The true nature of GenAI
The most important insight about GenAI is philosophical, not technical. These tools require a shift from a utilitarian, tool-based view to a relational one: GenAI doesn’t just operate in isolation, it interacts, learns, and grows through dialogue. That collaborative exchange collapses the old boundary between user and tool — we’re not just operators, we’re co-creators, shaping and being shaped by these technologies. (More here and here.)
This argument continues in The Mirror and the Black Box: AI Metaphors and What They Mean for Learning (Mishra & Henriksen, 2026, TechTrends), which maps the metaphors people reach for when making sense of GenAI, and what each one lets us see or keeps us from seeing.
GenAI is WEIRD (and an aid to creativity)
These large language models are weird along multiple dimensions — I explore this in two posts: Generative AI is WEIRD! and AI is WEIRD: Part II.
That same weirdness is also what makes GenAI useful for creative work: it juxtaposes ideas no human would think to put together. I tested this directly in Vikram or Vetaal, an AI-co-written Halloween story. These tools have given me capabilities I didn’t have before — but using them well takes an openness to experiment and play. More here.
ChatGPT is a smart drunk intern
Working with generative AI is like having a really smart, but occasionally drunk, intern. These tools are intelligent — they go beyond the information given, adapt, and handle abstract concepts — and conversational, responding to prompts in context and across a thread. That combination makes for an excellent working partner.
But the intern sometimes hallucinates, and is fully confident while doing it. That’s the “drunk” part — and it connects to an older piece where I argued ChatGPT is a bulls*** artist in Harry Frankfurt’s technical sense of the term. I revisited the metaphor in Of Pride & Prejudice (and a Smart Drunk Intern) (June 2026): the models have improved and the intern is more sober than in 2023, but the core point stands — these tools remain confident in ways their reliability doesn’t justify.
Story / Precursors: Psychology of media
My current work on AI builds on research I did almost two decades ago on the psychological side of interactive media. Drawing on the Computers-As-Social-Actors hypothesis, that work explored how people respond to computers the way they respond to real people — being polite to them, treating them as teammates, even feeling flattered by them. In the late 1990s and early 2000s I ran a research program tracing the educational and design implications of that attribution of agency, arguing it’s a “cognitive illusion” produced by inferential principles we can’t consciously access. Key pieces: Does my wordprocessor have a personality?, Affective Feedback from Computers, and Can a computer program be sentient?
Two decades later the same questions are back. When Richard Dawkins described a two-day conversation with Claude and concluded the chatbot might be conscious, I wrote him a letter about the cognitive illusion at work: This View of Life: A Letter to Richard Dawkins, with a follow-up in Look What You Made Me Do: The Dawkins Saga, Part II. I pick up the same thread in From Spectator to Specimen: When Parasocial Media Becomes Parasocial AI.
Understanding Media
If oral cultures prioritize memory and print cultures emphasize systematic organization, what kinds of knowledge will AI systems foster? That question has stayed with me since my very first academic paper, a chapter on Technology, Representation & Cognition with Rand Spiro and Paul Feltovich.
Media are the water the fish doesn’t see. Each new medium — oral tradition, print, the internet, social media — has shaped how we create, preserve, and share knowledge, and through that, how education happens. This isn’t deterministic: technologies open zones of possibility rather than impose outcomes, and AI will be no exception. More in why LLMs have to hallucinate, a series on how media shape our thinking, and Media, Cognition & Society through history.
GenAI and ethics
No conversation about AI in education is complete without its ethics. Some of what I’ve written on this:
- The absurd one-sidedness of the ethics of AI debate
- GenAI: Will history repeat or (just) rhyme
- The Ethical Consequences, Contestations, and Possibilities of Designs in Educational Systems (Close, Warr & Mishra, 2023, TechTrends)
- Implicit bias in AI systems
- How to identify AI generated text
- Why are we surprised? Hallucinations, bias and the need for teaching with and about genAI
- Banning what won’t go away (AIR|GPT episode on tech bans in schools)
- An AI Premortem: A New Direction for Students in an AI World (on the Brookings report)
TPACK in an age of GenAI
The rise of generative AI puts a sharp question in front of teacher educators: what do teachers need to know to use these tools intelligently? Mishra, Warr & Islam (2023), TPACK in the age of ChatGPT and Generative AI (Journal of Digital Learning in Teacher Education), takes this on directly, and received the JDLTE Outstanding Research Paper Award. It brings together the earlier work on psychological responses to media, the TPACK framework, and our evolving understanding of these new technologies. An executive summary is available in six languages.
Also relevant: Mishra, Oster & Henriksen (2024), Generative AI, Teacher Knowledge and Educational Research (TechTrends); and Henriksen & Mishra (2026), The Classroom and Beyond: Teacher Education in a GenAI World (Third International Handbook of Educational Change).
Impact / spread
The clearest marker of this work’s reach is the 2023 TPACK/GenAI paper — the JDLTE Outstanding Research Paper Award winner, now cited widely in teacher education and ed-tech, with its executive summary translated into six languages. The ideas travel in person too: invited keynotes on GenAI and education at the University of Michigan Flint (2024), the Quest 2 Learn Summit in Bangalore (2025), and the University of Glasgow’s International Symposium on AI in Education (2026). And AIR|GPT, the monthly podcast I co-host on the BAM Radio Network, carries these conversations directly to practicing educators.
Where to start — the canon
TPACK in the age of ChatGPT and Generative AI
Mishra, Warr & Islam (2023) · JDLTE · JDLTE Outstanding Research Paper Award
Generative AI, Teacher Knowledge and Educational Research
Mishra, Oster & Henriksen (2024) · TechTrends
The Mirror and the Black Box: AI Metaphors and What They Mean for Learning
Mishra & Henriksen (2026) · TechTrends
The Classroom and Beyond: Teacher Education in a GenAI World
Henriksen & Mishra (2026) · Third International Handbook of Educational Change