Taking a New Direction

Working on the AI-assisted seminar idea

I am going to reduce my typical posts in order to focus on the immersive seminar idea. Most of my posts here will be progress reports on that project.

This idea is at the intersection of two inflection points:

• the rapid progress in AI

• the need for dramatic change in higher education

On the second point, Philip Hamburger just wrote in the WSJ,

The current institutional players are in no condition to rethink higher education. Having cultivated and tolerated violations of civil-rights laws, universities and colleges can’t afford candid introspection, lest it be understood as an admission of wrongdoing. They are controlled, moreover, by administrators who generally don’t have the stomach to recognize the damage they’ve done to higher education, let alone what should be done with their jobs.

Agreed. But then later he writes,

Imagine webpages focused on standard undergraduate readings (say, a page on “Pride and Prejudice”) with brief video conversations between leading scholars and intellectually provoking questions—to help students explore unexpected perspectives on their readings and then bring them back into the classroom. That’s much cheaper than starting new universities.

No! This is pre-AI thinking. My thesis is that we can do much better with AI.

I believe that with Claude’s help in software development, I can produce a studio that will allow a professor, like me, to create a seminar learning environment that will scale up. You as a student will interact with AI characters representing a professor and other students. Your questions and comments will get responses from other characters, just as they would in a real seminar. These responses will not be pre-canned; they will incorporate the characters’ knowledge base and their personalities.¹

You can look at another “dry” demo page here. This one takes you to a discussoin of the Warriors and Worriers model. It is “dry” in that when you type in a comment, it doesn’t matter what you type in. You get a canned response.

Claude walked me through installing the API call to make it possible to create a live demo, where your comment would be responded to by the other students in the room and by the professor. If you’re excited to see that in action, let me know in the comments below. Otherwise, Claude and I will keep on working on “dry” pages, and save the “live” implementation for later.

In the near future, we don’t see computational resources being sufficient for this studio to be multimedia. Eventually, Claude and I expect major companies, like Microsoft or Google, to build multimedia studios, that would allow author-directors to create immersive movies or video games, not just seminars. For now, we are going to treat multimedia as off the table and instead aim for a text-based immersive seminar.

In the mid-1990s, when the Web was at an inflection point, I started one of the first commercial web sites. One of the most important lucky breaks that I had was finding Dirk, an all-around software developer. In those days, my development environment was Notepad. Dirk professionalized the software process. He installed the JavaWebServer, which we needed to handle our volume. He organized our web site as a database, so that we could rapidly implement changes across the site. He set up a version control system, so that we had separate test, deployment, and backup capability. He taught me how to use Unix and vi.

Claude is the new Dirk. Here is a snippet from a conversation that Claude and I had.

[me[ What will the work flow look like? Will you develop artifacts, and then I somehow copy them and move them to Vercel?

[Claude]

Great question! Here's how the workflow would work:

Development Workflow

Step 1: I create artifacts here

• I'll build React components, API functions, etc. as artifacts in our conversation

• You can copy/paste the code directly from the artifacts

• Each artifact will be complete, functional code

Step 2: You set up the project locally

• Create a new Next.js project on your computer

• Copy my artifacts into the appropriate files

• Test everything works locally first

Step 3: Deploy to Vercel

• Push your code to GitHub (or GitLab/Bitbucket)

• Connect your GitHub repo to Vercel

• Vercel automatically deploys when you push changes

Note: My readers will only be able to play with the software when I get to Step 3. If you’re really interested, let me know in the comments. Otherwise, I will just focus on testing and improving the seminar for my own satisfaction, and not worry about deploying it to the Web.

For me personally, one advantage of trying this project is that it will help me get a “feel” for what AI can and cannot do at this stage. With a live version, I can test and promote the thesis that seminars are the best form of education for humanities and social science.

As an aside, I’ll also learn something about software development nowadays, which seems in the last 30 years to have gotten beyond using Notepad, coding by hand, and deploying using a command-line interface and FTP.

I hope readers enjoy updates on this project. If not, you can always ignore them.

1. (Claude insists that this would be better executed by having each character be handled by its own AI, but I am thinking for now of having a single AI handle all of the characters.)