prior probability

I promised to say a few words about the recent “Teaching with AI” conference that I attended earlier this week. (See my previous post #TeachingWithAI, which I am reblogging below.) First off, let me begin with the big question regarding AI and higher ed: should college professors like me embrace new “GenAI” tools like ChatGPT, Midjourney, Bard, etc., or in the alternative, should we adopt a clean and simple bright-line rule prohibiting the use of A.I. altogether. Or more to the point, should we resist or join the revolution? This, in a word, is why I attended this AI conference: to try to answer this big question.

For my part, although I am already redesigning my courses in order to allow my students to test these A.I. “large language models” or LLMs for themselves (see here), I should disclose my Burkean priors: I am deeply suspicious of radical change, especially when done hastily, and my biggest fear or worst-case scenario regarding “GenAI” is that overuse of these new cheating genies could breed an entire generation of abject idiots! The pro-A.I. crowd, by contrast, tends to minimize the unprecedented and destabilizing nature of these powerful and super-fast A.I. cheating tools. By way of example, many speakers at the “Teaching with A.I.” conference compared ChatGPT to the pocket calculator, i.e. as just another step in our human problem-solving capabilities. Alas, I suspect this analogy is a false one for two reasons: unlike an abacus or pocket calculator, LLMs like ChatGPT are not only amassing and even stealing massive amounts of copyrighted materials and user data; they are also beginning to replace our ability to think for ourselves.

Given my priors and Burkean intuitions, my favorite talk at the conference was titled Thinking Slowly in the Age of AI. (Shout out to my colleagues Bruce Lenthall and Jessica Morris from the University of Pennsylvania for their excellent work. I have posted a link to their slide deck below.) In brief, Lenthall and Morris explained how college professors need to first decide what they want their students to learn from their courses before deciding whether and how to expose students to A.I. But how can we move more slowly and deliberately when the world around us is moving so quickly? On the last day of the conference, for example, OpenAI announced that it is now rolling out new voice and image capabilities in ChatGPT!!! No, you cannot make this up!

Will A.I. platforms like ChatGPT convert college degrees into worthless pieces of paper? Has higher ed now become all but obsolete? Thanks Google, Meta, and OpenAI (sarcasm voice)! Earlier this week, I attended the Teaching with AI conference at my home institution, the University of Central Florida. (Kudos to my UCF colleagues Kevin Yee, Kirby Whittington, Erin Doggette, and Laurie Uttich for making this happen!) Suffice it to say that I amassed a lot of information during this cutting-edge, two-day conference and will be sharing the main lesson I learned in my next post.

Mind blown! I just saw this real-life example of the Condorcet paradox on “Around the Horn” (ATH), namely the episode that aired on 26 September 2023:

To the point (pun intended), this year’s WNBA MVP award went to Breanna Stewart (who plays for the New York Liberty), even though Alyssa Thomas (Connecticut Sun) had obtained the most first-place votes and A’ja Wilson (Las Vegas Aces) had garnered the most second-place votes!

More broadly speaking, this voting paradox emerges when there is no clear, universally agreed-upon winner with three or more alternatives. By way of illustration, imagine a referendum to decide the best flavor of ice cream. For simplicity, there are only three voters (A, B, and C), and their individual preference rankings are as follows:

Person A prefers Chocolate > Vanilla > Strawberry

Person B prefers Strawberry > Chocolate > Vanilla

Person C prefers Vanilla > Strawberry > Chocolate

This set of preferences produces a Condorcet cycle because none of the options can consistently beat the other two choices in pairwise comparisons. In my simple ice cream referendum, each flavor can win against one flavor but loses to another. One way to solve this impasse is to assign weights to each rank, e.g. 10 points for first place, seven points for second place, and five points for third place (like in the WNBA vote pictured above), but in my ice cream example each flavor would end up with 22 points apiece, hence the paradox!

PS: I identified another real-life example of the Condorcet Paradox here.

Here it is. Alas, this new syllabus by Harvard law professor Stephen E. Sachs has come under fire from some academic quarters for excluding the work of feminist, queer, and minority scholars (see here, for example). Contrary to my colleague and friend Brian Leiter (here), I think the criticism is a fair one, but that said I find the omission of Lon Fuller to be the most unforgivable one!

I featured my first ChatGPT-inspired “micro essay” in a previous post. Below is a screenshot of the new “micro essay” for my module on the common law:

FYI: I will further discuss the motivation for this assignment in a future post (after my students have submitted their responses); in the meantime, I cut-and-pasted ChatGPT’s original response to my prompt below the fold:

Overall, according to this 2023 report by Al Jazeera, of the 486 attempted or successful military coups around the world since 1950, almost half of this total or 214 have occurred in our “Mother Continent”, and of those 214 African coup attempts, 106 have been successful. The map below, by contrast, shows the sundry coup d’etats that have taken place in Africa since the outbreak of the Wuhan pandemic in 2020.