Late Lessons, Jensen Huang and AI
Markdown version

B23 notes: 2025-03-30 to 2025-04-01 (2 posts)#

Reading notes on Andrew Maynard’s Substack posts in batch B23. Only his own prose counts as evidence. Quotes are exact, including his original typos, curly punctuation and italics (shown as asterisks).

Batch context: the end of March 2025. He had just given a keynote, “What does the future look like?”, at the 2025 Yidan Prize Conference. ASU’s Mary Lou Fulton College of Teaching and Learning Innovation co-hosted the conference to mark Micki Chi’s 2023 Yidan Prize for Education Research. The first post writes up that keynote. The second announces a Modem Futura podcast episode (with Sean Leahy) about the same keynote. Per the user’s instruction to skip Modem Futura podcasts, the second post is noted in one line and not analysed.

Relevance summary:

Date Slug Relevance
2025-03-30 reimagining-education-in-an-age-of-ai high
2025-04-01 when-ai-takes-the-wheel none (Modem Futura podcast post; skipped per user instruction)

HIGH#

2025-03-30 — reimagining-education-in-an-age-of-ai — “Reimagining learning and education in an age of AI”#

Subtitle: “Reflections and provocations from a keynote given at the 2025 Yidan Prize Conference”.

Provenance. Almost all of it is his own prose (about 2,500 of about 2,900 words). He wrote it up after giving the keynote. - Not his: two quotations he used on his slides. One is from Dario Amodei (“Machines of Loving Grace”, October 2024) on the “compressed 21st century”. The other is from Mark Daley (Western University’s Chief AI Officer, whom he calls “Chef AI Officer”; “AI isn’t running water”, March 2025) on machines “that can think” cutting “to the heart of who we are”. These are evidence only of what he chose to frame the talk with, not of his own words. - AI-related links: a “Top takeaways (generated by Perplexity)” link and a “ChatGPT summary and further reading” link. These are prompt links, with no AI text in the post itself. - Slides: about 16 slide images whose text is not in the corpus. Some of the argument (for example the “where we live” dimension in footnote 1, and the models he calls “simple”) was carried by the slides, so the prose is a partial record of the talk. - Header image: captioned “NOT generated by Midjourney!”. This is a small signal that he labels AI imagery and wanted readers to know this image was not AI-made.

Argument in his terms. - Framing: a conceptual model of how transformative technologies change “what we do, and who we are”. He calls it “a conceptual model that I find myself using increasingly frequently”. Technologies have always moved us along both axes. AI may push us “far further along these two axes than previous technology transitions”. In footnote 1 he adds a third dimension he only mentioned in passing in the talk, “how technologies alter where we live”, and argues that AI “is also in a category of its own” on that axis too. - Amodei stands for “what we do” and Daley for “who we are”. He pairs Amodei’s claim that AI will compress 50-100 years of progress in biology and medicine into 5-10 years with Daley’s claim that thinking machines set AI “on a different plane than plumbing or electricity”. He says both “push the limits of our imagination” but are “critical” for seeing how AI challenges “deeply held beliefs about our relationship with technology and the future”. - The AI trajectory is uncertain, and his own best guess is a plateau. He gives three trajectories. The first is an overhyped bubble and “yet another AI “winter.”” The second, which he rates “even more likely in my perspective”, is a “classic innovation “S” curve” with a ceiling at an unknown point. The third is continued exponential growth over 10-20 years, which he calls “more speculative” but with “some credence”. - Even a small chance of transformative AI must be taken seriously. The key move is decision-theoretic. “No matter which trajectory ends up being closer to the truth”, a chance of large capability growth “demands new ways of thinking”. If AI is as transformative as “Dario, Mark and others (including myself) believe it could be”, its role as an accelerant “in virtually every domain of technology innovation”, and its effect on who we understand ourselves to be, “put it in a fundamentally different category to every previous technology that we’ve developed as a species.” The claim is conditional (“If”), but he puts himself among those who think it could be so. - Why learning and education matter: a value-creation model. He deliberately asks educators why education matters, rather than assuming it does, which he calls “a little risky”. His model starts from our “evolved ability to imagine futures that are different to the present”, to imagine better lives in them, and to “problem solve” our way there. That is how humans create “value”, through “curiosity, imagination, creativity, and ability to solve problems (or innovate)”. He defines “value” broadly: health, wealth and wellbeing, but also quality of life, the planet and our relationship with it, social stability, community, dignity, equity, justice, kindness, and “our capacity to experience awe and wonder”. Learning “vastly” enhances these innate abilities. Education, “the formalization of learning”, amplifies them further. So education matters because it increases “the rate at which we can create value”. - The disruption. “But what happens when this essential part of being human is superseded by machines?” If AI “can problem-solve faster and better than any one person, or any collective of people”, where does the value of education lie? He says this is not hypothetical “given the rate at which AI capabilities are continuing to advance.” - Three provocations. 1. What does it mean to be human? AI can emulate many qualities “we think of as uniquely defining us”, which “does put a wrench in the works” of long-running philosophical debates. The educational task is to “center what it means to be human” when technology can “emulate much of what makes us “us.”” 2. What does value creation mean when intelligence is “free”? He takes up Daley’s idea and calls it “controversial, and not entirely accurate when the material costs of developing training and running AI models are considered”. Yet anyone with a smartphone may soon have on-demand “intelligence” that “potentially far surpasses what they can get from human educators — effectively making it free to them.” This challenges a centuries-old “scarcity model” of access to intelligence (see inequality below). It raises the question of which skills students need, and “what roles, if any, human educators will have”. 3. How do we equip future generations to create meaningful value? Humans are “not simply wealth-creation machines” but people who “yearn to create value that means something to us.” Here he turns the question around. Instead of asking what happens when AI threatens who we are, he asks “what happens when we learn how to work with AI to make us more than we are.” Educators must create “learning environments that leverage both human and artificial intelligence”. - Three closing perspectives. 1. “We cannot stop the march of innovation, but we can guide and steer it.” Technology is not deterministic, but innovation is in “our metaphorical DNA”. We have agency over which futures we aim for and how we get there. 2. Flourishing requires “the humility to question assumptions and embrace change”. “Change is the natural state of a universe governed by the flow of time.” 3. Advanced AI “challenges the deepest foundations of the why and how of learning and education”. If even part of the promise comes true, education faces “an existential crisis”. But leaning in “while staying tethered to what makes us human” opens “incredible” possibilities. - Closing line. We must reimagine education “or risk sacrificing what could be on the altar of a blind devotion to what is.”

How firmly. He openly labels the talk provocative and says “there’s a lot of hand waving here”. He calls his models “simple — some would say simplistic”. He also says he is increasingly unsure of the future: “week by week, I’m less sure what the future holds”. He is firm on four points: education must re-examine its why as well as its how; a small chance of transformative AI warrants serious rethinking; innovation cannot be stopped but can be steered; and humans must stay “tethered” to what makes us human. He is cautious on the AI trajectory (he leans towards an S-curve ceiling) and on “intelligence is free” (contentious, oversimplified, footnoted).

Concepts and frameworks he names or develops. - What we do / who we are (/ where we live): his model of the axes along which transformative technologies change us. He says he uses it “increasingly frequently”. AI is placed in “a category of its own” on all three. - Value-creation model of learning and education: imagine better futures, then problem-solve towards them, which creates “value” in a broad, plural sense. Learning enhances this; education formalises and amplifies it. - The “compressed 21st century” (Amodei’s term), which he uses as a framing device for acceleration. - “Intelligence is free” (Daley’s idea) versus a “scarcity model” of intelligence, which he describes as “limited supplies, hoarded by elite establishments”. - Three AI trajectories: AI winter, S-curve with a ceiling, and exponential growth. - Learning to be human in an age of AI, as opposed to only “grappling with the challenges of being human in an age of AI”. This is a shift from defending human distinctiveness to growing human capacity with AI. - Steer, don’t stop: non-determinism combined with the inevitability of innovation. - Humility, and change as the natural state of a time-governed universe. - An “existential crisis” in education. He means a crisis of education’s purpose and identity, not existential risk to humanity. - Learning and education as “a profound evolutionary self-improvement algorithm” (footnote 2).

Analogies and comparisons. - Past technology transitions, used structurally. Technology has always changed what we do and who we are. The claim is about degree (“far further”) and then about kind (“fundamentally different category to every previous technology”). - Plumbing and electricity (through Daley’s quotation): AI is not a utility like running water because it “cuts to the heart of who we are”. He endorses the “who we are” framing but does not use the utility comparison himself. - Evolution: human learning and education as an evolved self-improvement algorithm. In footnote 2 he asks whether “the same might not be possible in machines” that learn to problem-solve towards imagined futures. This is a conceptual analogy between human and machine development. - The innovation S-curve and AI “winters”: standard models from innovation history, used to frame uncertainty about the trajectory. - No comparisons with chemicals, nanomaterials, GMOs, nuclear power or biotech.

Views on AI. - What kind of thing it is: a technology that can “emulate” or “simulate” key human attributes. He hedges on whether anything lies behind this: “even if it’s blind emulation with nothing more than unthinking digital processes behind it”. The effects on who we are matter whatever AI’s inner nature. - What is new: (a) it is a general accelerant “in virtually every domain of technology innovation”; (b) it challenges human self-understanding by emulating qualities we took to be uniquely ours; (c) it may make “intelligence” nearly free to users. Together these could put it in a new category of technology. - Openness to machine agency: footnote 2 raises the possibility of machines that “learn to problem-solve their ways from the present they inhabit to futures they too can imagine”. This sits alongside the “blind emulation” hedge. He keeps both possibilities open. - Trajectory: genuinely uncertain. His best guess is an S-curve ceiling at an unknown point, but he treats exponential growth as possible enough to plan for.

Views on AI risk. He does not frame AI risk in hazard terms here. The risks are to human identity and purpose, and to the role of educational institutions. He speaks of “serious threats to who we are” alongside “new possibilities”, and of the value of education being “superseded by machines”. The “existential crisis” is institutional and conceptual. The structure of his reasoning is still worth noting: take a low-probability but high-impact trajectory seriously enough to rethink now (“This possibility — even if it’s small — demands new ways of thinking”). The final risk he names is the reverse of a technology risk: the risk of not changing, of “a blind devotion to what is”.

AI companies and leaders. He quotes Anthropic CEO Dario Amodei respectfully, as one of “two prominent thinkers and commentators on AI”, and uses “Machines of Loving Grace” as a legitimate if imagination-stretching projection. He offers no critique of Anthropic or other companies. He treats Amodei’s forecast as a possibility to be taken seriously, conditional on uncertain trajectories, not as a prediction he endorses outright.

Governance and who decides. He does not discuss regulation or policy. His governance-adjacent claim is about agency: technology “doesn’t just happen to us”, and “we do have the agency to determine what futures we aspire to and how we get there.” The “we” is collective and unspecified: society and, for this audience, educators. The “guide and steer” stance continues his earlier assumption (2024-08-18) that innovation cannot be switched off but can be channelled. Educators are cast as the actors who must redesign learning environments.

Cognition, language and human formation. This is the post’s centre, though framed through education rather than language. - Education is human formation that amplifies innate capacities to imagine, be curious, create and solve problems. - AI both threatens this (by superseding problem-solving) and could extend it (working with AI “to make us more than we are”). - The educational task becomes centring “what it means to be human” and learning how to “be human in an age of AI”. - He does not discuss cognitive offloading, deskilling or manipulation here. His focus is the purpose of education, not how AI affects thinking.

Inequality and justice. Access to intelligence has historically come “at a price”: “Something that the wealthy can purchase”, in a scarcity model run by “elite establishments”. AI is “beginning to turn this model on its head” by giving more people access. He calls the idea of AI “liberating access to intelligence from the shackles of a market-driven system” “deeply contentious”. Footnote 3 says the scarcity model is “far more complex” and lists tuition, consultancy, subscription and speaker fees. This democratising reading is hopeful but hedged. He does not mention the costs of AI access or the concentration of power in AI companies here, beyond noting the “material costs” of models.

What he criticises and who he engages. - He criticises: educators’ assumption that education’s importance is self-evident (“rationalizing their importance simply because we’ve always been told they matter”); the scarcity and “elite establishments” model of access to intelligence; and institutional inertia (“blind devotion to what is”). - He engages: Dario Amodei, Mark Daley, the Yidan Prize, Micki Chi and ASU’s Mary Lou Fulton College.

Change of view signalled. - He names the what-we-do / who-we-are model as something he uses “increasingly frequently”. It is emerging as a core framework. - He signals growing uncertainty about the future: “week by week, I’m less sure”. - He moves within the talk from AI as a threat to who we are to AI as a partner in becoming “more than we are”. This is a rhetorical turn more than a stated change of mind. - It is continuous with the 2024 technology-transitions work: innovation cannot be stopped, but can be steered; humility; embracing change.

Quotes. - “how transformative technologies are changing what we do, and who we are.” - “put it in a fundamentally different category to every previous technology that we’ve developed as a species.” - “Technology is not deterministic — it doesn’t just happen to us.” - “or risk sacrificing what could be on the altar of a blind devotion to what is.”

LOW / NONE#