AI Sensei Problems for Game Review

My newest go study obsession is to create problems in AI Sensei to reinforce my lessons with Clossius and BenKyo Baduk.  My lessons consist mostly of game reviews, so it was a no brainer to start creating problems for those games in AI Sensei based on my lesson content. My problems are based on the input that I get from my teachers. They are not just based on blue and green AI moves. Some of the problems I make are are simply reinforcement of good moves, so they are not problems in the usual sense.  They are memories.

I have found that the primary benefit of creating problems from lesson material has been that it forces me to rewatch my lessons.  This is something that I should be doing anyway, but I have not done it consistently enough in the past. It didn't really feel productive until I decided to try making problems.

The way I create my problems is to open the video of my lesson along with the AI Sensei review of the game under discussion. I create a note in Evernote to take notes about the lesson incase there are things I want to remember that don't lend themselves easily to problem creation. I also open a review of the game in OGS incase I want to save variations. AI Sensei is best for one move problems, so I find it beneficial to sometimes replay a variation from a lesson and drop it into OGS to look at it later.

The time I spend reviewing my lessons, and creating problems is credited to Lesson REPLAY. The time I spend solving AI Sensei problems is credited to the Go Activities subtask of Tsumego called AI-Sensei Problems. I spent a great deal of time on that subtask last week.

An hour of that time was spent sharing problems from one of my games with a few BenKyo League members.  I shared my screen with them, and we talked through the problems. Currently that is the only way to share problems from AI Sensei. There were 34 problems for that game alone, and we spent an hour going over them. That included going back through the problems that were missed the first time.

The amount of time spent creating custom problems is not trivial.  It requires me to watch the entire lesson, and it requires me to write unique prompt text, and answer text, for each problem.  Recently I spent two hours watching commentary about one game, and creating 20 problems from that lesson.  I estimate the time spent on each problem is actually between 3 and 6 minutes not counting the time of watching the lesson itself.  If you figure that I spent two hours watching one hour of video to create 20 problems the math comes down to 60 minutes divided by 20 problems. This equals 3 minutes per problem, which isn't bad, but it is an investment of time.

The value of that investment became clear to me during a lesson I had with Ben the day following that two hour investment of time.  The next day we had another lesson, and Ben referenced a situation from the game I had created the problems for. I knew exactly what he was talking about because I had spent two hours the night before reviewing the lesson about that game. We talked about it in abstract terms until he opened the game record and found what we were discussing. It turned out to be a hypothetical that we were both able to recall. It wasn't even part of the game. It was kind of a mind blowing experience for me to realize how valuable it had been to spend the time reviewing the lesson and making the problems.  I had even made a problem for the hypothetical that we were discussing.

To give you an idea of what AI Sensei problems look like I will offer three examples of problems.  I will show screen captures of the prompt text and the answer text for each problem. This is actually a series of problems for a kill that never happened.

Problem 1

Problem 2

Problem 3

Currently AI Sensei problems are presented to me using a Spaced Repetition Model.  There is value in that, but it also means that I have no control over which problems are presented to me, and when I will see them again.

1. I would like to have control over when I see problems again.  Part of this might involve having the ability to initiate a session involving all of the problems from a game from within the AI Sensei game review.
   
   
2. I'd like to be able to create problem sets based on a variation that would keep those problems together in sequence.
   
   
3. I'd like for the addition of problems to the Training List not be dependent upon the date at which the game was first loaded into AI Sensei.  I found out about this issue when I created problems for games In-Seong had reviewed perhaps five or more years ago, and those problems did not show up immediately in the Training List. I was so disappointed by this that it stopped me from creating problems for a while.

Last week I spent a great deal of time doing problems to see if I could force the In-seong reviews to the surface.  It worked.  There were three reviews of AYD games that eventually appeared.

Next I decided to try to solve all of my problems. I got myself to the point where I only had seven games left, and I hit "Start Training" to do them all.

Success.

It isn't that hard to stay current with my problems.  This morning I was presented with 25 problems. It took me nine minutes to do them. That amounts to about 1/3 of a minute per problem. Most of the time was spent reading the prompt text, and answer text, for each problem. New problems are presented to you in a short period of time.

Now for my usual Go Activities Report:

It was a very busy week for go activity largely due to my new obsession with solving AI Sensei problems.

Not a great deal of time was spent playing games.  There are only three of them, but I have more games from the previous week for my lesson with Clossius.  In addition to that I have a new rank to report.

I have a rating on the PlayGo server which translates to 8 kyu based on the levels of the go problems.