AI-augmented Learning? Or the end of unaided thinking?

I'm one of the (former) educators who feels that the advent and near future evolution of large language models is already driving a radical shift in pedagogy at all levels, and in particular in higher education. Like the rest of discourse about the evolution of AI (cure cancer? unleash Terminator?), the shape of change is not yet clear.

I have already experienced two shifts that are still playing out -- the advent of what are now called social media and the Zoomification of curriculum triggered by Covid. I remember when I started teaching a course that started out as "virtual community" in 2005 and ended up as "social media issues" by 2010. In 2005 I could identify only a half dozen professors who were actively using forums, blogs, chat, social bookmarking, RSS. When I asked the director of Stanford's Center for Innovations in Learning, he told me: "That's easy. This is a knowledge factory. Faculty are hired because of their expertise in their specialty. If they are supposed to teach and don't show up, that would be a problem. But there are no incentives for innovation in pedagogy."

The old model for college professors consisted of working diligently on a lecture, then delivering it (in many fields) every year for decades. YouTube blew that out of the water. Why show up at a specific room at a specific time when you could sleep in and catch the lecture on YouTube? Internet search and early online social networks, from MySpace ,Tumblr and Livejournal to Facebook, connected students to information and to each other in ways never before possible (although, as I started pointing out in 2009, the authority of the text flew out the window and the responsibility for verifying knowledge shifted from the authority to the seeker. )

Because online media constituted the subject I was teaching, it seemed natural and necessary to use the media students were immersed in outside class to extend the physical classroom and enable them to find new pathways of learning more suited to the world they inhabit outside the classroom. I retired a decade ago. From where I sit, higher education has failed to embrace the use of online media at scale. Then, in 2020, Covid forced elementary and middle school teachers as well as university professors to shoehorn their courses through videoconferences.

Zoom school was not highly regarded by a large number of teachers and students, for good reason. Drawing on my own experience, I wrote during the early months of lockdown about how to make online learning engaging and effective.

Now the driving force is potentially far more tranformational and destructive to real learning than social media and videoconferencing were: AI is a formidable challenge to old ways of doing things from the teacher's point of view. The ability to synthesize artificial essays in under a minute challenges the pedagogy that used writing to help students learn how to think. I was inspired to write this post when I read the claim by a contemporary graduate student that his generation is "afraid of thinking without AI."

Who is conducting large-scale research into what LLM pedagogy might look like? Chinese, educational researchers for one:

From MOOC to MAIC: Reshaping Online Teaching and Learning through LLM-driven Agents

Since the first instances of online education, where courses were uploaded to accessible and shared online platforms, this form of scaling the dissemination of human knowledge to reach a broader audience has sparked extensive discussion and widespread adoption. Recognizing that personalized learning still holds significant potential for improvement, new AI technologies have been continuously integrated into this learning format, resulting in a variety of educational AI applications such as educational recommendation and intelligent tutoring. The emergence of intelligence in large language models (LLMs) has allowed for these educational enhancements to be built upon a unified foundational model, enabling deeper integration. In this context, we propose MAIC (Massive AI-empowered Course), a new form of online education that leverages LLM-driven multi-agent systems to construct an AI-augmented classroom, balancing scalability with adaptivity. Beyond exploring the conceptual framework and technical innovations, we conduct preliminary experiments at Tsinghua University, one of China's leading universities. Drawing from over 100,000 learning records of more than 500 students, we obtain a series of valuable observations and initial analyses. This project will continue to evolve, ultimately aiming to establish a comprehensive open platform that supports and unifies research, technology, and applications in exploring the possibilities of online education in the era of large model AI. We envision this platform as a collaborative hub, bringing together educators, researchers, and innovators to collectively explore the future of AI-driven online education.

I add a grain of salt at the mention of MOOCs, which were touted as the successor to place-based higher education...and fizzled. But research into pedagogical innovation is worth monitoring. More concerning to me was this article: "Adults lose skills to AI. Children never build them." I have a 5 year old grandson. How do I, his parents, and his teachers guide his learning so he knows how to use AI-powered tools without losing his ability to think and write? So -- and I don't see this as contradictory -- I asked chatGPT where to start. My prompt:

generate a literature review with links to sources and an extensive executive summary of peer reviewed research regarding the possibility that using large language models diminishes the human ability to think well.

Of course I checked the links. You can read chatGPT's literature review here.

Or you can read the executive summary:

Yes, the peer-reviewed literature supports the possibility that LLM use can diminish the human ability to think well. The best-supported pathways are reduced mental effort, maladaptive cognitive offloading, weaker retention, and passive acceptance of generated answers. But the literature also shows that LLMs can enhance critical thinking when they are used as scaffolds for debate, inquiry, and reflection rather than as substitutes for those processes. So the most accurate conclusion is:

LLMs are not inherently corrosive to thinking; uncritical dependence on them can be.

I checked the sources. Links are in the chat.

One of the clues to combatting cognitive offloading, according to the literature, is cultivating metacognition -- a topic I have covered before, here and here.

Next, I asked chatGPT "How to teach metacognition to a 5 year old boy." The chat is here. The key appears to be making thinking audible rather than thinking about thinking. The exercises suggested by chatGPT were drawn from three main sources:

Lev Vygotsky → thinking becomes internal through dialogue

John Flavell → awareness of thinking develops through reflection

Carol Dweck → focus on strategy, not correctness

For example, modeling metacognition starts by narrating my own thinking during everyday moments:

“Hmm… I’m not sure where I put my keys. Let me think—where did I go last?”

“This puzzle is tricky. I’m going to try the corners first.”

“That didn’t work. I need a new plan.”

I'm also teaching scientific method, so before we do an experiment, I ask grandson what he thinks is going to happen (he knows the word "hypothesis"), what he observes happening, and what he thinks about why it happened the way he did. (Vygotsky asserted that children learn to internalize thinking by observing how others articulate their thought processes verbally.) Three questions that fit many childrens' activities: "what is your plan?," "what worked?," "what will you try next time?" I give him a pile of craft sticks and clothespins and encourage him to build without specific instructions. After he constructs a tower or a car, I ask him these three questions.

My hope when I started writing about social media literacies 16 years ago was that curriculum on crap detection at least would be included in high school and college curricula (elementary and middle school would be ideal, but those institutions are even more conservative in terms of change). I am less sanguine about teaching kids to think for themselves in the era of ai chat, given that experience. But I do intend to pursue my own research and try different approaches with my grandson -- and to make my experiences and resources available to others who want to try.