Late Lessons, Jensen Huang and AI
Markdown version

B14 perspective notes: 2024-02-11 to 2024-04-14 (18 posts)#

These notes read the batch for how Maynard thinks, not for the concepts he names. I read all eighteen posts in full. For the six “In Conversation With …” audio posts only his framing lines count as evidence. Quotes are exact, including his typos, apart from markdown italics, which are dropped.

Evidence rules applied

Context. The batch runs from February to April 2024. Sora has been announced but is not yet public. Gemini’s image generator has been paused. NVIDIA has just passed $2 trillion. Neuralink has shown its first patient, a Waymo car has been burned in San Francisco, and Orchid has started selling whole-genome embryo screening. Maynard is spending his first weeks with an Apple Vision Pro. He is writing reviews of Dune: Part Two and The Creator for Jurimetrics, and teaching “Pizza and a Slice of Future” with a surprise intake of retirees. He is also hosting Jonathon Keats for a week of recorded conversations produced by Sean Leahy, a precursor of the later podcast work. Several posts form one chain of thought built across weeks: Harari (03-10), then playgrounds (03-17), then Sora for kids (03-24), then the technology problem (03-31), whose afterword revises the Harari list.

1. How he thinks here#

He opens from something that happened to him#

Almost every substantive post starts from a concrete encounter, not a thesis:

The large claims grow out of the particular. In the technology-problem post, the irritation at being asked about optimism becomes an argument about how humanity frames every global challenge.

He finds out by playing with the thing and watching people#

The Vision Pro post is a record of finding out by doing. “I’ve been playing with the Vision Pro with a particular eye to how it might be used to substantially enhance or otherwise impact the things we do.” He does real work in it, and treats his own body and mood as data. Full immersion made him “feel isolated”. Passthrough made the virtual workspace feel like “an extension on the real world, not a substitute for it.” He also treats other people’s reactions as data: “I’ve been letting my students try out the Vision Pro and watching their responses”, and the students choose movies “every time”. The post is about his view moving: “I’ve found my thoughts on the headset changing almost daily this past week — and in ways I wasn’t expecting.” It ends with that change recorded: “I’m less concerned than I was a week ago that spatial computing could plunge us into a societal crisis”.

He also follows odd intersections because they look odd. The initiative is looking at the Vision Pro and the future of travel, “possibly an odd-seeming intersection until you begin to think”.

Research as a followed thread, with serendipity he admits to#

The playgrounds post (2024-03-17 undergraduate-playgrounds-not-playpens) shows his research method openly, with self-mockery. “I started playing around on Google after posting last week’s article — a research “methodology” that I feel is justified by where I ended up”. He finds Resnick’s 2004 critique of edutainment. He admits it suited his priors: “Mitch’s article engaged my cognitive biases to the full!” Then he checks it by writing to the author, “to see if he still holds this view”. Resnick points him to Lifelong Kindergarten, “And this is where things got interesting.” A chain of chance, checking and redirection produces the playpen and playground metaphor. In the postscript he names what happened: “a moment of validation though — albeit one that was probably tinged with confirmation bias.”

Reframing is his main move#

He states the other side’s case before judging it#

In the embryo post he sets out Orchid founder Noor Siddiqui’s moral case at length and with sympathy before disagreeing with it (“I’m paraphrasing here, but these are arguments that have a sizable following”). He concedes the strongest ground: “I would be the first to agree that there are strong arguments to be made for having as much information as possible”. He even supplies a defence the company might make: “Of course, Orchid may be using a counseling and decision-making regimen that counters my concerns.”

Imagining, then testing the imagining for plausibility#

Imagine and what if do real work in these posts:

He then checks each image. The embryo scenarios are “both technologically feasible and economically plausible”. Backlash-driven discrimination is “a remote but plausible outcome”. For Levin, the bioelectric premise makes it possible “to imagine plausible technologies”. His own Sora idea is “highly speculative at this point”. In the Waymo post he holds back until the evidence is in: “Without more evidence it’s hard to say exactly what transpired here”.

Analogy by structure, and history as a sense check#

Stories and films as ways of thinking#

Thinking that unfolds across posts#

Harari (03-10) prompts the playgrounds post (03-17), which gives the “playground of the mind” that the Sora post (03-24) borrows. The technology-problem post (03-31) then goes back to the Harari list and admits a gap: the people, technology and future relationship “isn’t overtly there”. He adds a cross-cutting area, “a warts-and-all deep dive into how technology defines us and how we in turn define the technologies”. He revises his own frameworks in public, from one week to the next.

Curiosity and delight drive the thinking#

This is the pleasure of someone who follows what fascinates him into biology, time, choreography and photography, well outside his home field.

He designs conditions for serendipity, not outcomes#

This is the clearest methodological thread in the batch.

2. What matters to him#

Human flourishing, and the essence of being human. - The aim of his work is “a future of human flourishing on an equally flourishing planet” (03-31). - In Dune he values the “unique “human-ness”” that brings “value to ourselves and those around us” (03-03). - The deterministic case for embryo screening troubles him because it seems to “boil human flourishing down to a balance sheet” and risks “problematizing the very essence of what makes us human” (04-14). - He takes care not to impose his own picture of flourishing. The Harari list adds “assuming that this is something people aspire to”, and asks what flourishing does “not look like”.

Learning, curiosity and freedom of mind. - He describes his own appetite frankly: “my curiosity and hunger to learn has always led to me getting frustrated with educational tools and environments that seem to throttle the flow of knowledge”. He wants “a learning firehose rather than a educational trickle”, and would take “a good book rather than a $3500 headset any day!” (02-11). - What frustrates him most in education is “unimaginative attempts to constrict and quantify learning to the nth degree”. He dislikes “playpens masquerading as playgrounds” and learning “measured, cataloged, compared, and quantified to the extent that all useful degrees of freedom are effectively shut down” (03-17).

Imagination and expression, especially for those without the skills to show it. “many of us are frustrated artists, or video makers, or storytellers — simply because our technical skills fall far short of our imaginations” (03-24). He loves children’s minds as “gloriously unconstrained”. He values technology that lowers barriers. For panoramas “the barrier to entry to creating immersive content is ridiculously low” (02-11).

Technology that is useful, human-connected and serves a purpose. - He has “a philosophy as an educator of putting learning and student success first, and using the lowest level of tech necessary” (02-11). He distrusts “complicated and expensive technology solutions in search of an educational problem to solve”. - He values virtual work that extends the real world rather than cutting you off from it. - He frames the headset against ASU’s Charter: “how will this technology support what we aspire to achieve?” (image caption, 02-11).

Beneficial technology not being lost, and development staying open and accountable. In the Waymo post his concern is that “lazy narratives” and mob action will drive companies away. That would be “hardly” a win “for people and communities that could benefit”. It would also make development “more secretive, less open, and far less accountable” (02-18).

The people who carry the risk. - Neuralink (03-21): patients whose lives become tied to a company. - Embryos (04-14): parents forced into “a state of painful knowledge where ignorance might have been better”. Also the future child’s “sense of self and autonomy”, the “perceived “validity”” of hearing- and sight-impaired and neurodivergent people, and those “on the margins of society” priced out. He also values diversity as possibly linked to “an evolved resilience as a species”.

Respectful dialogue and diversity of people and thought. - Pizza makes the class a place to “share ideas and food as friends and colleagues — even where there are differences of opinion” (04-07). - On embryo screening he calls for “hard but respectful dialogue” that allows “opinions to be modified” (04-14). - In the Lerman and Keats conversation he protects vulnerable moments. Its end “should be listened to, respected, and pondered” (03-15).

What frustrates him. - Binary framings (03-31). - Convenient narratives that stroke “confirmation biases” (02-18). - Hype such as Musk’s “telepathy” (03-21). - Naive open letters (02-25). - Edutainment’s sugar-coating (03-17). - The attention economy. His conversations are “the antithesis of the sort of content that dominates today’s attention economy” (03-10). - Disciplines that each see only a piece of the puzzle (03-31).

3. Risk as a way of thinking#

The explicit risk vocabulary sits almost entirely in the Neuralink post. The underlying habits of mind run through the whole batch.

He brings back the 2019 analysis of the most likely “orphan risks” standing between Neuralink and its success. Three features show how the ideas work.

He also refuses to write off the “fake it ’till you make it” playbook. “I’m not begrudging this approach as it’s clearly worked in many cases — in part because people are incredibly good at taking innovative ideas and products finding new ways to use them.” And: “the reality distortion field is real, and I suspect that critics will discount it at their peril.” Narrative and perception count as real parts of the risk landscape, not as noise.

The language of navigation is everywhere, and it describes a stance, not a procedure:

Novel technologies need a new mindset#

Risks from the social response, not only from the technology#

The Waymo post (02-18) is an unusual piece of risk reasoning. The danger he traces is a chain: a “lazy” narrative legitimises mob action, the action spreads to “Anything autonomous”, and developers respond by making technology “more secretive” or even building machines “that can give as good as they get.” He labels this “a dystopian future-tech fever dream”, but it is a risk that runs through narrative and social dynamics, not through the device.

Irreversibility, unknowns and humility#

Across the batch, then, the risk concepts work as mental models for seeing what conventional analysis misses and for steering towards value. They are not procedures.

4. Scholarship and public writing#

The Substack is his workshop. - The Dune post was written while he should have been finishing the Jurimetrics review. “The Jurimetrics review I should have been completing while writing this will be out later this year” (03-03). - He prepares his review of The Creator inside the Vision Pro (02-11). - The Neuralink post takes a 2019 journal paper, a Risk Innovation Nexus method and a 2018 book chapter and tests them in public against the news (03-21).

Scholarship and public writing are not two outputs here. They feed each other in real time.

Thinking in public, including the unfinished. - The technology-problem post (03-31) is “part of my process of trying to marshal my thoughts” and “very much a work in progress”. It turns down the standard ending: “At this point a well considered and written article would wrap up with a compelling call to action. I’m sorry to disappoint, but I don’t have one.” - The Harari post ends by asking readers: “But this is just my perspective — what am I missing”. - Two weeks later he answers his own question in an afterword.

Experimenting with form. - Five conversations posted in one week, “rather audaciously (foolishly even)”, on the grounds that “There is method in my madness here”, since together they have “a coherence and meaning” that would be lost if spread out (03-10). - The pizza class run as a “natural experiment” (04-07). - A week of daily use as a way to review a device (02-11).

How he treats evidence. - He names the thinness of evidence: a “single datapoint” (02-18), a “dearth of hard evidence” on edutainment (03-17). - He calls for evidence that does not yet exist, such as “greater insight into both the short and long term impacts on patient mental health” (04-14). - He checks sources with their authors (Resnick). - He admits his own bias three times: “idiosyncratic biases” (02-11), “cognitive biases to the full” and “tinged with confirmation bias” (03-17). - He separates hype from novelty (Neuralink and the Utah Array).

Modesty outside his field, without silence. - He approaches pedagogy warily: “a field full of scholarship, ideas, personalities, egos, and passionately defended assumptions … one that I should probably stay well clear of. Yet … I’m not sure this is an option” (03-17). - On Sora, his wife, an elementary teacher, “reminds me that I am not an elementary school teacher”. His footnote reads: “I checked with her — thankfully I got the green light for this piece!” (03-24).

Transdisciplinarity as a necessity. - Each of the six Harari areas cuts across disciplines (03-10). - The technology-problem post calls for knowledge “that transcends conventional disciplines and areas of expertise at a scale that we’re rarely seen before” (03-31). - The conversation series puts physicists, astrobiologists, a humanities director, a photographer, a choreographer and an experimental philosopher in one room (03-10 to 03-15). - The Levin post ties developmental biology to “the nature of intelligence and even the emergence of artificial intelligence” (03-27).

Accessibility. - Plain, first-person prose, with humour: Dune obsessives will “hunt me down!” (03-03), and ChatGPT’s meltdown as possibly “another indicator of emergent of human like behavior” (02-25). - Explanations built on everyday questions, such as “How do the cells in a ridge on your fingertip … know where they’re supposed to go” (03-27). - Embedded videos, and confessions (“I must confess” recurs across the batch).

Teaching as part of the scholarship. The postscript to the playgrounds post traces a line from his physics labs, through his research, to his classes. The teaching spaces (a sci-fi movie course, “Pizza and a Slice of Future”) are where his ideas about play and the future are tested with students (03-17, 04-07).

5. His role as he sees it#

A student of technology and society, whose job is to explore, not to rule on outcomes. He writes “As someone who studies and writes about the complex relationships between technology and society”. He accepts that “done right” is contestable and that “it may be that the world is indeed not ready for autonomous vehicles … I get that — it’s what I do for a living” (02-18). The same stance shows in the Harari post, where thinking about the future is “an occupational hazard of being an educator and someone who studies the intersection of technology, society and the future” (03-10).

A “pilot”, or a builder of pilots. He takes up a colleague’s term: “We need pilots — whether they are institutions, communities, or individuals”. He calls for “a new breed of institutions, thought leaders, scholars, graduates” (03-31). The Future of Being Human initiative, the pizza class and the playgrounds idea are his practical answers (03-17, 04-07).

A convener and host. In the conversation series he makes the room and then steps back. He modestly counts himself out of the “stellar cast” (“maybe myself excepted”). He is “my own worst critic when it comes to media with me in it” (03-14). He is candid about his nerves in the class (“I nearly had a panic attack … I had visions of a student rebellion!”) and ends up “humbled and inspired by every week this semester” (04-07).

With readers. - He treats them as fellow thinkers. He asks what he is missing, apologises for having no call to action, warns them of “challengingly “interesting” stuff” (03-14), and shares his irritations and delights. - He asks them to act once, over the deepfake letter, and explains why that was unusual for him (02-25).

With industry. He is fair and gives credit, while watching for incentives. - “I feel some sympathy here for Google … good on Google” for responding quickly (02-25). - OpenAI is “very sensibly being diligent” in stress-testing Sora (03-24). - On Neuralink: “I must confess that I’m excited by what Neuralink is achieving” (03-21). - On Orchid: “I also respect what Siddiqui and Orchid are trying to do” (04-14). - Beside that credit he tracks business-model drift: subscriptions for implants, rich clients paying for non-health screening, and “competitors using similar technology” (03-21, 04-14).

With other experts. - He disagrees with Harari constructively. “Harari certainly has a point”, and Harari’s comment “did challenge me to see if I could boil my thinking down” (03-10). - He praises Davies and Walker as “generous and humble” (03-11). - He is sharp only with peers whose narratives do harm. “writers and academics” who spin the backlash story “are legitimizing such behavior” (02-18). - He criticises his own fields, STS, ethics and responsible innovation, for lacking integration (03-31).

What he refuses to do. - Polarise. He will not answer the optimist/pessimist question on its terms (03-31). He will not choose between “demonizing” Orchid and “raising them, god-like, on a pedestal” (04-14). - Take lazy critiques in either direction. Dismissing Orchid as ““dangerous Silicon Valley tech-bro fantasy.”” would be “lazy and narrow minded” (04-14). The Waymo backlash narrative is “both lazy and dangerous” (02-18). - Preach. He has no ready answer or blueprint (03-31). - Hype. Musk’s “telepathy” is “the type of hyperbole most serious medical professionals and research scientists would avoid like the plague” (03-21). - Excuse violence as protest. “It’s one thing to be critical of the introduction of self-driving cars. It’s another thing entirely to embrace the willful destruction of something” (02-18).

Changes of mind and stance, all made in public. 1. Vision Pro (02-11). Within a week his interest moved from entertainment to productivity, and his concern about a “societal crisis” eased. 2. Deepfakes (02-25). He moved from trusting common sense to signing a letter he would normally avoid (“I tend not to sign open letters like this as they’re often deeply naive … But in this case I did”). 3. “technology apologetics” (03-31). He is reconsidering a decades-long professional habit: “one that intentionally seeks to support others where they are at in their thinking. But I’m beginning to question its wisdom”. This is a significant shift. He moves from accommodating colleagues who see technology as separate from us towards challenging them. 4. The Harari framework (03-31 afterword). He revises his own list three weeks after writing it.

6. What is distinctive#

  1. Technology treated as part of what humans are, not something added. The oxygen analogy (03-31) steps outside the optimist/pessimist axis that structures most AI debate. So does the claim that no global problem makes sense “outside the context of thousands of years of an evolving relationship between humanity and technology”. He argues that treating technology as “something we do and not something we are” is itself the risk. He also refuses to romanticise rejecting technology: Dune’s machine-free society is “deeply flawed” (03-03).
  2. An AI worry built around imitation, not intelligence. Through Herbert’s “counterfeit” line he moves the AI question away from arguments over “intelligence” and “consciousness”, which he calls “misnomers”. The question becomes the blurring of “what defines us as humans” and what machines imitate (03-03). The Levin post widens it further, since intelligence and agency may be properties of cell collectives (03-27).
  3. Orphan-risk mapping in the service of responsible success. He uses the risk landscape to help a company he finds both exciting and worrying see what it would otherwise miss. He checks the map against events five years later, and names a new orphan risk, company failure and stranded implants, that neither hype nor conventional safety review would raise (03-21).
  4. Risks that come from the reaction to a technology. The Waymo post (02-18) traces how critical narratives and mob action could make technology less accountable and more defensive. Most commentary does not look at this second-order dynamic.
  5. Generative AI as a window into children’s inner worlds. Where most commentary on Sora stopped at deepfakes, he asked whether it could let a four-year-old show what is in their head, and let adults see it. He keeps guardrails and safety as conditions throughout (03-24). This is his play and creativity lens applied to AI.
  6. Designing for serendipity as institutional practice. Couches, pizza, strangers paired on purpose, norms of “catalytic serendipity”, and outcomes not measured to death. He treats the conditions for unexpected learning as something to design, and he sees that as the answer to preparing people for futures no one can predict: “how to think in the future, not just what to build” (03-17, 04-07, 03-10 to 03-15).
  7. Holding respect and worry together. “I worry about the social and ethical risks. A lot. But I also respect what Siddiqui and Orchid are trying to do” (04-14). He locates the ethical problem in the burden of choice and in knowledge that cannot be unknown, rather than in the line between can and should.
  8. Criticising his own habits and fields in public. He questions his decades of “technology apologetics” and finds responsible innovation and STS too fragmented (03-31). Few scholars do this in front of a general readership.

7. Posts that best reveal how he thinks#

  1. 2024-03-31 we-have-a-technology-problem-and. The oxygen reframing, technology as part of what humans are, the critique of his own fields and of his own “technology apologetics”, navigation and pilots, and thinking in public with no call to action.
  2. 2024-04-14 welcome-to-the-age-of-swipe-and-select-embryos. Stating the other side’s case before judging, holding respect and worry together, the ethics of choice, irreversible knowledge, the question of who decides, plausibility tests, and a call for dialogue in which views can change.
  3. 2024-03-21 elon-musks-neuralink-plays-mind-games. The risk landscape and orphan risks at work as a lens for noticing, tested against events and extended to business-model risk. It also shows fairness to the Silicon Valley playbook.
  4. 2024-03-17 undergraduate-playgrounds-not-playpens. His research method as serendipity plus checking, open admission of confirmation bias, and play as the root of his own thinking: “So much of how I explore new ideas … is grounded in play.”
  5. 2024-03-24 could-openais-sora-be-big-deal-for-kids. Imagination and what if applied to generative AI, analogy by structure (pencils, KidPix, Minecraft), speculation kept honest, and modesty outside his field.
  6. 2024-02-11 one-week-on-with-the-apple-vision. Finding out by doing, his own experience and students’ reactions as data, and a change of mind recorded within a week.

The next two also say a good deal. 2024-02-18 setting-fire-to-self-driving-cars-is-bad shows second-order risk reasoning and a refusal of convenient narratives. 2024-04-07 multigenerational-learning-tech-future shows designing conditions for serendipitous learning.