B32 perspective notes: 2026-09-24 (1 post)#

These notes read the batch for how Maynard thinks, not for the concepts he names. I read the one post in full: introduction, lecture body, postscript, Claude’s process account and all nine footnotes. Quotes are exact, including curly punctuation. Markdown emphasis is dropped.

Evidence rules applied

2026-09-24 being-an-academic-in-an-age-of-ai is an edited version of a lecture he gave at King’s College London on 8 September 2026. The parts carry different weight as evidence.

His own written prose (most secure).
The six-paragraph introduction, from “A couple of weeks ago” to the disclaimer.
The first two paragraphs of the Postscript.
The bracketed aside “[AM — Claude read my handwriting better than I can!]”.
Lecture body and footnotes (his spoken words, AI-smoothed).
The body is his spoken lecture. He corrected the transcript against the video. Opus 5.5 turned it into an article under his rule, as Claude reports it: “Include nothing he didn’t say, cover or think about”. He then “line-edited the result against the original transcript”.
The ideas, examples, structure and most of the wording are his. Some sentence-level smoothing is the model’s, so I treat exact phrasing with mild caution.
The footnotes record his answers in the Q&A plus later clarifications. They have the same status as the body.
[^7] (“Actually, I suspect I use it more than I realize”) and [^3] read as his own later additions.
[^4] and [^5] are identical, an editing glitch, so I cite the text once, as [^4].
Not evidence of his views.
The block-quoted process account, written by Claude.
The Midjourney image.
I use Claude’s account only to describe how the post was made and the role he chose: setting rules, correcting the transcript, “a dozen or so decisions”, pushing back when drafts “didn’t sound like him”.
Earlier posts used only for context, and labelled as such:
2025-03-15 ai-playgrounds-in-higher-education, whose “categorical error” wording I checked;
2026-01-10 is-ai-a-cognitive-trojan-horse (epistemic vigilance);
2026-07-16 orphan-risks-frontier-ai-maynard.

Context. The talk was given in September 2026, at the invitation of King’s Vice-Chancellor Shitij Kapur, to an audience of academics. He was “asked to be provocative”. The title was Paradise desecrated: Is AI destroying the university of our dreams? And if it is, would that be such a bad thing? The post appeared in the week Anthropic released Opus 5.5.

1. How he thinks here#

He starts without knowing where the argument will land, and says so#

He does not begin with a thesis. He begins with a book and an admission: “I must confess that I wasn’t entirely sure where the narrative was going either (although I had an idea).” He adds: “I thought I’d start with a short anecdote — although even now, I’m still not entirely sure how it fits in. We’ll see at the end whether it does.”

His introduction confirms that the talk was “given without slides and with minimal notes”. The thinking happens in the telling. He trusts the anecdote (Bradbury’s Eating People Is Wrong, recommended by his O-level teacher in 1981 and read “40-odd years” later) to earn its place as the talk unfolds. It does: the reluctant cannibal who says “eating people is wrong” while “nothing ever changes” becomes the frame for academic inaction and returns in the crab-bucket close.

He treats the not-knowing as part of the subject matter: “that in itself is part of the challenge of the times we’re facing: there is so much uncertainty, so much novelty, so much speed”. He makes the uncertainty visible because it is what he is talking about, not a slip in his preparation.

He tells the audience where he stands before he argues#

“But first, you should know a little about where I’m coming from, just so you can calibrate.” He then sets out his position in unusually candid terms: - “What I see scares the life out of me sometimes.” - He sorts his own fear into kinds: “Some of this is rational, some of it is irrational; some of it, I suspect, is justified, and some of it probably isn’t.” - He holds both sides: “one of the scariest things I’ve ever seen in a career grappling with some of the most advanced technologies we’ve had — and, at the same time, that the potential is profound.” - “I do have really, really bad days with this technology, so bear that in mind”.

This is a scientist’s habit (declare your instrument’s biases) applied to his own feelings. Emotion is admitted as data, labelled and discounted where needed. It is not used as a lever on the audience.

He goes to the foundations: what a university really trades in#

He will not start from AI. “I want to start, though, by stepping back from AI”. He lays out the stories universities and academics tell: the public story (“pillar of society”), the private story (“maintaining a position… And then we spin a story on top of this”) and the academic’s self-story (“We are the saviors of future society”). He refuses to rank them: “all of these stories are right in one way or another. There are no wrong and right answers here.”

Then he strips them back to a structural claim: “We are part of a scarcity economy and a scarcity model, except that what we trade on is intelligence.” He flags it at once as a model to be tested, not a verdict: “a big, bold statement, and there are many ways of pulling it apart”.

The move is a first-principles reduction that makes the AI question sharp. What happens when “that scarcity model ends up as a model of abundance”? The threat he names is to identity, “an existential threat to what we think we are”, and he immediately holds open the possibility that it is not real (“it could be that all of this stuff… is mere hype”).

He insists on precision about what the thing is#

Before judging the threat, he says you must know “exactly what we’re talking about — as far as we can, in an uncertain world”. He criticises those who “bandy around the letters “AI”… with no grounding”.

His account of the technology is historical and mechanistic: from the 1950s, “human intelligence” in inverted commas “because we’re not quite sure what that means”, then next-word prediction scaling to “the next paragraph, the next page, the next book”. He picks out the interface as something “not to undervalue”. This is the physicist’s instinct to find what has actually changed before theorising about its effects.

Finding the new thing by structure: language is formative#

His analysis turns on one structural insight, not on a capability benchmark. Language “changes how we understand ourselves, others, and the world we live in”; “Language is formative… And now we had a technology that was actively taking part in the formation process.”

He hedges the premise (“This is somewhat controversial (there are a number of theories here)”) and then follows where it leads: - relationships with AI; - the split between “intellectually” knowing it is a machine and “emotionally and cognitively” treating it as human; - the ability to “alter, potentially, how we think, believe and act”; - the risk that AI uses “the medium of formation” to “slip beyond our cognitive defenses”.

Later he uses the same premise to give a non-AGI account of loss of control: “We’ve given AI the ability to use language as a lever”. One well-chosen principle carries the whole risk analysis.

He reads stories as archetypes#

For the claim that AI is “not just a tool”, he turns to stories: “How many books have you read, and how many movies have you watched, where somebody is given the opportunity to wield a technology that gives them incredible power — and yet the price is always that they have to be willing to be changed by the technology to do it?” He names Lord of the Rings and the Tesseract. He adds: “I find it hard to find any story in history where somebody is given the opportunity to wield unbelievable power that they don’t fully understand, and yet isn’t changed by doing so.”

The analogy works by structure: power bought at the price of being changed. It is not decoration. He says why he reaches for stories in [^8]: creativity “means being willing to be influenced or inspired by unlikely sources. It’s one of the reasons I used Terry Pratchett.” Using stories is part of his stated method.

He reframes by “flipping the lens”, twice#

The talk turns on two reversals, which he names as such: 1. From institution to society. “I’d argue, though, that you can flip this around in an interesting way. Instead of asking what the threats of artificial intelligence are to the institution or the academic, ask what the threats (or the consequences) are to society, because then the conversation both opens up and goes in a completely different direction.” The reframing is valued because it opens the conversation, not because it gives the right answer. 2. From self-preservation to service. “But things look very different if we flip that lens, and ask how we use the skills and the insights… to help navigate to a future that is vibrant”.

In both cases the same facts, looked at from a different vantage point, show different possibilities. This is his usual way of getting unstuck.

Working assumptions stated as such#

“Here I’m making the assumption (and it may be a flawed assumption) that powerful AI is inevitable… But run with this for the moment”. “We can’t run away from it. We can’t stop it. We can’t pause it.” This is scenario reasoning: a premise adopted so that he can think, flagged as possibly wrong. It is not a prediction. The same pattern appears with the scarcity thesis.

Holding tensions without resolving them early#

The post holds many unresolved pairs: - fear and “profound” potential; - “a lot of justification” for moral opposition to AI, alongside doubt that opposition “is going to work”; - leaning in as “very enticing” but “problematic”; - a crisis that is “two-edged”; - “I have days when I’m not optimistic… I’m not sure we’re going to make it”, alongside “a high chance of making it if we have organizations step up to the plate”; - “paradise” as a word he hates but uses (“I hate that word, but it works here”).

He resolves none of these by picking a side. They are the ground on which navigation happens.

Building to find out#

His own framing of the postscript is experimental: “Curious as to how good the model was, I decided to set it the challenge”. The introduction generalises this: “because my writing is never just writing”. Even a lecture write-up becomes a test of a new capability, carried out in public, with his verdict attached (see section 4).

Where curiosity, play, creativity and serendipity drive the thinking#

Serendipity in his reading.
A summer reading of Bradbury, prompted by a four-decade-old recommendation, generates the talk’s frame.
A Pratchett reread supplies the ending: “another book that I’m rereading at the moment”.
The ideas come from what he happens to be reading, not from the literature on the topic.
Joy and play named as the core of academic value. “It is the joy of playing around and serendipitously discovering something. There is a joy there. And that is something that can be harnessed.” In a talk about AI threats, he makes joy, play and serendipity the resource universities bring to the AI transition. The self-correcting footnote on “Joy is a word I don’t use often” ([^7]: “Actually, I suspect I use it more than I realize”) is a small, telling moment of self-knowledge.
Creativity as openness to unlikely sources ([^8]): “you cannot find those sparks of genius solutions, or partial solutions, to problems unless you have creativity.”
Humour and lightness keep the provocation humane.
“if you can feel your hackles getting up on your neck, run with that”;
“if anybody has seen an example of a government that moves really fast and really smartly, please do let me know (and I’ll tell you another story)”;
“my increasingly unreliable memory”;
the 😊 on the LLM-summary suggestion;
“[AM — Claude read my handwriting better than I can!]”.

2. What matters to him#

Human identity and formation. At bottom, what he wants to protect is people’s “sense of self, your sense of your relationship with others, your sense of your relationship with the world”. He wants to protect it for society as well as for academics. The “new paradise” is helping people “retain their humanity, their sense of purpose, their sense of self and their sense of belonging, within a future that is dominated by advanced AI”. The threat he worries about most is to who we are, more than to what we have.
Thinking for ourselves. Cognitive surrender troubles him because it “feels so good… it fools you, when all the time you’re beginning to lose those cognitive abilities” ([^6]). So does the “easy button”. What frustrates him is a harm that feels like a benefit from the inside.
Reading and language as embodied and relational. “The process of reading is a very human process. It’s an embedded process… we are biological beings in the world… And an AI knows nothing about any of that.” ([^1]). He shares a writer’s worry that AI will change how humans use language, and admits his hope that it won’t is “not… well-founded”.
Equity. “what does it mean when we say it’s free — because it isn’t? Somebody is paying somewhere.” ([^2]). He wants benefits shared “in an equitable way” and admits “I don’t think there’s any clear way forward here yet.”
Service over self-preservation. The moral centre of the talk is service. What is needed is “changing our perspective on what it means to be of service to society, rather than service to our organization or ourselves” ([^9]). Academics must stop “looking at ourselves”.
The long view of education. His own O-level teacher’s recommendation took four decades to bear fruit: “it isn’t immediate impacts that are important; sometimes these things take decades to percolate through”. This quietly undercuts metrics culture.
What frustrates him:
institutional self-storytelling;
the crab bucket;
ASU’s rhetoric (“Be the person you can be… Do not be tied down by academic norms”) contradicted in practice by P&T files asking “What is your h-index?”;
going fast with “a technology that we don’t understand”;
race logic (“because if we don’t go fast, somebody else will”);
speculation with “no nuance… and no humility”.
What delights him:
“the joy of discovery”;
people from “incredibly different ways of understanding and knowing the world” coming together for “combinatorial advances” and “sparks of creativity”;
satirical fiction (Pratchett “tells you more than you would ever want to know about the academy”);
the capabilities themselves (“I must confess that I was impressed”).

3. Risk as a way of thinking#

The word “risk” appears only once in the post (“the risk of over-reliance”). “Manage” does not appear at all. “Navigate” and “navigating” appear ten times. The risk thinking is in the structure of the argument more than in the vocabulary.

Understanding risk in order to reach the benefits. This is his clearest statement of method in the batch: “I’m going to talk about threats here — not because I think there are only threats associated with AI, but because you can only begin to realize the benefits of a technology if you understand what can possibly go wrong, so that you can navigate around it. This is fundamental to developing and using emerging technologies beneficially: you’ve got to understand what they can do that’s harmful, so you can avoid it or flip it, and so get to the good.”
Risk is a map of the ground between the present and a valued future.
“Avoid it or flip it” is the risk-innovation instinct: a threat can sometimes be turned into an opportunity, not only reduced.
It works as a mental model, not a procedure. No assessment method is proposed.
Threat to value, in substance. The threats he names are threats to things people value and to who they are, not physical hazards:
the academic’s “currency” and “identity”;
the university’s “business models and the social contract”;
society’s “sense of self”, purpose and belonging;
“societal systems… even down to our own identity”;
“things that have been central to being human, and to human society, for the last 20,000-plus years”.

The “existential threat” he names is “to what we think we are”. This is risk as a threat to value without the label. - Navigating, not managing. Navigation language runs through the talk: - “navigate advanced technology transitions to get to the sort of future we want”; - “helping society navigate the AI transition” (his introduction); - “navigating this AI transition to a future of human flourishing”; - “the ability to see the potential pathway between where we are at the moment as a society and where we might be”; - “What those ways are, nobody knows yet. We are in uncharted territory.”

Navigation here means orienting and finding pathways under deep uncertainty, not controlling a known hazard. He describes his own work in these terms: “this is a lot of what I do, thinking about how you navigate to these futures”. - A novel technology needs a new mindset. Three passages carry the claim: - Past frameworks mislead. “as soon as we start evaluating it within past frameworks, we make categorical errors”. He applies this even to the ethics of opposing AI. - This technology is different in kind. It is unlike “(I would argue) any other technology in human history” because it “interacts with our understanding of who we are”; it is “not just a tool — unless you consider a tool as something that changes who you are.” - Recalibration at every level. He reports that people at the frontier say “we do not even have the frameworks to begin to formulate the questions we need”, and concludes “we have got to completely recalibrate — as individuals, as communities, as a society”. - Context: the “categorical error” wording repeats 2025-03-15 ai-playgrounds-in-higher-education, where treating AI “as a leaning [sic] aid” was the error. So this is a settled part of his thinking, not a one-off. - Orphan-type risks, unnamed. The risks he stresses fall outside conventional risk frameworks and have no institutional owner: - formation through language; - bypassing epistemic vigilance; - cognitive surrender; - AI using “humans as another cog”.

His governance-gap analysis (companies, governments, civil society and publics each unable to lead) is a structural account of risks nobody is positioned to own. But he does not use the term “orphan risks” here, two months after his orphan-risks paper (context: 2026-07-16 orphan-risks-frontier-ai-maynard). The concept is at work in the argument, not in the vocabulary. - Humility against false confidence ([^4]). When a “self-described techno-optimist” asks for empirical observation over speculation, he rejects both extremes: - AGI and singularity speculation is “incredibly blinkered and naive… there’s no nuance there and no humility”; - “There’s nothing new under the sun” is “not evidence-based either. It’s speculation, and it’s dangerous as well.”

His middle way: “when the technology changes faster than we can generate data, you’ve got to have some degree of informed speculation, and some degree of imagination… don’t disallow speculation, but do it within a context of humility — knowing that it’s speculation, not reality; looking at possible futures rather than real futures; acknowledging that you need data to follow through; and bringing in different voices.”

This is the quantitative-foundations point in miniature. Data remains the goal (“you’ve got to have empirical data at some point”). Imagination bridges the gap while data cannot keep up. Humility guards against mistaking the bridge for the destination. - Capability, not AGI. “I am not talking about AGI… All of those might happen. But I think they’re irrelevant to this conversation.” He puts the risk in present capability: solving problems “unsolvable to humans”, pulling in resources “inaccessible to humans”, “using human behavior”. This is a risk thinker’s refusal to let dramatic but speculative scenarios crowd out plausible, nearer ones.

4. Scholarship and public writing#

A lecture that becomes a post, which is also an experiment. His introduction makes the three layers explicit. The post is a record of the talk; “a record of them for my increasingly unreliable memory”; and a capability test (“my writing is never just writing”). The post’s form is itself research into how AI can be used well.
An honest verdict on the experiment, and a demanding standard. “It was good enough for me to think it worth posting, although I’m not sure that it’s substantially better than if I’d taken the same amount of time to edit the transcript manually… And I must confess that I was impressed. It’s also a great example of how AI used well doesn’t necessarily make things faster if you’re going for quality, but it can allow you to achieve more with the time you have.”
He gives a mixed verdict rather than hype or dismissal.
He makes a finding (quality does not come faster, but more becomes possible) and generalises from it carefully.
Transparency about process as an ethic.
He asks Claude to document every step, including “more than 40 agents” and “eight and a half million tokens”. That wording is Claude’s, but the choice to publish it is his.
He adds his own bracketed interjection.
He discloses the risk of error: “There may also be slight inaccuracies… I think that most of these have been caught. But it’s worth the disclaimer, just in case.”
The process account shows him keeping the judgement for himself (“the judgement”, “pushed back when a draft didn’t sound like him”), which matches his worry about cognitive surrender.
Faithfulness over polish. “The article is unapologetically long as I wanted to make sure the ideas I explored and the points I made were captured as faithfully as possible.” He also kept the spoken idiosyncrasies: “I left most of these in.” He chose fidelity to what he actually thought over a tidier, shorter essay.
Readers as LLM users. “do feel free to drop it into your favorite LLM for a summary 😊”. He points to the “LLM-readable text version”. He expects, and designs for, readers who read through machines, which is itself an experiment in how scholarship reaches people.
Lived institutional evidence. His evidence for the crab bucket is three years on ASU’s university-level P&T committee, two as chair, “about 120 cases a year”. This is a particular kind of evidence, drawn from his own role inside the system and stated with its limits. He uses it to show a gap between rhetoric and practice, not to claim a statistic.
How he treats expertise.
He credits other researchers (“King’s has leading researchers focusing on AI psychosis”).
He reports frontier practitioners’ views (“Most people working at the cutting edge of AI will claim…”).
He calls for “really serious research” on cognitive surrender rather than asserting its extent ([^6]).
He is careful about the edges of his claims (“somewhat controversial (there are a number of theories here)”).
Transdisciplinarity as the university’s value. The case for universities rests on people from “incredibly different ways of understanding and knowing the world — whether you call them disciplines, areas of expertise or something else” coming together. He sidesteps the word “discipline” while describing the thing it names. Fiction sits alongside technical history and economics in the argument.
Accessibility. The register is conversational (“So we’ll see how this goes”; “Think about that.”), the analogies are drawn from popular culture, and the jargon is glossed (“our epistemic vigilance”; “(transformer systems)”).

5. His role as he sees it#

Deep practitioner and critic at once. “I spend a lot of time diving deep into frontier AI models… It dominates my life more than I’d like to admit”. Yet he says: “I’m not a strong AI optimist or advocate.” He keeps his authority to judge by using the technology intensively while refusing to cheerlead. He uses Anthropic’s model to produce the post while saying that “the Anthropics and the OpenAIs and the Xs” should not “decide for humanity”. He does not treat this as a contradiction, and it is not one: close engagement is how he earns the right to criticise.
Scholar of transitions, not of AI alone. He places his AI work within “my broader work asking big questions about how we navigate advanced technology transitions to get to the sort of future we want — and what it will mean to be human in those futures.” AI is the current case of a longer question.
The academic as navigator for society. He argues that at present only universities can fill the gap between technically capable companies, slow governments, under-resourced civil society and publics who are “critically important” but cannot be handed the problem. The university is “an accelerator and a catalyst for what other organizations can do”. This is his own model of the public scholar, made institutional: helping others “see how we can build this”, not deciding for them.
He includes himself in the critique. “We all do this. I do this. I think about what it’s going to take to keep myself funded, to keep myself in a job, to keep myself relevant.” “I raise this not to criticize universities and academics, because we’re all struggling here”. He criticises from inside, including his own institution (“my own, Arizona State University, where we have proclaimed that we are an AI university”).
Provocateur by invitation, not by temperament. He mentions the brief three times (“I was asked to be provocative”; “that was the brief I was given”; “Remember, I was told to be provocative here”). Provocation is treated as an assigned mode. The flags let the audience separate provocation from settled judgement.
How he treats people who disagree.
AI refusers: he gives their position real credit (“my sense is there’s a lot of justification for that”) before questioning whether it can hold in an AI-saturated world.
The techno-optimist questioner: he answers on the substance, and criticises hype and dismissal equally.
The writer in the audience: “I have exactly the same worry.”
Students. His students came to him “very excitedly” with early OpenAI APIs. That memory leads straight to his clearest change of mind (below).
What he refuses to do:
Polarise: “neither an AI optimist nor an AI pessimist”.
Fear-monger: he admits fear but discounts it openly (“some of it probably isn’t” justified).
Choose a story: he won’t adopt the AGI story or the “nothing new” story.
Call AI a mere tool: “we’re kidding ourselves”.
Hand the future to companies: “I hope not. I don’t think we can.”
Give the fatalist answer: “Maybe it’s the end of universities” is “a very dangerous one”.
Blame individuals: he criticises culture, not colleagues.
Be settled optimism’s advocate: “where my hope lies” is conditional on organisations stepping up.
Changes of mind.
Explicit and self-deprecating: “I remember looking at them and thinking, “It’s interesting. It’s a toy. It’ll never catch on.” I was wrong.”
A self-correction in a footnote about his own vocabulary ([^7]).
A shift in emphasis: from AI in education as a question of cheating (“That is important”) to “a problem far bigger”: the university’s identity and role.
A darker register than in many earlier posts (“really, really bad days”), kept apart from his analysis rather than driving it.
[^3] marks a change in the world between the talk and the post (“companies asking regulators to rein them in”).

6. What is distinctive#

Identity and formation as the core of AI risk. Most AI-and-universities talks focus on cheating, assessment or productivity. Most AI-risk talk focuses on AGI, misuse or bias. He puts AI’s newness in its role in human formation through language, and puts the main threat in identity: of institutions, vocations and people. “This is not just a tool — unless you consider a tool as something that changes who you are” compresses a position that few commentators in either camp take.
An economic analysis of academia, then turned outward. Treating universities as traders in scarce intelligence, whose identity and social contract depend on that scarcity, is an unusually blunt and self-implicating diagnosis for an academic insider. It gives AI’s threat a precise mechanism: abundance undoing scarcity. He then flips it into a claim about society’s need.
Non-AGI loss of control through language. “We’ve given AI the ability to use language as a lever”; “humans are just another cog in the works”. He builds a capability-based, near-term account of AI using people, with no need for sentience or superintelligence, and ties it to his account of formation. It rejects AGI framing without playing the risk down.
Humility as method ([^4]). The “informed speculation… within a context of humility” answer rejects both hype and “nothing new” dismissal as unevidenced speculation. It gives a four-part discipline: know it is speculation, look at possible futures, follow with data, bring in different voices. This is the stance of a risk scientist who knows the limits of data, not of a futurist or an empiricist alone.
Joy and play as policy-relevant resources. He argues that the joy of discovery, serendipity and the freedom to ask unasked questions are what fit universities to help society through the AI transition. This is rare in AI-governance discussion, and it follows from the principles behind his own initiative.
Fiction as serious evidence of human patterns. Bradbury, Flanders and Swann, Pratchett’s crab bucket, Tolkien’s Ring and the Tesseract are used as archetypes that carry structural insight. He explains why: creativity needs “unlikely sources”.
Criticising his own institution in public, from a position of authority. He is a P&T chair describing the “academic claws” in his own university’s files, and he critiques ASU’s “AI university” stance, in a talk at another university.
Showing his AI practice while critiquing it. He publishes an AI-assisted text with a full process account and his own verdict (quality is not faster, but more is possible), while arguing that AI threatens thinking for oneself. The demonstration puts the answer in practice: keep the judgement human, be transparent, check faithfulness.

7. The posts that best reveal how he thinks#

Only one post is in this batch, and it is a strong one for this purpose.

2026-09-24 being-an-academic-in-an-age-of-ai. It reveals, in one place: - the improvised, anecdote-first opening; - the declaration of his position and feelings “so you can calibrate”; - first-principles reduction (the scarcity model) and “flip the lens” reframing; - “language is formative” as the structural insight behind his risk analysis; - risk as understanding what can go wrong “so that you can navigate around it… and so get to the good”; - the claim that past frameworks produce “categorical errors” for a technology unlike any before; - joy, play and serendipity as the university’s distinctive contribution; - fiction as a thinking tool; - humility-bound informed speculation; - self-inclusive institutional critique; - a public, honestly evaluated AI experiment; - an explicit change of mind (“I was wrong”).

The most secure evidence of his own voice is the introduction and postscript. The lecture body and footnotes are the richest record of his reasoning.