With the keynotes out of the way, let's get to some sessions. First up, Tuesday - day 1. The morning was set for workshops. I didn't attend any of those but I did go to two afternoon sessions. One was alright but, at least for me, missed the mark and the other was really strong.

The first one was titled "AI Ready: Future-Forward CS Education." The talk was on how Gwinett County Public schools have been implementing AI tools and teaching AI.

The session left me a little flat. The talk as very much at an overview level. They talked about the fact that all schools offer some manner of CS and also gave some examples of AI related activities. Two that I recall are one, a Tic-Tac-Toe game that they cover in the earlier grades (K-5) and another is a discussion of the idea that in the NBA AI is used to determine player risk in contract negotiations.

The problem is, that they really didn't get to the meat - how do they implement the tools and how do they teach the AI. Also, how do they prepare the teachers.

With the Tic-Tac-Toe example, it sounded like they used a premade site where they "trained" a program to play Tic-Tac-Toe. It's unclear how much AI is being taught and what understanding is really being developed. I remember a few years ago, at SIGCSE, Google showed off an AI teaching platform that had students train a pizza pr not pizza classifier. They gave examples to the computer of pizza and not pizza and eventually the program could identify pizza. The problem was that the students were given no understanding of what was going on. The most a kid would take away was the fact that it was possible for someone to write this thing that could classify pizzas.

Same for this example, at least as shared in the session. It was clear that some manner of program can "learn" to play Tic-Tac-Toe effectively via repeated input of examples of good and bad moves but nothing under the hood is being exposed. Now, I'm not saying that the schools don't go further, but I left the session not knowing anything more about implementing an AI strategy than when I entered the session.

The NBA example was also lacking. It was just a discussion, apparently because using the NBA as a real world example can be motivating but to me it wasn't really even AI but rather data science and again, no exploration of the how.

Overall, nothing wrong with the session but, like some of the keynotes, was really more of an experience report and at CSTA, if I were an active teacher, given the title, I'd want something more actionable.

Fortunately, the second session I attended, "Partnering with AI: Enhancing Coding and Debugging Skills in the Classroom" did a much better job scratching the itch. I almost didn't go to it. There was another session "Beyond OOP" which was an ML session run by John Chapin. John ran a great session on teaching AI last year so I figured this one would also be great. In the end, I decided to go to the other session figuring I can always reach out to John for his materials thus getting to see a new presenter and presentation while also getting info on the session I skipped.

The session was run by Katie O'Shea of CodeHS. It showed explicitly how AI might be used in a programming class, included a discussion of the limitations and challenges while also providing ideas on how to takes these ideas to the next level.

The talk covered strategies and difficulties in debugging and also discussed abilities and limitations of using AI as a coding assistant.

The meat of the talk was on using AI coding tools to develop code. Code that invariably has errors and then have the class dive in using debugging strategies already discussed and normalized. The speaker emphasized the PRIMM pproach (Predict, Run, Investigate, Modify, Make).

O'Shea also explicitly mentioned using bool or v0.dev to generate the code. Not that those were the only options but so that the audience had something specific they could use to get started on their own. The audience also had a chance to actually go through the process live.

At the end, O'Shea discussed some of the limitations and issues with this approach and also other possibilities for using AI in the CS class - to do a code review, to present other coding styles or solutions and more, while acknowleding that these other possibilities also have their own challenges. For instance, while an AI can share an alternate approach to students, it could also share coding techniques and constructs that students may not know or that a teacher might not want to introduce at the present time.

Wonderful session.

Given the time constraints, there were of course things that couldn't be covered. For instance, how do we know that the AI solution will have one or more errors and how do we know they'll be the types of errors we'd be looking for.

Still, I really enjoyed the session. The audience was given an overview of the problem - teaching debugging, given specific tools and techniques they could use, and even walked through the process.

Slam dunk.

I'll write about Wednesday's sessions next time up.