Courtney Milan: So a thing about AI is that, not understanding the material itself, it cannot understand what the gap is in someone else’s understanding. Students already have access to materials that will repeat important points over and over. They don’t need more.

So a thing about AI is that, not understanding the material itself, it cannot understand what the gap is in someone else’s understanding. Students already have access to materials that will repeat important points over and over. They don’t need more.

Also known as “training your replacement.” 😩

Morehouse to be the first university to use virtual AI teaching assistantswww.axios.com The Atlanta HBCU will have AI teaching assistants this fall.

They particularly don’t need a thing that will repeat important points incorrectly.

As a chem TA, one of the most valuable things I ever did with struggling students in office hours was have them narrate for me how they would solve a problem, and then figure out what it was they weren’t getting. That’s how you get the lightbulb moments.

We have access to an extraordinary amount of materials that will explain the same concepts in twenty different ways. What most people often need is to listen to the person in question to figure out what you’re missing.

This will also makes graduate school much, much less accessible—because teaching helps students finance their education—and less effective, because teaching the basics patiently, over and over again, and figuring out the pitfalls, really helps cement your own understanding.

But the idea behind this—that the robot can convey information—fundamentally misunderstands what is being taught in those core classes. You’re teaching people how to do problems: how to check your work, how to solve problem sets, how to generalize solving problem sets to other problems…

Most people will forget 80% of the information they learn in these classes. The real learning is teaching people how to tackle problems.

It's almost like most of the industries/services that are rolling out AI have suffered from decades of not being led/managed by someone who is a subject matter expert in the related fields, but instead someone who passed an MBA course.

I have heard it said that the main thing college teaches you is how to learn.

Yep, so much of the point of these classes is teaching you how to make observations, ask questions, do research, write clearly, etc. The facts you learn may be forgotten (or proved wrong!) after the class, but the SKILLS hopefully stay with you!

I remember when a Lexis AI contractor breathlessly informed us about how AI could replace research assistants! Neat, right? NO. First of all, students find actual stuff. Second, being an RA is a valuable career development skill for them and helps them earn $$. Just. These fuckin' people.

AI is going to kill entry level jobs

From this thread, it seems like the main effect will be to derail academic careers. No $$$ from TA jobs, no valuable experience from helping other people understand concepts. It’s salting the ground instead of nurturing the next crop of PhDs

Also fewer people majoring in a discipline because they don't get the help they need at the start.

But at this point, most newer PhDs are wilting on the vine (to continue your metaphor) because there are so few jobs, and every university is cutting tenure-track lines. In every discipline. It's truly been horrifying to watch.

Not to mention how bad it is for grad students who NEED teaching experience to get jobs and need the training!If you want to be anything other than full research they want you to have TAd and then planned and taught whole classes. This will mean new teaching profs come in with less experience.

(Which is not to say I love the system as it is: wherein you get teaching experience by being thrown to the wolves and not actually learning pedagogy. But denying even that bit of training?)

True. My daughter is paying for her second year of grad school by being a GTA. The bank of mom and dad is down to paying for her car insurance and phone.

THIS! I handily passed all my English classes in my own schooling, but I understand the what and why SO much better now that I teach it!

And the "what's missing" is so different from student to student! I've been teaching college math for 20+ years, and there are definitely "here are my 5 best guesses of what you're struggling with here"... and students *still* surprise me on the regular.

Or when the problem isn't actually math - I spent 20 minutes talking through all the trig with a student who was struggling with a word problem. They had all of that *down*. It eventually turned out that they didn't know what a ferris wheel was, and clearing that up meant they could do the problem.

In programming, it's fairly well known that debugging is harder than writing new code. What's often missed is that debugging *people* is also much harder than debugging code, and that's a large part of what good teaching is. Figuring out exactly where the mental block is can be tough.

Interesting. That sounds similar to the way software engineers use "rubber duck debugging" - explaining your problem out loud can often lead you to determine where you're stuck

Also true for plot issues in novel writing

Teaching is hard work - emotional and intellectual labor, both.

Even when it is right, it isn't learning. I had a student being up a code snippet they had gotten from chatGPT. It worked (not surprised, those llms have trained a lot on code) - but my student had no idea what it was doing, and they were so focused on the answer, they'd skipped all the figuring out

And I spend a fair bit of time in my class modeling debugging and providing potential avenues for solutions. So far as a teacher my experience with LLMs is that students are using it when they are overwhelmed - and it feels less intimidating than talking to a person. 🫤

Which is unfortunate, because learning is really a group exercise in many ways. And it is already hard to teach them to communicate with each other.

Yeah, that's a crucial element in my thoughts about AI in education - The fact that it gets things wrong is not inherently unworkable... One technique for education even involves intentionally doing that to help students learn critical thinking. The opacity of any reasoning process is unworkable.

I remember an education professor telling us we knew all our subject material already; he was there to help us learn how to teach that material. This set up seems like it will not do much except for a small subset of students, and certainly not have the touch education needs.

I’m teaching a summer course; I’ve been helping a student who can barely formulate questions because the material is so unfamiliar to her An AI responding to her wouldn’t be able to do more than word salad! I’m incredibly angry at the idea of AI grading or teaching assistance. It cannot do my job!

It cannot do your job *well*, but it can do it poorly for less.

Yeah, solidarity with all the writers and artists out there

Which is, of course, the big attraction. It’s so frustrating.

*So* much of what I did when I was a TA in grad school was “okay, you’re clearly still fuzzy on this concept, what if I explain it like this”, or “how about this as another example” Not to mention how being a TA was absolutely vital training for teaching my own courses

I know it's not the most important thing but I'm always impressed by how the corporate AI/VR future just _looks_ like the most depressing soulless hellscape you can imagine. They probably pipe in the sound of humming fluorescent lights just to further sap your will to live.

Seriously I like some minimalism, but the kind that comes from care and human attention (like Japanese joinery or Shaker boxes) Humans evolved in a rich and varied environment and stripping it down to the ugliest, cheapest shit is violence

It kills me how much contempt tech has for literally every other profession on the planet.

Also: Teaching is about human interaction. The whole concept of handing this over to machines is throwing 150 years of research in didactics and developmental biology out the window

It also kills me that tech companies have so very little understanding of every job or function they attempt to replace.

Wait until they start using it for grading.

Oh, I’m sure they already have.

They have.

Wonder how hard it is to get plagiarism past it.

It’s not created to identify plagiarism so it randomly spits out an answer that has nothing to do with whether the paper was plagiarized or not. www.digitaltrends.com/computing/ch...

Professor falsely fails entire class for ChatGPT plagiarism | Digital Trendswww.digitaltrends.com A college professor not familiar with how the AI chatbot ChatGPT works fails an entire class after accusing them of AI plagiarism.

Turnitin is bad enough at detecting plagiarism.

Post