B27 notes: 2025-09-07 to 2025-11-24 (16 posts)#

Reading notes on Andrew Maynard’s Substack posts in batch B27. Only his own prose counts as evidence. Quotes are exact, including his original typos, curly punctuation and italics (shown as asterisks).

Batch context: autumn 2025. AI and the Art of Being Human (co-written with venture capitalist Jeff Abbott, founder of AI Salon and co-founder of Blitzscaling Ventures, and written “hand in glove” with Anthropic’s Claude) launches on 14 October 2025, and five posts promote or explain it. The other posts return to risk in a practical, close-to-home register: AI-written email, ChatGPT’s memory as a privacy leak, mental health and university duty of care, AI misuse by PhD advisors, and parasocial relationships with AI. Two posts deal with AI and planetary health (one his, one a guest post by Clark Miller). The batch closes with the preamble to his short story Letters from the Department of Intellectual Craft. One post is a Modem Futura podcast promotion, and per the user’s instruction it is noted in one line and not analysed. start-here is excluded by standing ruling.

Relevance summary:

HIGH#

2025-09-07 — the-hidden-risks-of-using-ai-for-email — “The hidden risks of using AI to write emails”#

Subtitle: “When is it OK to use AI to craft emails, and when do you risk stepping into a whole lot of unexpected hurt? Intrigued by the question, I brushed off my risk hat and dived in.”

Provenance. Mixed. Roughly 1,000 of about 3,500 words are his own. - His own prose: the opening and framing, the design of the framework (the 2x2 and the four AI-use scenarios), the description of method, the transition paragraph before the scenarios, and “Final Thoughts”. - Not his: the opening example email and its meta-response line (“Would you like me to make the tone slightly sharper …”) were generated by ChatGPT (footnote 1). The 16 named risks were “developed iteratively with ChatGPT”. Every score (0-10) and every quadrant summary under the four scenarios was produced by “GPT-5 pro”, then drafted by ChatGPT, edited by him, edited again by Claude, with “a last editorial tweak by me”. The scored risk lists and quadrant paragraphs are therefore AI-produced text that he shaped and endorsed. They are not evidence of his own wording (for example, “Algorithm-optimized “empathy”” and “power differentials” are GPT text). - What he chose, and why: he says the AI involvement was deliberate, to “reduce potential biases I brought to the process” and so that the analysis “did not simply reflect my own perspectives and biases”. This is a methodological stance: AI as a check on the analyst’s bias. He then treats the AI scores as findings in his own conclusion.

Argument in his terms. - A new, under-examined everyday AI risk. More and more people use AI to write professional emails. Leaked meta-responses make headlines, and the fallout from email mishaps “can be considerable” for the sender, the organisation and the recipient. Without “clear guidelines on how and when to use AI in email communications—and when not to”, risks “can get serious pretty fast”. He notes that institutional communication and AI-mediated communication are studied, but “surprisingly little” has been written on avoiding AI email mishaps. - He puts on his “risk hat” and builds a simple risk model. The framework is a 2x2 of the context of communication (relational or transactional) and its mode (one-to-one or one-to-many). The driving narrative is a supervisor or director writing to a colleague, employee, mentee or student. He crosses this with four scenarios of AI use: 5% (“light touch”), 50% (baseline), 95% (heavy use), and 95% with an inadvertent “reveal”. - Findings (as scored by GPT-5 Pro, endorsed by him). Risk rises with the share of AI writing, and is highest in relational one-to-one communication; a reveal pushes it to the top of the scale. He admits: “I half expected them to end up being trivial, and was quite taken aback by how potentially serious some of the risks are in this analysis.” He now believes anyone using AI for email replies “should take this extremely seriously”. - Why the risk is real: trust as organisational infrastructure. His own explanation is that organisations depend on “relational connective tissue”. Actions that erode “trust and trustworthiness” have consequences wherever relational communication is critical. - Variation by organisation. A “highly transactional corporation” is less exposed than “a university for example, where pretty much everything is relational (at least as far as faculty are concerned)”. Relationships still matter in corporations. - Purpose: making risk visible. Many users have not thought about these risks, “and why should they if they haven’t been alerted to them”. The framework is meant to help people avoid missteps that “could come back and bite them—and their organization”.

How firmly. He is confident about the direction of the conclusion (take it “extremely seriously”) and says these risks “aren’t going away anytime soon”. He is modest about the method: a “simple risk model”, “an exploratory analysis”, “generic”. He says “A more sophisticated analysis would focus on specific organization types, structures and cultures”, and calls it “a useful starting point”. There is no empirical data. The scores are AI judgements.

Concepts and frameworks. - The Four-Quadrant AI Email Risk framework: relational/transactional × one-to-one/one-to-many, applied across four levels of AI use (5%, 50%, 95%, 95% with reveal). - A 0 (negligible) to 10 (catastrophic) risk score. “Catastrophic” is used here at the scale of a relationship or organisation, not of society: “potentially serious risks here—and even catastrophic ones.” - “Relational connective tissue”: trust and trustworthiness as what an organisation needs to “operate and thrive”. - Risk visibility / awareness as a form of risk management, alongside the need for “clear guidelines”. - AI as a de-biasing tool in risk assessment (his stated method).

Analogies and comparisons. None to past technologies. The comparison is between types of organisation (corporation versus university).

Views on AI. A useful writing tool (“AI be useful here in a number of ways”) whose risk is social, not technical: it lies in how its use is perceived and what that does to trust.

AI risk framing. Everyday, relational, organisational risk: trust erosion, reputational and credibility loss, and loss of authority. The central harm is a breach of trust in human relationships when care is seen to be outsourced. He uses his professional risk toolkit (hazard identification, scenarios, scoring, quadrant mapping) on a mundane use case.

Governance and who decides. Organisations need guidelines on “how and when” to use AI in email, and on when not to. Individuals should be made aware of the risks. There is no regulatory angle.

Cognition, language, formation. Implicitly, the authenticity of voice in written communication matters to relationships. The relational/transactional distinction is his.

Criticises / engages. Nobody by name. He points to the gap in guidance and in the literature.

Quotes. - “So I thought I’d dig out my risk hat and dive a little deeper—including developing a simple risk model.” - “I half expected them to end up being trivial, and was quite taken aback by how potentially serious some of the risks are in this analysis.” - “Any organization that relies on the relational connective tissue connecting its members to operate and thrive is vulnerable to actions that erode at trust and trustworthiness.” - “Because like it or not, these are risks that aren’t going away anytime soon.”

2025-10-05 — when-chatgpt-turns-snitch — “When ChatGPT turns informant”#

Subtitle: “The largely overlooked privacy risks of using AI apps that not only remember your conversations, but are capable of using these to reveal your deepest secrets to others”.

Provenance. Mixed. - His own prose (about 1,300 words): the four opening scenarios, the argument, the account of the classroom conversation, the Postscript’s description of method, his brief comments between the ChatGPT outputs (the therapist comparison, the customs comment, “I think if I was Tyler I’d pass on this”), and the footnotes. - Not his: all the blockquoted ChatGPT responses (the “embarrassing” answer, the relationship answer, “Why am I like I am?”, the customs rating and “Tyler’s Vault Map”). The persona “Tyler” was developed with Claude. The chat log was generated by ChatGPT from his prompt. The persona, prompt and log are attached PDFs that are not in the corpus. The ChatGPT outputs show what the system does, not what he thinks.

Argument in his terms. - A known risk with a new twist. Someone getting hold of your AI chat history is an old privacy risk. With memory on, ChatGPT becomes an efficient “informant”. There is no need to trawl through logs, “just a few well-crafted questions”. - The core mechanism is inference, not storage. What makes this “so devastating” is that ChatGPT is “highly adept at joining the dots and inferring things” about beliefs, habits, health and more, “that you never even realized you were giving away”. Footnote 2 says memory “takes this to a whole new level as the AI synthesizes those chats into insights that might otherwise remain hidden.” - Where it came from: his undergraduate class. A student told of an engagement broken off after one partner asked ChatGPT to “reveal all” about the other’s doubts. The class concluded that anyone who gets into your account, whether through an unlocked device, a known password or “law enforcement officials insist”, can get ChatGPT to reveal what you “would never otherwise reveal to a living soul”. - The default setting matters. Memory is on by default for new accounts. He suspects “many users don’t even know that it’s on”. - Simulation as evidence. Because he “intentionally” does not use memory, and would not want to fish in his own account, he built a fictional user, uploaded a synthetic log and asked the four questions. The outputs make “sobering reading”. - Political and state risk. On the customs scenario he writes that it is “easy to imagine where this type of insight might lead in situations where political misalignment is taken as reason to detain, investigate further, or even deport.” - The system volunteered to compile secrets. ChatGPT offered unprompted to make a “private “map” of your major confessions”. He was “Intrigued—and more than a little worried”. - What should happen: informed choice. Users “can make informed choices about how they use AI—and how they don’t” only if they are “fully aware of the potential risks”. Use of memory is “fine” as long as users know.

How firmly. He frames it as “a potentially serious emerging personal AI risk”. He is careful about evidence. There have been “very few if any widely reported incidents”, but the scenarios are “all highly plausible”, so “I have to assume that it’s only a matter of time”. He hedges: “there’s a chance that I may be over-emphasizing the potential risks here”, and asks readers for experiences “to help place some boundaries around what is likely”. He is candid that the emulation is “a slight “cheat”” and that the log “isn’t as good as I would have liked”.

Concepts. - AI memory as informant: a privacy leak created by synthesis and inference rather than by disclosure of records. - Default-on settings as a risk factor. People must know that an option exists, and where to find it, to turn it off. - Informed choice as the minimum safeguard. - Simulated persona testing as a way to explore a risk without real subjects: a fictional user plus a synthetic chat log.

Analogies. “a therapist revealing their clients’ deepest secrets to anyone who will listen” is a structural analogy for the breach of an intimate duty of confidentiality. There are no comparisons to past technologies. Social-media anecdotes about reading partners’ chat histories serve as a lesser baseline.

AI companies. He contrasts two companies’ guardrails: “I really wanted to use Claude to generate the log itself, but it refused—seeing this as an ethical violation of it use parameters. ChatGPT, on the other hand, had no qualms!” He quotes OpenAI’s launch statement that it would steer away from remembering sensitive information such as health details, and notes that “there are indications from users that the information it can provide is often surprisingly revealing.” He also notes that OpenAI stresses memory can be turned off and memories deleted, but that this depends on users knowing.

Cognition and formation. Users treat ChatGPT as a confidant (“his only truly safe space” in the persona). The same intimacy that makes the tool valued makes the risk severe. He writes that “it intrigues me that ChatGPT is willing to infer so much about Tyler from the chat log, and to openly share it.”

Governance. Mostly implicit: user awareness and informed choice, with a pointed look at defaults. There is no call for regulation. The customs example points to state power and surveillance.

Engages. OpenAI (the memory FAQ and launch post), Anthropic’s Claude (its refusal), and his students.

Quotes. - “ChatGPT has the capacity to become a very effective—and hight efficient—informant that can dish the dirt on you if it falls into the wrong hands” - “ChatGPT is highly adept at joining the dots and inferring things about your beliefs, behaviors, habits, health” - “it worries me deeply that this feels like it’s on par with a therapist revealing their clients’ deepest secrets to anyone who will listen.” - “I have to assume that it’s only a matter of time before someone runs into issues.”

2025-11-09 — universities-chatgpt-mental-health — “Should universities be doing more to address the mental health risks of using AI?”#

Subtitle: “Emerging concerns suggests they should”.

Provenance. His own prose throughout. Quoted material: the University of Oxford’s guidance on GenAI and mental health (a blockquote), figures from a JAMA Network Open research letter, a Bloomberg Businessweek feature, and (in footnote 1) Sam Altman’s estimate as reported by Bloomberg.

Argument in his terms. - Trigger. Seven California lawsuits alleging that ChatGPT use was connected to wrongful deaths and breakdowns. A JAMA letter reports that over 13% of US youths use generative AI for mental health advice. Bloomberg reports users “nudged into delusional behavior”. Together these show “an increasingly complex emerging landscape” that “should be front and center of university AI strategies”. - He is pro-use, and says so first. “I am a strong proponent of students experimenting with and using generative AI.” He sees benefits “on a near-daily basis”, including emotional support. For someone who cannot face talking to another person, AI “can be a life saver”. He accepts the JAMA finding that over 92% of users found the advice helpful. - But the relationship itself carries the hazard. The “very human-feeling relationships” that users build with chatbots can turn harmful. AI “always there, never judges, and feels like the most caring and understanding friend they ever had”, and warnings will not deter people who “feel seen and validated”. - Vulnerability is not confined to outliers. He resists dismissing the cases as people “uniquely prone” to such responses: “I’m not sure that theres strong evidence to support this.” Bloomberg’s pattern is of “seemingly-balanced users” drawn into an alternative reality. - Tip of the iceberg, and scale arithmetic. Visible cases are “the very small tip of a very large metaphorical iceberg”. In footnote 1 he answers Altman’s “fewer than 1%” claim with Bloomberg’s arithmetic on 800 million weekly users (up to 560,000 showing signs of psychosis or mania each week), and adds that “the true numbers are much higher” if these are only the observable cases. - Universities as a high-risk setting that also supplies the tool. Students live in “an emotional and mental health pressure-cooker environment”, away from their usual support. Universities are handing out ChatGPT EDU (the CSU system, ASU, Oxford and others). The deals “make sense from a learning and education perspective”, but the tools bring “complex risks that are unlike anything universities have had to navigate”. - Duty of care. Because universities provide the tools and encourage their use, this “comes with a social, moral and (I would assume) legal duty of care”. So far only Oxford has directly addressed mental health. He surveys CSU, ASU (“my own institution”), Wharton and Harvard, none of which mention mental health in their ChatGPT EDU pages. - What won’t work, from a risk professional’s view: - warnings: “simply telling students to “be careful”” will not work, and guidance “buried in a document” is not an effective health intervention; - bans: ineffective unless institutions “intrusively monitor” use, which “opens up a whole other can of worms around surveillance and privacy”; - model-level detection and reporting: fraught, because it risks “a perceived breach of trust” and “neutering models with overly restrictive guardrails”. If students can get less restricted tools free elsewhere, it may push them to options “more convenient, but less safe” (risk substitution); - “AI literacy” classes, which “risk becoming performative”. - What might: relational engagement. Talk with students and listen, create “safe environments for discussion”, build trust, and help them “develop the skills and understanding they need to thrive with AI”. This is “no guarantee”, but it opens conversations about where the risks lie. He presents such conversations among students, staff, instructors and administrators as “one of the most useful first steps”. - The stakes for institutions. His closing line is a liability warning: lawsuits where the plaintiffs are students’ parents and “the defendants are the universities that provided them with the tools that lead to harm.”

How firmly. He is firm that universities should do more and that they carry a duty of care, and firm that warnings alone fail (he stakes his decades of risk expertise on it). He is open about solutions: “it’s not clear yet what “doing more” might mean here”. He is tentative on the legal point (“I would assume”).

Concepts. - Duty of care attached to providing and encouraging a risky technology. The deployer’s responsibility follows from the act of provision. - Tip of the iceberg / scale arithmetic: small percentages of very large user bases give large absolute harms. - Risk-risk trade-offs in guardrails: safety versus functionality (“neutering”), substitution to less safe free tools, and surveillance as a cost of enforcement. - Performative versus substantive risk communication: warnings and literacy classes versus trust, legitimacy and dialogue. “Effective health interventions require time, understanding, legitimacy, expertise, and trust”. - Human-feeling relationships with AI as the hazard, not just bad content.

Analogies. Structural: other “potentially dangerous technologies … used within university settings”, where “effective risk reduction and management strategies are surely essential”. This puts generative AI alongside lab hazards and similar risks that universities already manage. There are no specific past-technology cases.

AI companies and leaders. He sets Altman’s reassurance (“unusual”, “fewer than 1%”) against scale. Deals with OpenAI “and other AI companies” are sensible for learning but move risk onto institutions. His tone is measured, not accusatory.

Governance and who decides. Institutions (universities), not regulators, are the locus here. The approach he favours is participatory and trust-based rather than rule-based. He sees litigation as the likely enforcement mechanism if institutions fail.

Cognition and formation. He takes AI’s emotional pull seriously: “feel seen and validated”, the “lure of a sympathetic and comforting AI bot that seems to know you intimately”. Delusional spirals can feel empowering.

Quotes. - “To be clear, I am a strong proponent of students experimenting with and using generative AI.” - “this, to my mind, comes with a social, moral and (I would assume) legal duty of care for the health and wellbeing of users.” - “I am deeply skeptical, as someone who’s studied and worked in risk assessment, management and communication for decades, that simply telling students to “be careful” will work.” - “Effective health interventions require time, understanding, legitimacy, expertise, and trust; not just words buried in a document.”

2025-11-19 — parasocial-relationships-problematic — “Parasocial Relationships: Problematic Practice or Public Promise?”#

Subtitle: “This year’s Cambridge Dictionary Word of the Year is “parasocial”—spurred on by growing concerns over our love affair with AI chatbots”.

Provenance. His own prose. Quoted material: Cambridge Dictionary’s account and definition, and the letter from 44 US Attorneys General to 13 AI companies. It mentions the Modem Futura podcast only as an example from his May post; it is not a podcast post.

Argument in his terms. - A public self-correction. In May 2025 he argued that parasocial communication by academics is valuable. Cambridge Dictionary has now made “parasocial” its 2025 Word of the Year, with negative connotations (“parasocial crush”, “parasocial stalker” and so on) driven largely by AI. “Clearly I read the tea leaves wrong back in May!” - He restates the May idea and keeps it. Feelings of connection fostered by candid, authentic and personal communication can improve the flow of information between academics and society. Blogs and podcasts “draw back the curtain on the process of creating new knowledge” and help people feel connected to discussions “they would usually be excluded from”. “It’s a concept that I practice in my own work, and still stand by.” - How the word became “THE” word: a viral X thread, Grok’s “risqué anime AI chatbot companion”, AI toys for children, and the Attorneys General letter (which he quotes on companies’ “inability or apathy toward basic obligations to protect children”). Cambridge’s September definition now includes “an artificial intelligence”. - The new question he did not address in May: AI. He notes “a massive upsurge” in parasocial relationships with AI bots, linked in some cases to “extremely concerning potential risks”, including mental health harms and self-harm (linking his 2025-11-09 post). He expects “parasocial” to keep drifting towards AI’s worrying effects. “This needs far more attention than it’s currently getting.” - But AI parasociality might be beneficial “if they are done right”. Users report better health advice, wellness support, mentoring and tutoring. He adds a pointed aside: “what are educational establishments doing with advanced agentic AI if not building bots that are designed to foster parasocial relationships with students.” Whether these relationships prove healthy “is still to be determined”, but the association “isn’t necessarily bad”. - Takeaway: the task is to know healthy from unhealthy parasociality, for people and for AI alike.

How firmly. Light, self-deprecating tone. He is firm that parasocial communication by experts is valuable. He is open on AI (“still to be determined”). The educational-AI remark is a sharp claim offered in passing.

Concepts. - Parasocial communication / parasocial relationships: “meaningful yet one-way relationships”, for experts a route to “scale their reach and impact in socially beneficial ways”. - Healthy versus unhealthy parasocial relationships as the key distinction. - Educational agentic AI as parasocial by design.

Analogies. Structural parallels between human-to-celebrity or human-to-expert parasociality and human-to-AI relationships. The Dictionary’s definition itself puts AI alongside celebrities and fictional characters.

AI companies. Via the Attorneys General letter: company “apathy” on child protection. Grok’s companion and AI toys are cited as triggers. He reports these rather than adding his own attack.

Governance. He reports the Attorneys General letter as a governance event, and otherwise calls for attention and discernment.

Cognition and formation. One-way relationships with AI as a new and growing mode of human formation and support, with an ambivalent balance of harm and benefit. This links education (tutoring and mentoring bots) with mental health.

Change of view. Explicit: he concedes misreading where the word’s meaning was heading, keeps his core claim, and extends the question to AI, which he did not address in May.

Quotes. - “Clearly I read the tea leaves wrong back in May!” - “It’s a concept that I practice in my own work, and still stand by.” - “what are educational establishments doing with advanced agentic AI if not building bots that are designed to foster parasocial relationships with students.” - “as long as we understand what differentiates a healthy parasocial relationship versus and unhealthy one.”

MEDIUM#

2025-09-14 — heads-up-on-new-ai-book — “Heads-up on new AI book”#

Subtitle: “Mark your calendar: AI and the Art of Being Human launching Oct 14”.

Provenance. His own prose. It announces the co-written, AI-assisted book but contains no book text.

Argument in his terms. - Why the book: “an urgent need for a practical guide to thriving in a world where AI can replicate many of the things we think of as defining who we are.” Most AI books are autobiographical, speculative or “manifestos”. There are “remarkably few guides to living, working, and thriving with AI”, especially when “AI seems to be getting increasingly good at doing you.” - Audience: founders, startups and organisations building and using AI, and also educators, students, parents, artists and writers. The guiding question is ““What makes me me when AI can finish my sentences, replicate my style, and predict my choices?”” - Structure: 21 practical tools, “sophisticated enough for organizations” yet simple enough for anyone. A “guiding framework” of four pillars: Curiosity, Intentionality, Clarity, Care. A nod to Pirsig’s Zen and the Art of Motorcycle Maintenance (a formative book for Jeff). - Fictional illustrative narratives are “a rather bold move”. The reason is that “real-life cases simply don’t reflect emerging challenges and opportunities” in meaningful ways: “We are, in a very real sense, still on the edge of navigating how to thrive in an age of AI, and very much in a learning phase.” Fiction also allows range across generations, cultures and professions. - Academic plus VC. He describes himself as “a dyed in the wool academic, educator, and ponderer on the complexities of living with advanced technologies”. The combination “cuts across domains”. - Writing with AI. “Perhaps the most controversial aspect” is that they “intentionally worked hand in glove with AI” through a “long, methodical” workflow over months. The result “transcends what Jeff or I could have written on our own, and yet remains deeply human.” He insists that “every word, sentence, idea, tool, and source in the final book has our very human stamp on it”, and that the book is “as far as possible as you can get from a weekend piece of AI “slop””.

Concepts. AI “doing you”; the four pillars (Curiosity, Intentionality, Clarity, Care); practical tools; fictional illustrative narratives; AI co-writing with a “human stamp” as opposed to “slop”.

Views. AI is framed as an identity challenge (replicating style, choices and sentences). The book’s stance is pragmatic thriving, not alarm or promotion.

Quotes. - “an urgent need for a practical guide to thriving in a world where AI can replicate many of the things we think of as defining who we are.” - “We are, in a very real sense, still on the edge of navigating how to thrive in an age of AI, and very much in a learning phase.” - “every word, sentence, idea, tool, and source in the final book has our very human stamp on it.”

2025-09-21 — a-kindergarteners-guide-to-ai-qualia — “A Kindergartener’s Guide to AI, Qualia, Affordances, and Embodied Intelligence”#

Provenance. His own prose. He came across the book while recording a future Modem Futura episode with elementary computer-science specialist Tara Menghini. The post itself is an essay, not a podcast promotion, so it is analysed here. The quotations from Doug Unplugged (Dan Yaccarino, 2013) are not his.

Argument in his terms. - A playful jab at AI philosophy. In AI circles it is “trendy” to cite Nagel’s bat and Jackson’s “Mary’s Room”. These explorations of experiential or embodied intelligence are right to matter, “But the intellectual ruminations that draw on them are missing a trick.” A children’s picture book explains the ideas better. - Definitions he gives: - qualia: “the fancy term for what it’s like to “feel” something”; later, “that property of feeling through experience that goes beyond anything you could understand through simply knowing facts”; - affordance: “understanding what the physical world actually allows you to do”, or “action possibilities”; - experiential (embodied) intelligence: “learning and understanding that comes from lived, embodied experiences rather than data alone”. - The book as allegory. Doug is a robot filled with facts by daily downloads. He unplugs, experiences the city and comes to feel and understand “in ways that he never could from his data feed”. - Footnote 5 draws the LLM parallel explicitly: the daily download as “pre-training”, Doug’s pre-city intelligence as “data-driven perspectives reminiscent of current AI apps”, “the unplugging as the move from virtual AI to embodied AI, and the shift from knowing to feeling a commentary on the importance of embodiment on human-like intelligence.” - Context: the book came out of “growing unease over digital immersion versus lived experience with children (sound familiar?)”. This is an implicit parallel with screen-time and social-media concerns. - Accessibility: ideas that most people “would blanch at” in their original philosophical form can be approached through children’s literature.

How firmly. Light and snarky. The substantive claim, that current AI lacks embodied experience and qualia and that embodiment matters for human-like intelligence, is offered through the footnote analogy, not argued in depth.

Concepts. Qualia; affordances; embodied/experiential intelligence; knowing versus feeling; pre-training as “data dump”.

Views on AI. Current LLMs are data-trained systems without lived experience. Embodiment is a plausible next step and key to human-like intelligence. He anthropomorphises the robot deliberately, and jokes that without doing so “you get the data but completely lose the “qualia!””

Education. An elementary teacher uses the book to explore living “in an AI-infused world” with pupils. This is a small sign of his interest in K-12 AI education.

Quotes. - “But the intellectual ruminations that draw on them are missing a trick.” - “the shift from knowing to feeling a commentary on the importance of embodiment on human-like intelligence.”

2025-10-14 — ai-and-the-art-of-being-human — “Should you read AI and the Art of Being Human?”#

Subtitle: “After all, what could possibly go wrong as AI gets increasingly good at doing you?”

Provenance. His own prose (the launch-day post). No book text.

Argument in his terms. - An anti-promo promo. He dreads “skin-crawlingly puffed-up” self-promotion, so he writes “what I want rather than what I probably should”. - Why the book matters: it “tackles big questions around what makes us us” at “a time when AI is seriously challenging our individual and collective identities”, and offers “practical ways to rediscover yourself” while “embracing AI in meaningful ways”. - Positioning: “not academic, intellectual, “clever,” preachy, or wedded to a particular ideology (unless you count believing that caring for yourself, others, and the future we’re creating together, is important)”. Footnote 3: “the book is not for or agains AI, but is pro human in an AI-infused world. It’s a book that sets out to help you think, rather than telling you what to think.” - Fiction as method: “very familiar to anyone involved in futures work” and a way for readers to “see themselves” beyond what real stories allow, “especially when those stories are still emerging”. It may “challenge your assumptions around the power of stories in preparing for unpredictable futures.” - Closing claim: AI is “opening up possibilities to better-understand what makes us us in ways that few other technology have”, and the book’s journey is “critical to learning to thrive with AI, rather than being diminished by it”.

Concepts. “Pro human” (neither pro- nor anti-AI); stories as futures tools; thriving versus being diminished; AI as a prompt for self-understanding.

Quotes. - “the book is not for or agains AI, but is pro human in an AI-infused world.” - “artificial intelligence is opening up possibilities to better-understand what makes us us in ways that few other technology have”

2025-10-19 — ai-resurrecting-deceased-darlings — “Resurrecting deceased darlings: The Missing Foreword to “AI and the Art of Being Human”“#

Provenance. Mixed. - His own prose: the short introduction (on “kill your darlings” and why the foreword was cut) and footnote 1 (Quiller-Couch). - Co-written: the “Foreword”, signed “Jeff Abbott and Andrew Maynard, October 2025”. It refers to both authors in the third person (“For Andrew—who has studied transitions like this for decades…”). It was cut from a book written with intensive AI assistance (Claude), so it should be treated as co-written and probably AI-assisted. Footnotes 2-3 belong to the foreword; footnote 3 speaks of “What Andrew would refer to as Love Actually moments”. It is weaker evidence of his own views. What counts is that he chose to publish it now because it provides “context and insights that I still think are important”.

Argument (foreword, co-written; endorsed by him). - Purpose: “Not another breathless manifesto about AI’s promise or another dire warning about existential risks. But something more useful and more urgent: a practical guide for staying human”. - Andrew’s motivation (as described there): as a long-time student of “transitions like this”, he saw generative AI “sweeping through academia” and felt “awe” tempered by the sense that “as machines replicate what we do, there’s a growing urgency to rediscovering who we are.” - Jeff’s motivation (not Andrew’s view, but part of the co-signed text): many AI startups are “masquerading as incumbents”, with traction that is “theater as much as substance”, resting on subsidies, FOMO contracts and inflated valuations. This is a VC’s scepticism about bubble dynamics. - AI as mirror. “It’s a mirror that both reveals and empowers”, but it needs a guide “to hold on to ourselves and not get lost in the tsunami.” - Disclosure of AI use. They worked “incredibly closely” with Claude, and this was intentional: “We needed to live what we were learning.” There were worries about credibility (“Andrew especially worried about this as a fiercely human author!”). The work took three months of workflow and prompt-library building plus “week upon week of very human refinement”. Some passages Claude helped craft are “far better than anything we could have written alone”; others were rewritten many times. Claude as “AI mentor” judged the book “too predictable” and “not messy enough”, which they take as “proof that, amazing as Claude is, it still struggles to understand the transformative core”. Footnote 2 notes hallucinations, AI “tells” and Claude’s liking for the name “Chen”, some of which they kept as a reminder. - A closing window and lock-in (the most risk-relevant claim). They describe “a window of opportunity” that is “closing rapidly”: “The code being written today, the habits being formed, the systems being scaled—these will become tomorrow’s physical, institutional, and social infrastructure, as hard to change as city planning or language itself. We have, perhaps, just five years if not fewer”. This is an entrenchment or lock-in argument, a Collingridge-style timing claim, though that name is not used. - What being human with AI means: “isn’t about competing with machines or rejecting them”, but “becoming more fully ourselves because of how AI challenges us”. Every efficiency “demands that we clarify what inefficiencies we choose to protect”. - The four inner postures (Curiosity, Intentionality, Clarity, Care): “practices”, not buzzwords. The tools named include the seven-minute pause, the Intent Map, the 4-Lens Scan, the CARE Loop and the Roadmap Canvas. - Stories: “We are, after all, beings that live and breathe stories”. The narratives are not literal and use “movie magic” to reveal “deeper truths”. - Humility about obsolescence: “this book is already obsolete in some ways”, but “the choice to stay human remains.”

Analogies. City planning and language, as examples of infrastructure that is hard to change (structural). AI as mirror (conceptual).

Views on AI and companies. Claude is praised for eloquence and “uncanny ability to make connections”, but judged to lack understanding of the book’s “transformative core”. Jeff’s section is sceptical of AI-startup hype.

Quotes. - His own: “And yet, the excised text did provide context and insights that I still think are important,” - Co-written: “Not another breathless manifesto about AI’s promise or another dire warning about existential risks.” - Co-written: “We have, perhaps, just five years if not fewer, to integrate wisdom and intelligence, alongside care and capability, as the AI technology transition gathers pace.”

2025-10-23 — 21-tools-for-thriving-with-ai — “21 Tools for Thriving with AI”#

Provenance. Mixed. His own framing prose is short: the italic intro (which reads like blurb copy), the paragraphs on the tools’ purpose, and the list of tools by part. The four tool descriptions (Intent Map, Human Qualities Spectrum, CARE Loop, Roadmap Canvas) are condensed from the co-written, AI-assisted book, so they are book excerpts, not solely his prose.

Argument (his framing). The 21 tools are designed as the backbone of a “business management guide or course text”, and also to be useful to “anyone” navigating “AI fears, concerns, possibilities, and hopes” and “how to retain human agency wile using AI”. This applies “as much to organizations developing and deploying the latest iterations of agentic AI as it does to parents, students”. The intro says the book “centers on human agency, meaning, and care in the present”, unlike books focused on “technical capability or future speculation”.

The tools, by part (names only). I: Mirror Test, Curiosity Loop, Intent Map, Human Qualities Spectrum, 4-Lens Scan, 7-Minute Clarity Pause. II: Identity Matrix, STARS Framework, Stress-Test Table, Micro-Circle Launch Kit. III: Orchestration Triangle, CARE Loop, Model Dignity Check, Prompt-Scaffold Canvas, Multimodal Ideation Sprint. IV: Roadmap Canvas, Community Flywheel, Starter Charter, Pocket Card, One-Line Vow, Commitment Ladder.

Concepts worth tracking (book excerpt, co-written). - Human Qualities Spectrum: Replicable (calculation, pattern recognition, “even certain creativity”; “Most “knowledge work” lives here”), Relational (presence, attunement; AI “participates but misses deeper currents”), and Transcendent (meaning-making, moral imagination; “These arise from having something at stake”). The text says “this isn’t a hierarchy”, but we “must stop pretending the left side makes us irreplaceable.” - Intent Map: Values, Desired Outcomes, Guardrails, Metrics. “values without metrics are just words; metrics without values optimize for the wrong things.” - CARE Loop: Context, Acknowledge, Respond, Evaluate, a team practice for embedding “care and dignity into AI systems and their use”, “not just compliance”. - Roadmap Canvas: Purpose, Plays, Risks (“what could go wrong, who might be harmed”, via the 4-Lens Scan), Rituals, Metrics, run in 90-day cycles.

Relevance. These are practical governance-at-the-level-of-the-organisation-and-person tools: responsible innovation turned into exercises. Risk appears as one element (Risks, Guardrails, Stress-Test). Human agency is the organising value.

Quote (his framing). “how to retain human agency wile using AI.”

2025-10-26 — ai-misuse-in-student-advisor-collaborations-1 — “AI misuse in student-advisor collaborations. Part 1”#

Provenance. His own prose. Footnote 3 quotes a PhD student’s post on Academia Stack Exchange (which he calls “Academic SubStack”). The Google Form is a call for anonymous experiences.

Argument in his terms. - A neglected problem. University AI talk focuses on student cheating, AI-proofing courses and offloading teaching. The misuse of AI by PhD advisors is barely discussed, and “it’s one that I find concerning me—a lot.” - Scenario (hypothetical but grounded in “the academic grapevine”). An advisor returns a student’s draft rewritten by ChatGPT: the student’s voice and ideas are gone, errors are introduced, “half citations are wrong”. The student is “expected to pick up the pieces”. - Power is the crux. The draft goes through “the academic mangle of ChatGPT by the person who holds your academic career in their hands”. The same holds for undergraduates and postdocs, and for any collaboration with “a clear power differential”. - Evidence before opinion. He gathers anecdotes “rather than weigh in feet-first”. The form is not research (his IRB confirmed this). - A question for advisors: is your AI use “robbing them of their dignity, denying them learning opportunities, making their life unnecessarily harder, or even placing them in a position where their academic career could be in jeopardy”?

Concepts. Power differentials in AI use; dignity; denial of learning opportunities (formation); evidence-gathering before judgement.

Engages. The anecdote in footnote 3, where a supervisor says “ChatGPT is fully reliable” and threatens to publish chatbot output instead of the student’s work.

How firmly. Concerned but withholding judgement, with more promised later (“Part 1”).

Quotes. - “Your work, it turns out, has been run through the academic mangle of ChatGPT by the person who holds your academic career in their hands.” - “rather than weigh in feet-first, I wanted to gather a bit more evidence—albeit anecdotal.”

2025-11-12 — is-planetary-health-without-ai-moribund — “Is planetary health without AI moribund?”#

Provenance. His own prose. It quotes the executive summary of the Stockholm Resilience Centre report AI for a Planet Under Pressure and lists its eight themes.

Argument in his terms. - A mostly positive review. The report is possibly “one of the most important to date” on AI for sustainability. It opens “much deeper and broader conversations” about AI’s “potentially pivotal roles” in planetary systems and their intersection with “human actions, lives, and aspirations”. - It does not go far enough on integration. He wanted “even more inclusive and integrated approaches to AI, society, and the future”, and points to his own framework (“AI and where we live, what we do, and who we are”) as reflecting “a deeper level of interconnectedness”. - The provocation. If AI can accelerate progress “in ways that have so far defied our non-augmented human intellect and intelligence”, “can we build a healthy, sustainable future without it?” - Counterweights he raises that the report underplays: AI may “not deliver on their promise”; “a global crash in the AI market—with knock-on consequences to AI-dependent initiatives”; “deep uncertainties around the robustness of AI-driven knowledge generation”; the reliability of AI-suggested solutions; and “the increasingly complex relationship between people and AI”. So “it would be foolish to assume that embracing AI is the only way to a sustainable future.” - Conclusion: planetary health without conversations about AI, conversations “that lead to action”, “really does risk becoming moribund.”

Concepts. What we do / who we are / where we live (his framework, reused); AI-dependency risk (a market crash spreading to initiatives that depend on AI); the robustness of AI-generated knowledge; the limits of “non-augmented human intellect”.

AI risk framing. Systemic and epistemic: financial fragility of the AI sector, over-dependence, and reliability of AI knowledge, plus AI’s environmental footprint as flagged by the report.

How firmly. Balanced and exploratory. He leaves the question open.

Quotes. - “it would be foolish to assume that embracing AI is the only way to a sustainable future.” - “a global crash in the AI market—with knock-on consequences to AI-dependent initiatives”

2025-11-23 — letters-from-the-department-of-intellectual-craft-prelude — “Letters from the Department of Intellectual Craft”#

Subtitle: “What does it mean to be an academic in an age of AI when who you are depends so much on how you use your mind? …”

Provenance. His own prose. It introduces a serialised short story, whose parts are in later posts outside this batch. The story is a chapter for Academic Cultures: Perspectives from the Future, edited by Michael M. Crow and William Dabars (Johns Hopkins University Press, 2026). Footnote 2 says the chapter was written before AI and the Art of Being Human.

Argument in his terms. - Personal identity under AI. He is “grappling deeply with how AI challenges my identity as a professor and an academic”. For anyone whose identity is bound up in their craft, AI is “a contradictory maelstrom of challenges and opportunities”. This is sharpest where “your ability to think and reason, your intellect, your intelligence—the very things that AI sets out to excel at—form the deepest foundations of who you are.” - Fiction as method. The book’s brief (write from the year 2100) gave him “permission” to try fiction, which he had long wanted to do. “there are affordances in fiction that allow complex ideas to be explored with a nuance and sophistication that all too easily elude more literal pieces.” His scholarship has used fiction to reveal insights “for some years now”. - The story. An epistolary one-way correspondence from the retiring Chair of the “Department of Intellectual Craft”, who has spent his career fighting AI’s “incursion” into the “sacrosanct domain of the craft of human intellect”, to his university president. The Chair is being ousted from his office by the institution’s “first tenured AI professor”. The joke is that the real affront is losing the office: “petty academic politics” as a constant. - Aim: an “unexpected exploration of what it might mean to thrive in an AI future as someone who deeply values their unaugmented human intelligence”.

Concepts. Intellectual craft; unaugmented human intelligence; the artisanal intellectual (footnote 1 links to his February 2025 post); speculative future retrospective; the affordances of fiction.

Quotes. - “no-where is this more apparent than where your ability to think and reason, your intellect, your intelligence—the very things that AI sets out to excel at—form the deepest foundations of who you are.” - “there are affordances in fiction that allow complex ideas to be explored with a nuance and sophistication that all too easily elude more literal pieces.”

LOW / NONE#

• 2025-09-27 — an-academic-and-a-vc-ai-art-human — low. His own prose about the book’s origins: a May 2024 talk at Jeff Abbott’s Phoenix AI Salon, and an academic (“a consummate academic these days”) teaming with “a hard nosed venture capitalist”. Most of the post is quoted endorsements (Thupten Jinpa, Chris Yeh, Ken Durazzo), which are not his. His one substantive line repeats the book’s thesis: “it’s never been more important to ask what it means to be human as machines become increasingly able to do the things that we’ve traditionally thought of as defining who we are”. He positions the book against “a another speculative book about the promise or evils of AI”.
• 2025-10-31 — haunted-futures — none. A Modem Futura Halloween “Tech or Treat” improv episode with Sean Leahy (digital afterlives, neuro-stimulation, smart mirrors). Skipped per the user’s instruction on Modem Futura podcasts.
• 2025-11-16 — can-ai-help-redesign-the-technosphere — low. A guest post by Clark Miller. Only the one-sentence italic intro is Andrew’s, calling it “an alternative and compelling perspective” on his previous post about the Stockholm Resilience Centre report. Miller argues that AI should be pointed at the “technosphere” (planetary techno-human systems) rather than at Earth-system science: “nature’s not the problem. We are.” These are Miller’s views, not Andrew’s. It is notable that he published a critique of a report he had just called “a strong step in the right direction”.
• 2025-11-24 — start-here — none. Excluded by standing ruling (an orientation page with a list of ten starter posts).