AI Enabled By Default Hurts Teaching

[BenKyo_Baduk]

Right, meaning that an option that the reviewer controls is a good idea right? ie. disable AI for everyone in the review (at least at that time, or at least switches AI off for everyone even if we give them the ability to still manually turn it on after that). Are we thinking the same thing? I feel like we are

[fuseki3]

I agree, it would be better to have it turned off by default. (or a syncable option, to be turned off for everyone by the reviewer)

From personal experience also in teaching reviews, collaborative reviews (with other players or studying together) and also with pros, I agree with all of this being problematic :

@BenKyo_Baduk

• Students frequently mistake the AI indicators for what I’m trying to say and show, and get confused by the lack of cohesion between the two, meanwhile I’m unaware.

@BenKyo_Baduk

• Students get overwhelmed by the AI stuff on the board, especially in combination with what I’m trying to show and my own board mark-ups with the (awesome) pen tool.

This is definitely the case – when reviewing, things like tsumego, sequences of any sort, deep reading or yose, it’s complex to solve a tsumego or see possibilities and the AI can display something like a dense 5-6 options in one area, which make it difficult to read and clutter the board even to me, even at ~5-6d, with excellent reading and the ability to quickly follow sequences laid out.

The indicators also appear over the AI numbers, or the letters “A” “B” etc, are smaller and difficult to find in the web of numbers – in general, it seems to clutter the board a lot and make it visually confusing for a lot of kyu players, to have all of the extra info whilst reviewing.

@BenKyo_Baduk

• Students get influenced and/or spoiled by the AI, ruining quiz questions and other teaching prompts.

This happens often too, when jumpiing into a demonstration with a tsumego or whole board problem, unfortunate for interesting positions I or others want to think about

@BenKyo_Baduk

Learning is incremental, teaching Go is largely about concepts and ways of thinking, and different students need different tailored suggestions. Students will come up with lots of moves that are wrong to AI, but they will improve their reading and shape sense and stuff over time, so what’s important sometimes is how they’re thinking and what they’re trying to accomplish on the board, not whether their move is AI-optimal. Lots of AI moves are optimal only due to specific variations and details, what’s the use of that when the student would never be able to read like that in their game and doesn’t even have underlying Go concepts down yet.

It’s also true – and even when I review with strong pros, their suggestions can be to play X simpler move, or that from experience even top pros will make a mistake 80% of the time trying to play X moves like the AI.

It can make a big difference to play and learn from what one can understand or follow-up, as some AI moves may rely on superhuman ability, for example, to read accurately 200+ different 30-move deep sequences in a moyo when playing in a riskier way to live there – and in practice, even top human pros can often die where the AI will tend to find a way to live.

If reviewing with a 20k, also, it may not be the most useful thing to point out there was a 20-move long sequence with 25 branches, which means they could have gained a few points with a move even I or a pro may have difficulty reading.

It’s possible to say things like “oh, the AI sees some sort of aji there” but it can also be extra clutter to have the indicators+sequences and complex stuff being shown, if we’re focussing, for example, on things we could consider playing or basic good shapes.

@BenKyo_Baduk

Plus, often during teaching I have constructed only a local position instead of a whole-board position, so the AI will complain about unplayed corners and such. I’ll stop here… I feel like I could respond to this endlessly. I take it you’re not a teacher, that’s okay.

(this, too

– also, OGS’s AI is not always accurate – in discussions with strong pros, when we intentionally try it out, it can lead down rabbit holes in which it suggests various sequences which are dying 10 moves later, which we can both see doesn’t work (the analysis may initially be that X is the best move but later, it ends in the position after all of the optimal moves in the sequence are played to 8-10 moves deep, being judged by the AI’s “score” judgement at -10 points worse than at the beginning )

@Groin

Well if the choice made by the reviewer only has local temporary effect on the review users, I don’t see much inconvenience (and the users may switch back if they feel so, so it’s not a big deal for them to accept the default wish of the reviewer)

@BenKyo_Baduk

Right, meaning that an option that the reviewer controls is a good idea right? ie. disable AI for everyone in the review (at least at that time, or at least switches AI off for everyone even if we give them the ability to still manually turn it on after that). Are we thinking the same thing? I feel like we are

A question, though, sometimes we may want to check the AI during a review out of curiosity, either for the reviewer to quickly look at it by themselves without disruption & see the current AI idea of the score, or choose a sequence to lay out as the AI suggestion, or maybe everyone – if the case, will the AI be able to be turned to “on” to adapt correctly in this case ?

Would the option be only to “turn off AI analysis for everyone by default” (with the ability to turn it on asynchronously whenever), or “disable ability to turn it on” ? Or both options as a possibility ?

[vilgo]

Thanks for starting this thread, I was considering creating it and I’m glad to see that it is already being discussed. I have disabled it in order to self-review my games myself, and often activate it afterwards. Instead of the graph popping up, two buttons “review game manually” and “AI review” (or a button inside the review tool “show AI analysis”) would meet my needs.
Also, I feel like AI analysis of every game is a waste of resources (ok the community of OGS is not that big, but don’t we wish for it to grow ?).

[Shen0927]

@vilgo

I have disabled it in order to self-review my games myself, and often activate it afterwards

I’ve done the same and found this to be the best way to review. AI analysis hinders deep thinking and self-reflections, which is key to grow.

It’s like practicing Tsumegos that already show the next best moves on the board - It’s hard to actually think and calculate what works and what doesn’t.