Late Lessons, Jensen Huang and AI
Markdown version

B25 perspective notes: 2025-04-13 to 2025-05-25 (10 posts)#

These notes read the batch for how Maynard thinks, not for the concepts he names. I read all ten posts in full. For the three Modem Futura episode posts I took only his own framing lines. Quotes are exact, including his typos (“discipled”, “alter”, “Un the US”, “AI responsibility” where he means “responsibly”). Markdown italics are shown as asterisks where they carry emphasis.

Evidence rules applied

Context. Spring 2025, at the end of the ASU academic year. Colossal’s “dire wolves” have just been announced. The US has issued an AI-education Executive Order a week after China’s announcement. Agentic AI (reasoning models, Manus) is the talk of the moment. Maynard is writing a responsible-AI module for a new undergraduate AI-literacy course and preparing to teach Moviegoer’s Guide to the Future “for the 9th time I believe”. The Substack is two years old, and he co-hosts Modem Futura weekly with Sean Leahy.

1. How he thinks here#

He starts from a practical problem, not a thesis#

Several posts begin with something he has been asked to do, or something that is not working. He then treats the constraint as the thing to think with.

He reframes by turning a concept back on the people who use it#

Analogy by structure, often carried over from physics#

Plausibility tests and first principles, applied to hype in both directions#

He builds things to find out, and so that others can play#

This batch shows a steady habit of making small working things:

Building is how he tests an idea and how he hands it to others. A framework becomes something you can turn around in your hands.

Films and stories are how he gets into a question#

Holding tensions rather than resolving them early#

Curiosity, play and serendipity drive the attention#

2. What matters to him#

Knowledge that reaches people and empowers them on their own terms#

This is the value stated most clearly in the batch. His aims for public writing (2025-04-20) begin with “Making emerging research and insights (including mine) as accessible as possible to as many people as possible”. Of the four purposes of expert–public communication (instruction, ego, impact, empowerment), he puts himself with empowerment: “providing others with access to information that they are able to utilize on their own terms” (2025-05-25). He wants relational communication “not just to bring about pre-determined outcomes, but to enable participants to benefit from this flow in ways that benefit them on their own terms.” The phrase “on their own terms” appears twice. He wants knowledge to move “from where they are concentrated to where they are needed”, “from a trickle to a flood”.

Care, and a broad sense of what counts as value#

The responsible-AI module rests on “developing a culture and a mindset of care around AI development and use” (2025-05-18). Value is not narrowed to money or safety. It may be “wealth, health, and wellbeing. Or it might be dignity, autonomy, and the ability to be part of building a better future.” Responsibility belongs to everyone: “everyone who uses the technology, or whose life is impacted by it in some way, has a stake in” it.

Students, learning and agency#

He treats students as sources of knowledge. “I came away having learned a lot from them.” Peer-to-peer learning about AI “is part of how learning should work — at least in my books” (2025-05-06). What moved him most was Bella’s question, “How can we navigate the future if we’re not currently exposed to it?”, which “struck such a deep chord with me as it’s at the core of what I do as an academic, professor, and writer.” He takes the lesson as a duty for educators: “it does put the onus back on us as educators to create learning environments that expose our students to the future”.

Humility, listening and power used well#

From FFTF: responsible science is “about more than just having good intentions. It also means having the humility to recognize your limitations, and the willingness to listen to and work with others who bring different types of expertise and knowledge to the table” (p. 39). Power is to be understood and used, “not eschewing it, but understanding and accepting the sometimes-complex responsibilities that come with it” (p. 45). The aim is to “build a better future for as many people as we can, without sacrificing the health and well-being of communities and individuals along the way” (p. 45).

Governance that fits the technology#

He wants oversight that neither smothers nor misses: approaches “that neither stifle innovation because they treat agentic AI as a one-size-fits-all technology, or fail to ensure it’s responsible development and use because they are designed for technologies that don’t match reality” (2025-05-04).

What frustrates him#

What delights him#

Students: “smart, candid, funny, and insightful”. “I was almost tearing up at this point!” (2025-05-06). Making small tools. Serendipity. The audacity of people he criticises. AI doing something surprisingly well. The small human details of his readership: “no subscribers in Mississippi”, “still waiting for my first subscriber from Greenland!” (2025-04-20).

3. Risk as a way of thinking#

Threat to value, stated as the hinge of responsible AI#

The clearest statement in the batch comes at the end of the 2025-05-18 script: “understanding effect as a threat to what is valuable to us and those around us — and what we care about, or what might actually enhance that value — becomes a powerful way of thinking about responsible AI.” Footnote 4 ties it to his own framework: “This is an approach that is used in Risk Innovation”. Note the form of words: “a powerful way of thinking”. It is offered as a way of seeing, not as a method for scoring. The question it leaves students with is open: “how our decisions or actions in the present will increase what we value in the future, or potentially jeopardize it”. Value can be enhanced as well as threatened, so benefit and harm sit in the same frame: “remembering that effects can be good as well as bad”, and the tool helps “avoid the ones we don’t want, while enabling those that we do.” Missed opportunity also appears as a risk: the US “cannot afford not to invest” (2025-04-24).

A landscape of plausible futures, already present in 2018#

The FFTF chaos passage (p. 41) reads as an early version of the risk-landscape image. Futures have boundaries that separate the plausible from “sheer fantasy”. There are “points of stability” that action can reach, and good futures that “can be squandered”. Unpredictability is real, “Complex systems behave in unpredictable ways” (p. 39), but it is bounded. Foresight therefore makes sense even when prediction does not.

Novel technology, new territory, new mindset#

Mental models, not procedures#

He is explicit that his devices are for opening thought. The six curves are “just six possible ways of approaching cause and effect that are designed to help think through potential consequences”. They are “useful for opening up new insights” (2025-05-18). The 2018 book says the same about itself: “there are no easy guidelines or rules of thumb”, but “much of this book is devoted to ways of thinking that reduce the chances of making a mess of things” (FFTF p. 39). He values the agent framework because it gives governance a multidimensional space to think in, and he turns that space into something readers can explore rather than a checklist.

Quantitative foundations, built on#

The six shapes (linear, S-curve, exponential, hysteresis, jagged, chaotic) are the vocabulary of a physicist and exposure scientist: response curves, thresholds, saturation, irreversibility. He does not use the phrase “dose–response” here; the link is my inference. What is distinctive is that the curves are the starting point, not the end. They lead to “effect” as value and care. The same foundation shows in the agentPlot assumption that risk rises “closer to exponential than linear” along each axis, and in his physics-based corrections of both hype and doom.

Echoes of orphan risks: harms off the usual axes#

The term “orphan risks” is not used, but the harms he notices consistently fall outside conventional risk categories:

Humility against hubris and false precision#

The Colossal framing (“The hubris here is palpable”) and the FFTF chapter carry the same lesson: “the dangers of thinking you’re smart enough to have every eventuality covered” (p. 38), and “if Hammond had read his Perrow” (p. 43). He applies it to himself as well. The parasocial essay concedes “much of the evidence we have to go on here is anecdotal rather than systematic”, and his reading of the agent framework is “not fully convinced … yet”.

4. Scholarship and public writing#

One continuous practice across formats#

The batch moves ideas between the book, the classroom, the podcast, the Substack and web tools without marking a boundary between scholarship and public writing:

He names writing as a way of thinking: “Using writing as a discipled way to stay on top of new developments while developing and extend my own thinking” (2025-04-20). The Substack exists partly to “scale the reach and impact” of the ASU Future of Being Human initiative.

Experimenting in public, with readers as participants#

The AI buttons are “still somewhat experimental, so please do let me know how useful it is!” (2025-04-20). The code for Prompt2URL and agentPlot is on GitHub. The Deep Research workflow is disclosed rather than hidden (2025-04-24). He publishes the audience numbers, low ones included: “These are still low compared to what I’d like to see” (2025-04-20).

How he treats evidence and expertise#

Transdisciplinarity#

Within ten posts he draws on chaos theory and hysteresis, genetics and conservation biology, Perrow’s normal accidents, science-communication and media-studies literature, education policy in two political systems, AI governance frameworks, and film and television. The students he interviews are from SFIS and the College of Global Futures, programmes that are themselves transdisciplinary.

Accessibility as care for the reader#

He provides the trailer transcript because “not everyone who sits down to read this Substack appreciates multimedia segments when they were simply hoping for a quiet read” (2025-05-11). The script is marked as “written to be read aloud”. A footnote explains who Stan Lee is: “I was informed that many undergrads won’t know who Stan Lee is — I’m getting old!” (2025-05-18). The summary buttons are for readers who “don’t have time to read the whole thing” (2025-04-20).

5. His role as he sees it#

A relational public scholar, working toward empowerment#

The most direct self-description in the batch: “In many ways it’s what I’ve aspired to for much of my career: communication and engagement that focuses on empowering others through relationship building, and that opens the way to potentially transformative exchanges of understanding” (2025-05-25). The podcast’s cold opens, personal stories and choosy guest list are all defended as relationship-building. The goal is “the permission they grant us as a result to talk about things that matter”, and “It’s why we focus on building connections rather than instructing listeners.” The work is “grindingly hard”, and it is worth doing because the topics are “far too important to be sacrificed on the alter of misguided ideas about what constitutes right and proper communication.”

An educator who exposes students to the future#

Bella’s question marks “the core of what I do as an academic, professor, and writer” (2025-05-06). He sees his job as helping people take part in “navigating toward the future”, and “imagining and designing the types of future we aspire to”, not delivering answers. The trailer asks questions and does not give conclusions: “What are the technologies we want in our lives? Where do we want these technologies to take us? And how do we spot the dangers before it’s too late?” (2025-05-11).

A learner among students and colleagues#

He comes away “having learned a lot” from sophomores. He calls Jamey Wetmore a “long-time friend and colleague” and frames the EV episode around “why informed skepticism matters” (2025-04-15 subtitle). He hosts a filmmaker who “looks far beyond Hollywood stereotypes and tropes” (2025-04-22).

A reformer of his own institution, from inside it#

He writes as someone inside academia’s reward system (chair of promotion and tenure), arguing for “a recalibration in how experts and institutions across multiple domains think about the purpose and practice of expert-public communication” (2025-05-25).

With industry, policymakers and experts#

What he refuses to do#

Changes of mind, shown in the text#

6. What is distinctive#

  1. Responsible AI taught as the shape of consequences, then turned into a question of value. Most AI-ethics teaching lists principles. He notes there are “Hundreds of ethical AI guidelines”. Maynard gives students six response curves from physics and systems thinking, including irreversibility (hysteresis), uneven distribution (jagged) and tipping points (chaotic). He then redefines the y-axis as what people value and care for (2025-05-18). The quantitative science is the base, and the value question is the aim. Few people discussing AI combine the two in this way, and fewer do it for first-year undergraduates.

  2. Formulating the problem comes before governing it. On agentic AI he says the field has not yet formulated the problem (2025-05-04). He then builds a tool so that others can explore how agents are characterised. His one substantive addition to the framework is to ask where effects on beliefs and behaviour fit, a concern with the human and social channel that runs through his work.

  3. Governance matched to the technology’s reality. His test for oversight is fit: neither “one-size-fits-all” nor “designed for technologies that don’t match reality” (2025-05-04). This is neither the “pause” position nor the “don’t regulate” position. It is a risk scientist’s insistence on characterising the thing before choosing controls.

  4. A 30-year-old film as the lens for a decacorn. He answers the dire-wolf news by republishing a chapter on Jurassic Park (2025-04-13). His use of chaos theory is unusual: he stresses the bounds and stable points that separate “plausible futures from sheer fantasy” (FFTF p. 41), not just unpredictability. His reading of power also refuses the morality tale.

  5. A theory of his own public role, and a defence of it. The parasocial essay is a scholar theorising his own communication practice. He places it within a sorting of motives (instruction, ego, impact, empowerment) and argues for it against institutional disdain. For the AI discussion, which is dominated by one-way expert instruction on one side and personality-driven influence on the other, he offers a third path: relational communication aimed at people’s agency “on their own terms”. He knows the same mechanism can manipulate (2025-05-25), and his companion-bot example in 2025-05-18 shows the risk. He does not connect the two posts himself.

  6. Students as primary evidence and co-thinkers. He takes seriously the students’ view that AI “will reveal and amplify” whatever educational values a student brings, and their informal “AI code-switch[ing]”. He treats both as findings with consequences for institutions, not as anecdotes (2025-05-06).

  7. Refusing zero-sum framing. In a moment dominated by “AI race” talk, his US–China analysis ends with collaboration and resilience, not with a scoreboard (2025-04-24).

  8. Play and serendipity designed into the scholarly platform. A serendipity button, AI “continue the conversation” links, a prompt-to-URL tool and interactive frameworks (2025-04-20, 2025-05-04, 2025-05-18). His scholarly platform works as a place to play with ideas, and he says so openly.

7. The posts in this batch that best show how he thinks#

  1. 2025-05-18 exploring-ai-through-cause-and-effect. A teaching problem (pace) becomes a mindset tool built through play. Physics metaphors (hysteresis, chaos, S-curves) carry social meaning. “Set it and forget it” management is rejected for resilience and course correction. The ending turns “effect” into threat to value and care, with Risk Innovation cited.
  2. 2025-05-25 why-parasocial-communication-is-important. His role in his own words. The deficit model is inverted, the four purposes of communication set out, empowerment “on their own terms” chosen, thinking in public (“still evolving”), and a change of mind prompted by the podcast. He pushes back on AI feedback and also concedes to it.
  3. 2025-04-13 de-extinction-conservation-futures (with FFTF ch. 2, pp. 27–45). He holds hubris and possibility together and uses a film as the lens. Chaos theory is read for its bounds and stable points. Humility is framed as listening across expertise. Power is to be used responsibly, not abdicated.
  4. 2025-05-04 an-important-new-model-for-guiding-agentic-ai-oversight. Formulating the problem comes first. The critique of others’ work is constructive and specific, the effects on beliefs and behaviour are flagged as missing, an app is built with the authors’ blessing, and the governance test is fit to reality.
  5. 2025-05-06 what-students-really-think-about-ai. Listening as method, honesty about small samples, delight in students, and “navigate the future” named as the core of his work.
  6. 2025-04-20 surprised-by-serendipity. His three aims for public writing, serendipity designed into his platform, candour about reach, and joy in making things.