That counterintuitive finding — that asking students to explain their reasoning during worked examples can actually reduce their effectiveness — likely reflects the additional cognitive load that self-explanation imposes on novice learners still building basic schemas.

Instructor presence consistently pulled attention away from the learning material itself.

In a world where machines will increasingly drive breakthroughs in climate, medicine, energy, and warfare—often while simulating human knowledge and emotion—the faculty member's ultimate function is to teach students and the public how to understand that work and to lead public debates about what to do with what technology makes possible.

faculty must learn to love students again

First, faculty must learn to work with AI teaching assistants—shaping how these agents interact with students, directing them toward specific learning goals, and knowing when to override or supplement the machine's recommendations.

Think of AI as a "genius teaching assistant" who assumes much of the work of basic knowledge transfer, unlocking learning when students get stuck and providing real-time assessment.

The goal is not to add one more obligation to an already long list, but to help institutional leaders, IT professionals, and accessibility advocates connect accessibility to other institutional priorities.

In a way, what we were seeing in that classroom was a small fractal of a larger pattern: how humans and tools interact inside a complex system, and how trust and curiosity can help a group reorganize its thinking together. The local pattern of one group of students making sense of Sheraden was reflecting something bigger about how our institution, and even our culture, is learning to live with AI.

They were not just reading about a persona’s lived reality; they were stepping into that mindset and then testing it against the neighborhood itself, using ChatGPT as a thinking partner. The tool could simulate roles and perspectives, but the students were the ones inhabiting the personas. That is where my foray into using AI in this way really began.

Anything involving risky human interaction. Clinical psychology, nursing, drug-abuse counseling. The danger there, of course, is that you don’t just need guardrails, you need castle walls. You have to be HIPAA-compliant, absolutely. But more than that, you have to make sure that what the student says never, ever leaves the classroom and can never be used in court against them. It has to be absolutely safe to totally screw up. And it has to be tested to hell.

We’re really talking about Goodhart’s Law: when a measure becomes a target, it ceases to be a good measure. The system becomes prey to gaming.

The problem is, it only gets you about 80% there, and that last 20% is garbage. And you have to be able to spot the 20% that’s garbage.If you can’t, you’re going to be in deep trouble when you get out into the real world and try to get the machine to do your job for you. You have to know how to do it yourself. You have to be able to do the process. You have to be able to do the real thinking yourself.That’s what we’re really all about in school, even if it doesn’t show. The product is merely a proxy for the process, because I can’t crack your head open and watch you think.

In some respects, GenAI is just the latest shiny thing to come along, and education has seen a lot of those. In some respects, we can treat it like that. On the other hand, AI is the printing press, it’s radio, it’s the internet. And it is going to fundamentally change things in ways we cannot predict. We need to be prepared for that. It is an absolutely disruptive technology, and it is going to change things in dramatic, fundamental ways. We just have to accept that and be prepared to ride it.

Right now there’s stigma in both directions—shame about using AI, judgment about refusing it.

commit to an answer before seeing AI’s suggestion

testing interface designs that reintroduce friction.

Being an expert neuroscientist doesn’t make you an expert at perceiving your own cognitive states.

But the AI pragmatists presuppose specific metacognitive abilities: Can writers learn to recognize, in the moment, whether a particular difficulty is productive versus merely frustrating?

The barriers that non-native speakers face are real and well documented; if AI can lower them, that’s a genuine benefit for equity in science.

Talking through your ideas with a colleague, defending your approach at a lab meeting, explaining your project to a collaborator from another field

There is also strong evidence from fields such as aviation and medicine that automated assistance will accelerate skill decay, even in experts.

The worry of “cognitive debt,” or skill decay, or however you want to frame the general issue, feels legitimate to me. But so does the counterargument that these worries are overblown

The same thing happened in the 2010s with massive open online courses, or MOOCs. Tech evangelists promised that we would not need as many professors, for one expert could teach tens of thousands online! But MOOCs were a mid technology that could barely augment, much less replace, deep expertise. Receiving information is not the same as developing the facility to use it.

The problem is that asking the right questions requires the opposite of having zero education. You can’t just learn how to craft a prompt for an A.I. chatbot without first having the experience, exposure and, yes, education to know what the heck you are doing. The reality — and the science — is clear that learning is a messy, nonlinear human development process that resists efficiency. A.I. cannot replace it.

Early film production up to about 1904 was overwhelmingly dominatedby actualities.

In 2016, Jeff Stein, a veteran journalist covering the US intelligence community, got a tip-off: a small insurance company that specialised in selling liability insurance to FBI and CIA agents had been sold to a Chinese entity.

By tracking how a student responds to failure within the AI tutor, we can generate a productive struggle score

By analyzing the digital trace data from student interactions in shared documents and team chats

systematic, personalized reinforcement

But we no longer live in an age of information scarcity. The lecture is a solution to a problem we no longer have. The challenge for colleges and universities in the twenty-first century is to deliver artisanal-quality learning at an industrial scale. For decades, this has been an impossible dream. Until now.

In the liberal arts, we often critique capitalism’s exploitative systems, yet we reproduce the same patterns in our own knowledge economy. We externalize the costs of learning and call it normal.

And here’s the truly jarring part: many of those same publishers are now selling our work again. This time to AI companies without our consent or compensation. I’ve come to label it as academic fracking: extracting value from our intellectual commons, layer after layer, until nothing of public good remains.

The “fetish” for particular kinds of writing—certain tones, certain sentence structures—reasserted itself, this time through algorithms and bias detection tools.

the professor released a custom GenAI prompt designed to interact with learners and tutor them on topics from the quiz.

This time-saving effort allows the instructor to focus time and effort on interacting with learners.

However, it is oriented toward the instructor, as the GenAI must consult the instructor for additional materials.

Instructor Proxy. In this mode, the GenAI's response is advocating on behalf of the instructor while interacting with learners.

Instructor Assistant.

intent and the orientation of the user affects interaction

GenAI is introducing novel ways for learners to interact not only with their peers and instructors but also with autonomous entities

Then we will need to teach students how to work with AI

Talking directly and candidly with her undergraduate students

Read: Models for Leading Classroom Discussions

clicking on the +Group button

It turns out that by adding your assignments to the assignment area, many of these functions of the course management system are automatically activated

maybe this goes in Module 1?

(as an aside.)

either by mouse drag-and-drop or by the Move option under an item's triple-dots menu.

Imagine, instead, if a student had to...

Creating a Pages Front Page (Special Instructions)

s clear/consistent navigation, a logical layout, and easily-accessible resources:

The video [1:58]

Submit a GLOW Sandbox Request Links to an external site..

long with annotations regarding the reasons for the edits.

The AI generates a draft of the learning design

It could improve significantly if it were trained on annotated data.

The end goal is to develop standards for integrating the ongoing ITS research (and other data-backed research streams) into continuous improvement of AI tutors.

How are those behaviors grounded in specific research findings about cognitive load?

Let’s start with the authors not providing outcomes data anywhere in the 50-page paper

The paper proposes evaluation rubrics for five dimensions of generative AI tutors:

But ChatGPT and similar programs are, by design, unlimited in what they can “learn” (which is to say, memorize); they are incapable of distinguishing the possible from the impossible.

On the contrary, the human mind is a surprisingly efficient and even elegant system that operates with small amounts of information; it seeks not to infer brute correlations among data points but to create explanations.

critically and systematically examines the cultures

ecting some text and clicking the button