what lines for sevens the bot might have allowed you.
Just got tile bagged.
I think the 2 tile fish by the bot signaling a strong bingo rack made it more likely to bingo thru the only open S hook, which interfered with his two backups. If the bingo percentage on the leave where you say is good though I can see that not mattering
Yeah I have no regrets about HM, and great point about all the As and Es, didn't even look at that part exactly. Drawing WQ will ruin any good play though! :)
29:36 you keep forgetting that BestBot doesn't do any inference! Kind of scary to think about because how good will it be with strong inference?! Because I'd agree that playing AI in that spot is a mistake
Yeah haha I just keep assuming since it's so dang good it must be using some kind of inference, even though of course I know somewhere in the back of my mind that it doesn't 😂
Would you try the game mode that the french people usually play? It is called "duplicaté", I believe.
I'd be happy to! I don't think Woogles offers it though, but maybe they'll add it sometime
You have mentioned drilling on words ending INGS and ILY in preparation for the SPC. Any other patterns well suited for study?
Does BestBot know how to abuse the random number generator to get better draws?