I generally enjoy both panels and special sessions at SIGCSE over paper sessions. Don't get me wrong, paper sessions can be great but with those you can always read the paper and all to frequently there's not a lot of value added beyond the paper itself. Panels and special sessions can't just be read in the proceedings.

For one panel session I had to choose between

Enduring Lessons from ‘Computer Science for All’ for AI Education in Schools

and

AI in Computing Education from Research to Practice

I opted for the latter. The panel was interesting but not what I expected it to be based on the title and description. The moderator provided a list of questions and the audience voted to order them using an app. The moderator then asked the top ranked questions along with additional audience questions.

Many of the questions went over what is now well trodden ground - will kids go into CS if AI will take their jobs? How can kids be motivated to learn to program since they have AI assistents? How can we detect the use of AI tools in assignments? What are useful current AI tools? Not a lot new here.

Still there were some interesting points that came up.

One panelist discussed how through the use of AI early CS classes, CS0 and CS1 can get much further - student can do more advanced work more quickly. I started to wonder - has anyone actually checked this? I mean, yes it's easy to see that they can get through more material but do they learn it? Have the later courses been adjusted to accomodate? Has anyone studied the AI enhanced CS0/1 students vs those without it in CS2?

I've seen countless times where students seemingly achieve advanced results through the use of libraries and other tools along with extreme coaching. Many code schools as well as some highly lauded youth programs do this but the students invariable have little retention and hit a wall or regress soon after the class or camp ends.

Now I'm not saying that you can't make a better early CS experience with AI. I'm just saying that one, it's not as easy as saying "they can do more" and two, the jury's still out.

Another "positive" the panel brought up was that LLMS can help or guide students at scale - LLM based TAs and tutors. Now, I'm all for adding resources for students but people can be so hyped for some new technology they miss the problems. In this case, adding LLM based tutors to hide a courses shortcomings - something I wrote about dealing with CS50 here. Another potential danger I see is using LLM based teachers and tutors to replace human interaction.

Another concern that came out of the panel when a panelist claimed that AI tools will allow for the creation of a "curriculum in a box" - the panelist framed this as a good thing - a school could offer a class they otherwise couldn't but I read it as another sign of the deterioration of the craft of teaching.

I'll end my thoughts on this panel though with an LLM strength - that they can be multilingual. I haven't seen this exploited yet but this could certainly be a boon.

My second panel was

Data Science and Computer Science in K-12

The panel started with an interesting fact. In 2020 no states had anything in their requirements or standards dealing with data science. Now in 2024, nineteen do. That's some pretty fast growth.

The panelists brought up a number of issues:

• Data Science and Computer Science frequently overlap
• Data Science provides opportunities to develop CS skills
• CS has DS in their standards but frequently they need to be expanded on
• The question as to if DS (and CS) should be standalone classes or embedded in other subject areas

It's doubtful that data science will spin off into its own discipline in K12. It's much more likely to be integrated into other subjects but the placement of data science and computer science is an interesting one. Currently, the trend in the early grades is integration. In the later K12 grades you've got mainstream players like the College Board with standalone classes like APCS-A and APCS Principles but also integration efforts like Bootstrap which integrates into classes like high school algebra.

Now, in general I'd love to see more subject integration but looking at CS in particular, at least at the high school level, I've always pushed for separate classes.

Why? Well, first off, separate classes means separate teachers and ultimately departments. Yes, this further silos the system but without it, schools won't have long term advocates for compute science and the subject will be permanently relegated to second class citizen.

Another problem with integration is that subject teachers are overloaded as it is. Now, a science teacher or history teacher might already be doing some data science and can just relable things but CS is another matter.

At the panel, I thought of another reason. What happens when the curriculum changes? Bootstrap is integrated into a typical algebra curriculum. What happens when the curriculum changes. On the one hand, if the curriculum is just tweaked, things will probably be fine but what if there are bigger changes. Over the course of my career there were changes to both the Geometry and Algebra 2 and Trig curricula. Big enough that if the embedded curriculum is scripted and the teacher doesn't really know the CS well, that add on is going out the window. This will likely be the case with minor changes - moving or removing a single unit or changing an emphasis but what about bigger changes - adding calculators in math or the types of changes we've seen in APCS-A over the years.

This isn't even counting major changes like when we went from Algebra to integrated math to a new version of integrated math back to Algebra all in my career.

This is all to say that integration probably isn't sustainable if you're just providing a drop in and some PD. It can work if the teachers are truly trained to know both their core subject and the CS or DS but that's a big ask.

So those are my thoughts on the two panels I attended. Good stuff and good food for thought. Next up paper sessions.