B20 perspective notes: 2024-11-06 to 2025-01-12 (18 posts)#

These notes read the batch for how Maynard thinks, not for the concepts he names. I read all eighteen posts in full. The five Modem Futura episode posts were read for his own framing lines only; the episodes themselves were not listened to. Quotes are exact, including his typos, apart from markdown italics, which are dropped.

Evidence rules applied

• 2024-12-15 advanced-ai-challenges-opportunities-native-american-communities is mostly three guest reflections (Al Kuslikis, Sean Dudley, Leonard Bruce). None of that counts. Only his introduction does, and what he did: he convened the piece and handed the space to others.
• 2024-12-17 navigating-the-challenges-and-opportunities-of-advanced-biopreservation-technologies quotes the guest editors’ introduction and a multi-author framing paper (“we further re-enforce this”). Those lines are collective text. The underlying risk-innovation paper is co-authored (Maynard, Oye, Scragg, Tripp, Wolf, JLME 52(3), 2024). The evidence is his post prose and his own narration of the slides from his October 2024 talk.
• 2024-11-17 navigating-the-ethical-dilemmas-of-brain-computer-interfaces quotes long passages from Gordon and Seth. Those are theirs. The evidence is what he picks out, what he extends, where he pushes back, and his footnotes. His 2019 paper with Scragg is mentioned but not quoted.
• 2024-12-01 geoengineering-aerosol-monitoring-john-aitken quotes his own 2015 Nature Nanotechnology piece. That piece is his own sole-authored prose, so it counts, dated 2015.
• 2024-12-22 why-modem-futura-is-more-than-just-another-tech-podcast quotes text he says he wrote for the initiative’s website in 2022. That counts. It also quotes an answer from an ASU News interview that he calls “our response”, shared with Sean Leahy, although it is written in the first person singular. I use it with that caveat, as something he endorses (“This, to me, gets to the heart”).
• 2025-01-12 chatgpt-faq-education: the FAQ document itself was produced with ChatGPT and is not evidence. His account of how he made it is.
• 2024-11-12 waymo-we-go: the episode summary reads like standard show notes. Only his opening framing line is used.
• 2024-12-29 fantasy-top-ten-lists-2025: the Sora header video is ignored. The footnotes are his and carry most of the substance.
• Embedded posts in 2025-01-05 five-voices-five-pieces are other writers’ work. His own descriptions of why he chose them count.
• Midjourney images are ignored. One caption (2024-12-08) is his own prose about choosing an image and is used lightly.

Context. The batch opens the morning after Trump’s 2024 election win and runs to mid-January 2025. OpenAI’s “12 days” release o1 in full, Sora to the public and Canvas, then announce o3. Google ships Gemini 2.0 and Apple Intelligence rolls out. Altman’s Reflections predicts AI agents will “join the workforce” in 2025. Maynard is six weeks into co-hosting Modem Futura with Sean Leahy and records the tenth episode before a live audience. The ATP-Bio ethics collection comes out in JLME. He is fielding colleagues’ questions about students and AI. Three posts form a chain built across a fortnight: the MIT materials-science study and scientists’ joy (11-10), a reader’s LinkedIn reply to it, and the amanuensis thought experiment that reply set off (11-24). The chain resurfaces in the fantasy list’s footnote on joy (12-29) and, inverted, in the FAQ post (01-12).

1. How he thinks here#

He opens from something that happened to him or landed in front of him#

Nearly every substantive post starts from an encounter, not a thesis:

• 2024-11-06 ai-in-a-world-of-trump: re-listening to an episode recorded before the election: “it was with some anxiety that I re-listened to it this morning.”
• 2024-11-10 is-ai-poised-to-suck-the-soul-out-of-science: a friend’s Substack post. “I was quite shocked”, and it “disturbed me more than I was expecting”.
• 2024-11-24 artificial-intelligence-agency-human-amanuensis: a reader’s reply on LinkedIn: “and so began the seeds of the thought experiment below!”
• 2024-12-01 geoengineering-aerosol-monitoring-john-aitken: a New York Times article about stratospheric balloons, “and this is where the aerosol physicist in me kicked in”.
• 2024-12-13 are-educators-falling-behind-the-ai-curve: a recurring conversation: “It’s a common occurrence: A colleague asks me about students using artificial intelligence in class”.
• 2024-12-20 sora-has-a-bias-problem: “Playing around with the platform, which was only released to the public a few days ago”.
• 2025-01-12 chatgpt-faq-education: his college links to an old FAQ of his, and he is “stung into action by the guilt of imagining colleagues using a long-outdated document.”

He is also candid about the emotional trigger. Shock, anxiety, guilt and delight are all named as the reason a piece exists.

He builds simple models to think with, and says where they break#

Three posts build deliberately small conceptual models in public and then test them against reality.

• The amanuensis model (2024-11-24). He starts with a two-actor spectrum from AI as tool to AI as assistant (“In very simple terms (remember, this is a thought experiment)”). He then extends it to AI as amanuensis. Then he notices that the model is unrealistic, because “very few people have complete control over what they do”. So he adds a third actor, a “primary agent”, and watches what happens when that agent “switches to working directly with the AI”. The human becomes the amanuensis. He then lets AIs be primary agents and chains the blocks into networks. He is open about what this costs: “Limiting the model to just three actors is a gross oversimplification”, chosen because it lets “multi-way influences to be explored on the smallest possible scale, while also being scaleable”. Later: “This is where the simplicity of the model begins to hit its limits.” He flags the resemblance to Actor Network Theory in a footnote: “much, much more simplistic.”
• The three s-curves (2024-12-13). He moves from one curve (capability) to two (capability, then a lagging “utilization” curve) to three (a “perception curve” lagging both). The argument arrives as the figures accumulate. The insight is that the gap educators face is not in the technology but in the lag between what it can do, what people do with it and what people think it is. He hedges the premise: “Although I’m sure some will disagree”.
• Three intersecting foci (2025-01-07 universities-need-to-step-up-their-agi-game). “Where we live”, “what we do” and “who we are” are offered as a “domain-agnostic” frame for university work on AI. It is explicitly an alternative to tagging AI onto disciplines.

In each case the model is a way of seeing, not a claim to be right. The amanuensis post ends with “like all thought experiments, this could be completely wrong.”

“What if”, then a plausibility test#

He lets imagination run and then reins it in. That is the core move of the BCI post (2024-11-17). He singles out Gordon and Seth’s decision to “set practicality and feasibility to one side and give imagination free reign”, followed by a hard return to plausibility, and recognises it as his own method: “It’s a technique that I’ve used in the past to explore the plausible boundaries of emerging technologies, and one that works well for closing the shutters on hyperbolic speculation”. The point of the exercise is ethical: it is “a critically important step toward ensuring ethical questions and constraints are grounded in plausibility rather than hyperbole.” Their line about imagination needing to be “reined in by what is … practical, feasible, and ethical” makes him want to add “an “amen”” (fn 5).

The same pattern shows elsewhere:

• 2024-12-17 biopreservation opens with pure “what if”: “Imagine it was possible to extend the time between a donor heart becoming available and a transplant patient receiving it from hours to days”. It then turns straight to “a mountain of ethical and social challenges” between capability and use.
• 2025-01-07 universities: faced with Altman’s prediction, he neither dismisses nor endorses. “It’s tempting to dismiss this as hyperbole. But given recent advances … I suspect there’s a reasonable chance he might be right.” The plausibility test cuts both ways: against hype and against reflexive scepticism.
• 2024-11-24 amanuensis: “This may seem far fetched. But I don’t think it is.” He then grounds it in the MIT study as evidence the flip is “at least partially” happening already.

Reasoning by structural analogy#

• Rocket science (2024-11-17). “I love the analogy they use here of space flight”. Spaceflight is engineering on well-understood science. BCIs are engineering without the science: “BCIs are not “rocket science” because, unlike rocket science, we don’t yet have the science to underpin the engineering”. He picks out their warning against approaches “that substitute understanding with brute force”. The analogy maps directly onto AI, although he does not draw that line here.
• The amanuensis. A historical role (the composer’s scribe who shapes the work) is used to see a possible future relationship. He first establishes that the historical amanuensis was never a mere recorder. That makes the analogy richer, because the human in the flipped relationship will “through necessity, bring some creative insight to the process.”
• Aitken’s dust counter (2024-12-01). He traces NOAA’s geoengineering monitors back to an 1889 “dust-counting apparatus” carried round Europe. The structural lesson is from his 2015 piece: “seemingly novel challenges don’t always demand novel solutions, and sometimes, the key to moving forward safely, is to look back at what’s already known.”
• Authenticity (2024-11-17). He generalises out of the case: “This question of authenticity is one that isn’t unique to eBCIs. Rather, it applies to pretty much any technology”. The same goes for cheapened achievements: “we’ve been using artificial means to enhance our minds in order to achieve stuff for millennia.”

He experiments with the thing itself#

• Sora (2024-12-20). An informal replication. He repeats “exactly the same prompt sixteen times”, counts the results (sixteen men, fourteen white, thirteen beards) and posts all sixteen links as data.
• Chatbots as thinking partners that resist (2024-11-24, fn 2). He tried to explore the amanuensis idea with ChatGPT-4o and Claude 3.5 Sonnet. Both kept pulling the human back into the driving seat unless he used “some fancy prompting footwork”. From that friction he draws a hypothesis: “I wonder if the guardrails built into these platforms resist explorations that diminish the potential role or centricity of humans.” The experiment produces a finding about the tool as well as the idea.
• The FAQ (2025-01-12). He reports his method in detail: “The process ended up being both iterative and generative”. He logs the time: “around 4 hours from start to finish” while juggling meetings. He compares it with an estimated six-month committee.

Curiosity and serendipity as drivers#

• He says outright that the amanuensis post came from chance: “this thought experiment was the product of a very human process of serendipitous inspiration.”
• The Aitken post is a detour he takes because it fascinates him: “This is not an article about geoengineering as such. But it is inspired by one.” It ends on an open invitation: “it would be fascinating for someone to replicate this set of measurements now!”
• On the podcast: “to be surprised by novel perspectives, and to be delighted by the serendipity of unexpected insights” (2024-12-22). The year-end episode is “a delightfully serendipitous 2024/25 retrospective/prospective”, with “plenty of irresistible rabbit holes” (2024-12-31).
• Curiosity is also the thing he defends. For him the value of science is “the freedom to ask “how” and “what if” — even when it’s hard to see the relevance” (2024-11-10).

He holds tensions rather than resolving them early#

• Science’s purpose (2024-11-10). Science should benefit society: “And I would agree — to a point.” He then argues that “societal good” must include long horizons and “what scientists get out of the deal.”
• Productivity and joy. He celebrates AI productivity in the FAQ post, “Generative AI is limited and flawed. But it’s also a game changer”. Eight weeks earlier he warned that the same productivity, applied to discovery, drains joy. The two are consistent because in the FAQ he stays the primary agent: “ChatGPT did the bulk of the work. My primary role was to curate everything”. The work was a chore, not the point. This is the distinction he approves in Mollick: don’t use AI “when the effort’s the point” (2024-12-13).
• What o3 is (2025-01-05 five-voices-five-pieces). He values Mitchell’s post for holding both sides: models are “pushing far beyond critiques that large language model-based AIs are simply “stochastic parrots”” but “we still don’t know what this means”.
• Hyperagency (2024-11-17). Doubtful, then leaves it open: “I’m not sure I can see eBCI users going into an existential tailspin … But I may be wrong. Either way, this is a possibility that we should probably be discussing more than we are.”

Play and wit as ways of thinking#

• 2024-12-29 fantasy-top-ten-lists-2025 is a genre joke with a purpose: a “top ten list of top ten lists that no-one’s writing”. The footnotes carry serious tests. The “pencil and paper” item is “one of those tongue in cheek tests I use to assess hyped up educational tech”, asking “whether we’re simply trying to manufacture problems that we claim emerging technologies can solve”.
• Humour does argumentative work throughout. Sora: “just to get another white dude”; “Sigh …”. The committee estimate is followed by “I’m waiting for the emails telling me that my estimate was way too optimistic!” (fn 4). He notes on the Risk Innovation Planner: “(we almost nailed it)”.

2. What matters to him#

Joy, wonder and curiosity, treated as things of real value#

This is the emotional centre of the batch.

• 2024-11-10: “the soul of science lies in the delight and wonder of exploring the unknown rather than just being a cog in a knowledge production line.” Curiosity is “the fuel”, and the loss is civilisational as well as personal: “the moment we frame science as simply a way to manufacture knowledge as fast as possible … I think that a little bit of all of us dies in the process.” The alternative he fears is “a utilitarian tool to support a utilitarian world heading for a utilitarian future — and one where curiosity, joy, and wonder, have no place.”
• 2024-11-24: the thought experiment’s closing worry is a future that “begins to suck the joy out of what we do, even if it does lead to increases in productivity.”
• 2024-12-29, fn 7: “The concept of joy is much under-appreciated in how we think about the roles and impacts of technology in our lives — as a result many advances have a nasty habit of sucking the joy out of what we do without us realizing it!”
• The close of 2024-12-29: “here’s to a 2025 that’s threaded through with meaning, joy and peace”.

He speaks from inside science here, “From my career-long experience as a scientist”, and treats wonder as “core to what makes us us”. Even so, he declines to be merely nostalgic: “Fortunately, I suspect there are ways of using AI in scientific research and discovery that spark curiosity and increase the joy and wonder”.

Human creative agency#

The amanuensis post worries about “a shift away from human-centered professional or creative agency, and toward AI-directed and human-executed implementation”. Its postscript holds on, provisionally, to “a spark of creativity that leads to novel value creation which remains uniquely human”, then immediately asks “how long it’ll be before even this aspect of what makes us uniquely human is challenged.” His deepest worry is organisational: that the model is “so compelling that organizations adopt it at scale — without fully understanding the potential long term consequences to human creativity and innovation.”

What it means to be human, held open#

• 2024-12-29, fn 12: “it’s getting harder to hold on to what we’ve thought of as the bedrock of being human as if it’s an immutable truth.”
• 2024-11-17: he gives the last word to the essay’s question of “who, and what, we are—and who and what we can, and ought, to be” and lingers on the skull as a boundary.
• 2024-12-22: the 2022 initiative statement he wrote imagines “what it might mean to be human a hundred years from now”. He does not treat being human as fixed ground to defend. It is a question that technology keeps reopening.

Inclusion and whose future it is#

• Sora (2024-12-20): “But I’d expected better from them.”
• 2024-12-15: he frames the Native American piece around “the need to be more inclusive and imaginative”, and pairs harm with possibility: “the news isn’t all bad”, “a call to action for ensuring that the equally real potential of AI is also fully realized in these communities.”
• The podcast (2024-12-22): “spaces where everyone — no matter who they are — can be part of exploring the futures we collectively aspire to”.

What frustrates him#

• Hype. “Hyperbole is the primary currency in an attention economy. What else can I say.” (2024-12-29, fn 9). BCI ethics should be “grounded in plausible challenges rather than hyperbole” (2024-11-17).
• Hubris in power. “a bunch of tech bros forget everything they ever knew about the Dunning-Kruger effect when it comes to governance and policy” (2024-12-29, fn 11). Neuralink’s Musk-first authorship is an “ethically dubious choice” (2024-11-17, fn 1).
• Brittle rules. “black and white ideas” and policies that will be “highly brittle” (2024-12-13).
• Institutional inertia, including in his own world. Universities are “mired in tradition, convention, and self preservation” (2025-01-07). Committees would take “3 - 6 months producing a mediocre set of FAQ” (2025-01-12).
• Writing that ignores the reader. He dislikes “opinion pieces that focus on what the author wants to say, not what I want to read” (2025-01-12, fn 3).
• Forgetting the past. The 2015 anecdote about a student asking “can we trust papers more than a few years old?” left him “taken aback” (2024-12-01).
• The attention economy’s unfairness to quieter voices. It is “sometimes hard to be heard as a writer, even if you have something interesting and worthwhile to say” (2025-01-05).

What delights him#

• A good essay: Gordon and Seth’s imaginative passage “is a delight to read”; their “new hole in the head” line is “my favorite quote” (2024-11-17, fn 4, with a 😄).
• An 1889 table of dust counts: “a quite delightful record of observational notes on aerosol concentrations on a jaunt around Europe” (2024-12-01).
• Weirdness: “we’re definitely going through a “weird science and tech” phase in human history — which is exhilarating” (2024-12-29, fn 10).
• People: talking with Cady Coleman was “an absolute blast — and it was so humbling” (2025-01-07 a-conversation-with-cady-coleman).
• Being surprised: the mental-monoculture risk “intrigued me. It’s not one I’d considered before” (2024-11-17).

3. Risk as a way of thinking#

The explicit statement: risk innovation for biopreservation (2024-12-17)#

This is the clearest account of the framework in the batch, and it is his own narration of his talk.

• Definition. Risk innovation is “a way of bringing an innovation mindset to understanding and navigating the types of risks that are often ignored but have a tendency to bit hard in a world dominated by rapidly advancing technological capabilities.” It is presented as a mindset applied to risk, driven by the pace of technological change.
• Reframing risk. “reframe risk as a “threat to value” — whether this threat is to the creation of new value, or a threat to existing value.” Risk innovation “explores novel ways of protecting and growing value.” The unit of concern is what people care about, not a probability of harm.
• Value, not values. “the latter are important, but the former is more effectively operationalized in policy and decision-making.” This is a pragmatic choice made to open conversation with innovators, not a dismissal of ethics.
• The risk landscape as terrain between here and there. It is “a risk landscape that lies between where an enterprise currently is, and where it wants to end up.” Crucially, the landscape includes the conventional: “risks that can often be handled using existing tools — including quantifiable risks to human, environmental, and fiscal health.” It also includes “risks that are often overlooked because they’re messy, subjective, and hard to deal with — we refer to them as “orphan risks”.” Quantitative risk science is built on, not discarded.
• Reciprocal threats. Orphan risks arise both from an organisation’s own actions and “from an organization threatening value associated with key stakeholders in ways which lead, in turn, to a threat to value to the organization”.
• Tool or mindset? Both, but the tool serves the mindset. The Risk Innovation Planner was meant “to take 30 minutes to complete and help reveal new ways of thinking about risks and opportunities”. The key takeaway is that “the process raises awareness of factors that may otherwise be overlooked” and helps innovators “navigating around potential roadblocks early on.”
• Responsibility without preaching. The approach works “not necessarily by impressing on researchers and developers the need to “do the right thing”, but by making it clear that their success is intimately intertwined with how they impact (and threaten what of value to) others around them”. Ethics is also framed as the reason for the technology, not an add-on: ignoring it “would negate the very reason why members of the engineering research center were doing this in the first place.”
• Humility about his own evidence. “This plot is dense and subjective — there’s no data here that can be taken as repeatable or absolutely representing reality.” He tested whether his own framework would transfer at all: “we didn’t know whether this framework … would translate over to a multidisciplinary and multi-sector initiative — which is why we embarked on this study.” He ends with “Of course more work is needed — there always is.”

The same logic, unlabelled, in the AI posts#

The terms appear only in the biopreservation post. But the reasoning runs through the AI posts, which is where it does its most interesting work.

• Joy as threatened value (2024-11-10). The MIT study is, in his reading, a threat-to-value finding that productivity metrics miss. “Societal good” must count “the human investment needed to generate societal good. And this includes paying attention to what scientists get out of the deal.” The argument has the reciprocal structure of the framework. Threaten the value scientists get from their work, and you threaten the enterprise of science itself: “who would become a scientist who sets out to make the future a better place if all the soul had been sucked out of what you love?” Joy is exactly an orphan risk: hard to quantify, easy to ignore, and potentially corrosive.
• Value creation as the lens on human–AI relationships (2024-11-24). The amanuensis model “deeply challenges how we think about human-AI relationships in the creation of value”. Its end state is humans who “codify the work of artificial intelligence systems into outputs and outcomes that create value for others.”
• Stakeholder networks in place of regulation (2024-11-06). He describes the likely shift under Trump in matter-of-fact terms: “responsible and beneficial AI will depend less on government oversight and more on a tapestry of soft governance mechanisms that rely increasingly on developers and their key stakeholders — including consumers.” This is the enterprise-and-stakeholders picture of risk innovation, applied to AI governance as a realistic starting point.
• Institutions judged by value (2025-01-07). Universities that cannot lead should “ask what value they actually bring to the table”.

Navigating, landscape and transitions for AI#

• 2025-01-07 universities: “navigate” or “navigating” appears eight times, counting the subtitle and a figure caption. Humanity must “successfully navigate the coming advanced AI transition”. The three foci are for understanding “the emerging landscape at the nexus of AI, society and the future; and how we might approach developing the insights and means to successfully navigate it.” It is a landscape to be crossed, not a hazard to be controlled.
• 2024-12-13 educators: students need “an ability to parse and navigate the shifting limitations and affordances of AI tools”; educators face “a rapidly changing AI landscape”.
• 2024-11-17 BCI: he picks up Gordon and Seth’s six questions as “critical to navigating the path between what is possible, and what is ethical”. The post title uses “Navigating”, as does the biopreservation post.

A novel technology calls for new mental models, but not amnesia#

• Mental models, explicitly (2024-12-13). ChatGPT’s 2022 release “helped form and crystallize mental models of what generative AI is”. “These are mental models that, in my experience, many people still hold.” The danger is decisions “being made on assumptions that are out of date”. His remedy is a way of thinking rather than a rulebook: “nurturing new ways of thinking about AI that are agile and adaptable”, “guidelines on how to think about AI”, and “being aware of critical failure modes that are likely to persist across multiple generations of AI”.
• His own mindset shift (2025-01-12). “But, of course, this was me in pre-ChatGPT thinking mode.” Even the language fails: “I’m not sure we have the appropriate language yet to describe working with machines that are fundamentally different from passive devices” (fn 1).
• Breaking stovepipes (2025-01-07). Tagging AI onto disciplines “would be a mistake. As well as constraining thinking” it would make AI “just the latest buzzword”. He wants “integrated and domain-agnostic ways” of thinking. He wants to “empower researchers to embrace radical creativity without having to bow to conventional metrics of success”.
• The counterweight (2024-12-01). “seemingly novel challenges don’t always demand novel solutions, and sometimes, the key to moving forward safely, is to look back at what’s already known” (2015). In the BCI post, some worries (authenticity, cheapened achievement) are old worries raised “to a new level.” His stance is discriminating: work out which parts of a technology are truly new and which have a long lineage.

Humility against false precision#

• “Of course we can’t predict the future.” (2024-12-29, fn 10)
• The amanuensis argument rests on a low bar: “even if there’s even a small chance of a future where people begin shifting into the role of human amanuenses … we should probably be thinking about whether we want such a future”. This is anticipatory reasoning under deep uncertainty, stated calmly.
• “it’s always good to stay grounded in what a transformative technology is not doing — at least not just yet” (2024-12-29, fn 5).
• On the MIT paper, a later update (May 2025) flags doubts about its integrity: “I’m keeping this post here for the moment, but please read with caution.” He corrects in public, and does not quietly delete.

4. Scholarship and public writing#

Turning papers into posts#

• 2024-12-17: a peer-reviewed JLME paper becomes “the quick(er) version”, built from his conference slides, with a pointer to the free full collection.
• 2024-12-01: his 2015 Nature Nanotechnology piece, “a favorite paper of mine (despite a rather sad number of citations)”, is brought back into a current debate. He links free access to Aitken’s papers via HathiTrust and attaches the 1890 data table. His geoengineering-ethics blogging goes back to 2009 (fn 1).
• 2024-11-17: he revisits his 2019 JMIR response on Neuralink with Scragg, recommends a new essay, then extends it. They “stop short of pushing it to a number of plausible conclusions — one being eBCI users being incentivized to “share” their neural data through free or reduced cost subscriptions services.” He adds business models and incentives, “creative ways of companies extracting value from users”, to a philosophical analysis. That is the risk-innovation habit of asking where value flows.

Experimenting in public#

• The amanuensis thought experiment is worked out on the page, figure by figure, with an open request: “I could not find any other work that clearly explores the concept … If I did, please let me know in the comments” (fn 1).
• The Sora experiment publishes its raw outputs. A second update adds OpenAI’s System Card text, which gives the company’s own position a fair hearing.
• The FAQ post publishes method, time taken and uncertainty about authorship (“I went back and forth on this”). The result is released “under a Creative Commons license so they can be freely shared and adapted”.

How he treats evidence and expertise#

• He is precise about the limits of evidence: “participant numbers were relatively low — 17 in all”; the synthesised areas of value “should not be considered as robust or authoritative” (2024-12-17).
• He discloses conflicts: “in full disclosure, I’m a member of the CIFAR Research Council” (2024-11-17, fn 3).
• In other writers, he values “a respect for following the evidence, challenging uncritical thinking, and encouraging open dialogue” (Revkin, 2025-01-05) and a “clear-eyed look” (Mitchell).
• He defers when others know more. Of Mark Daley on o3: “Mark is way more insightful and eloquent here than I could ever be.”
• He is generous to people who were wrong. Aitken’s climate reasoning ran “contrary to what we now know”, “And yet, 135 years after he first presented his research, his thinking is nevertheless thought provoking.”

Transdisciplinarity#

• ATP-Bio’s ethics group “includes transplant surgeons and lab researchers as well as social scientists, bioethicists and policy experts” (2024-12-17).
• His five recommended writers span AI across disciplines, the sociology of electric and autonomous vehicles, climate journalism, practical AI use, and complexity science (2025-01-05).
• Universities need “researchers and scholars who comfortably work across domains and excel at breaking the mold of academic conventions” (2025-01-07).
• He moves in a single month from aerosol physics to neuroethics to organ preservation to Indigenous data sovereignty to classroom policy.

Accessibility as principle, not packaging#

• A footnote addressed to students: “its not how you write but what you say that’s important — as long as there’s rigor and scholarship behind it” (2024-11-17, fn 4). Rigour and plain speech are not in tension.
• The podcast is a way to take ideas out of “the inaccessible corridors of academia” (2024-12-22).
• The FAQ succeeds because it answers “the types of questions that educators who are floundering are likely to ask” with “clear and direct answers … without getting lost in the weeds” (2025-01-12).

5. His role as he sees it#

Catalyst, not centre#

His 2022 statement for the initiative, quoted in 2024-12-22: “Our vision was not one of a research center, or even an educational endeavor (although we do both), but of an initiative that would catalyze thinking at scale”. The initiative answers “a “growing need for bold ideas and visionary insights that transcend the constraints of conventionality”. Modem Futura is “part of this vision of catalyzing thinking at scale.” The role is to start thinking in others, at scale, not to hand down conclusions.

Thinking out loud, with the audience as companions#

• “Each week, the conversations that Sean and I have are part of our own intellectual journeys.” Listeners “are invited to join us” (2024-12-22).
• The style is chosen: “intentionally authentic and conversational”, ““in the raw””. It “eschews any sense of being “talked-at” in favor of creating a space where listeners feel they have permission to explore nuanced and complex ideas with us.”
• “At its core, Modem Futura is an invitation”.
• The fantasy list is “an open invitation for others to fill in the gaps” (2024-12-29).
• He invites correction repeatedly: “please let me know in the comments” (2024-11-24), “If you know of anything comparable, please let me know!” (2025-01-12).

What he refuses to do#

• Preach, polarise or bore. “you can’t do this by being preachy or polarizing or boring — you have to be worth listening to, you have to be engaging and you have to build meaningful relationships with your listeners, whether they’re high school students, retired or anyone in between” (2024-12-22, from the jointly credited interview answer he endorses).
• Take a partisan line. The morning after the election he writes without denunciation. The ideas “should be informing thinking around AI policy in the incoming administration” and matter “whoever’s at the political helm” (2024-11-06). He addresses power as a possible audience, not as an enemy.
• Moralise at innovators. Risk innovation works “not necessarily by impressing on researchers and developers the need to “do the right thing”” (2024-12-17).
• Show off. He resists end-of-year lists that “show off their expertise and insight” (2024-12-29). He is wary of talking about his own book: “it’s rarely a good idea to talk about your own books” (2024-12-08).
• Fear-monger. Sora’s bias gets a “Sigh …”, not outrage. The amanuensis future is something “worthy of further exploration”, not a warning siren.

Towards the people around him#

• Industry. He is fair and firm. AI companies are “trying hard” but “still lack the breadth of vision and understanding that’s necessary” (2025-01-07). On bias, “Companies like OpenAI and Midjourney have made huge strides”, yet Sora shows “I’d expected better”. He takes Altman seriously enough to test his prediction.
• Governments. They have “the societal mandate”, but “in most cases they lack the imagination, vision, or agility” (2025-01-07).
• His own institutions. He is at his sharpest here. Universities are what they claim to be only “if you buy into the rhetoric”. Most are “comfortably stuck in bystander mode as they play with the latest AI tools”. He ends by questioning what value academia brings “as advanced AI is poised to transform every aspect of our lives.”
• Colleagues. He is honest, but not cruel, about lagging perceptions: “18 months and a lifetime out of date. Im exaggerating of course” (2024-12-13). He writes FAQs because colleagues are “floundering”.
• Students. He sees them as ahead of their teachers: “students are learning through creative thinking and hands-on experience how to extract value from emerging AI tools and capabilities far faster than most of their instructors are.” He worries they will be “penalized if they don’t confirm to outdated expectations” (2024-12-13). He encourages students who fear they “can’t write academic”. He brings in a “good colleague and former student” (2024-12-15).
• Communities and other voices. He curates and convenes: “I asked Sean and Al to reflect … I also asked good colleague and former student Leonard Bruce”. He uses his platform to “carve out a bit of space” for other writers, preferring “people who aren’t well know but who are, nevertheless, interesting thinkers” (2024-12-29, fn 3). He recommends a colleague’s Substack called Tech Skeptic Goes Electric.

Changes of mind and self-revision in this batch#

• On Future Rising’s central metaphor: “while I suggested that “humans are, in a very real sense, architects of the future” back in 2020, I’m still not entirely sure how useful this metaphor is” (2024-12-08). Whether the book is the guide we need “is, of course, for others to decide.”
• On his own thinking about AI work: “this was me in pre-ChatGPT thinking mode” (2025-01-12).
• On the MIT study: he flags later integrity concerns openly and keeps the post up with a caution (2024-11-10, May 2025 update).
• On uses of “we” with a machine: he decides it is appropriate and expects disagreement (2025-01-12, fn 1).

6. What is distinctive#

1. Joy and wonder as risk categories. Most AI risk discussion counts jobs, bias, safety, misinformation or catastrophe. Maynard treats the loss of delight, curiosity and meaning in human work as a serious, under-appreciated threat to value, and follows it from scientists (11-10) to organisations (11-24) to technology in general (12-29, fn 7). He also treats it as a reciprocal threat: drain the joy and you erode the societal good science is meant to produce.

2. AI read through agency and organisation rather than mind. The amanuensis model is not about whether AI is conscious or creative. Indeed, “This model does not imply intrinsic creative ability within the AI”. It is about where AI sits in chains of delegation and accountability, and how that could quietly invert who generates and who executes. His observation that model guardrails may themselves “resist explorations that diminish the potential role or centricity of humans” is an early, unusual note: the tools shape what their users can think about the tools.

3. Plausibility as an ethical discipline. He runs imagination to its limit and then reins it in, and he applies this both ways. It closes “the shutters on hyperbolic speculation” (BCI), but it also refuses to dismiss Altman as hype. This sits between the booster and critic camps without being a midpoint. It is a method.

4. Responsibility as enlightened self-interest in a web of value. The risk-innovation framing lets him talk to founders, engineering centres and AI developers without moralising. Their success is “intimately intertwined” with the value they threaten for others. Under a deregulatory administration, he reads this stakeholder web (“a tapestry of soft governance mechanisms”) as where responsible AI will actually be decided.

5. The problem is the mental model, not the machine. The third s-curve names something rarely named: a perception lag that makes policies obsolete before they are written. His answer is to teach ways of thinking that last across model generations, not rules tied to 2022’s ChatGPT.

6. A physicist’s long memory. Few AI commentators reach back to 1889 dust counters. His insistence that “seemingly novel challenges don’t always demand novel solutions” sits deliberately beside his call for new mindsets. The skill is telling which is which.

7. Institutional self-critique from the inside. He argues universities are the best-placed actors for “navigating advanced AI transitions” and, in the same breath, that they are “mired in tradition” and may have to justify their existence.

8. Personality in the margins. Much of his candour and play lives in the footnotes: the “amen”, the 😄, “Sigh …”, the confession about committees, the pencil-and-paper test. The footnotes are a second voice in which the scholar admits doubt, humour and delight. The form is part of the method: it models the curious, unpreachy stance he wants readers to adopt.

7. The posts that best reveal how he thinks#

1. 2024-11-24 artificial-intelligence-agency-human-amanuensis. Serendipity (a reader’s reply), a model built and broken in public, experiments with chatbots that push back, a plausibility argument under deep uncertainty, and the joy of work as the value at stake.
2. 2024-11-10 is-ai-poised-to-suck-the-soul-out-of-science. The clearest statement of what he cares about: wonder, curiosity and “what if”. The argument treats scientists’ satisfaction as a stakeholder value whose loss threatens the enterprise. Also shows public correction (the 2025 caution).
3. 2024-12-17 navigating-the-challenges-and-opportunities-of-advanced-biopreservation-technologies. Risk innovation in his own words: threat to value, the risk landscape that includes quantitative risk, orphan risks, and a tool meant to “reveal new ways of thinking”. Honest about subjective data, and non-preachy about responsibility.
4. 2024-12-13 are-educators-falling-behind-the-ai-curve. Model-building (three s-curves), explicit talk of outdated “mental models”, and a preference for ways of thinking over brittle rules.
5. 2025-01-07 universities-need-to-step-up-their-agi-game. Navigating AI transitions as a landscape, the anti-stovepipe argument, radical creativity freed from conventional metrics, and sharp self-critique of academia.
6. 2024-12-22 why-modem-futura-is-more-than-just-another-tech-podcast. His own account of his public role: catalysing thinking at scale, thinking in the raw with the audience, serendipity, and refusing to be “preachy or polarizing or boring”.

Also revealing: 2024-11-17 (the imagine-then-rein-in method, in his own words), 2024-12-29 (play as critique, joy as neglected value) and 2024-12-01 (the physicist’s long memory and delight in old data).