Late Lessons, Jensen Huang and AI
Markdown version

Batch B06 notes (13 posts, 2021-03-28 to 2022-12-13)#

All 13 posts were read in full. This batch has no Modem Futura podcast posts. The 2021-05-28 post announces his Future Rising podcast, which is a different show. Most of these posts started on Medium (the slugs carry hash suffixes) and several were also published in The Conversation or on the ASU College of Global Futures blog. Quotes below were checked against the corpus text.

Relevance summary: - High: 2021-03-28 graphene masks; 2021-04-09 bounded infinities; 2021-08-03 AI risk vs AI ethics; 2021-09-07 Tesla Bot; 2022-02-10 advanced-materials standards; 2022-12-08 ChatGPT and responsible innovation - Medium: 2021-10-05 chaos theory / Jurassic Park; 2022-02-12 Alexa ad; 2022-06-13 Jurassic World: Dominion; 2022-09-16 Midjourney images - Low: 2021-05-28 Future Rising podcast; 2021-06-09 intro video; 2022-12-13 five AI movies

2021-03-28 — how-safe-are-graphene-based-face-masks-b88740547e8c — “How safe are graphene-based face masks?” (HIGH)#

Provenance: His own prose. It quotes Health Canada’s 25 March 2021 memo, links to reviews and a CDC test report, and was updated several times, most recently in an August 2021 note that Health Canada had re-permitted four Shandong models. In the acknowledgement he thanks Jim Thomas of the ETC Group (a civil-society technology watchdog) for raising the issue and for “some of the initial digging.”

Argument: Graphene-coated face masks (Shandong Shengquan “biomass graphene,” sold under many brands) were put on the market without anyone showing whether they shed respirable graphene particles. Health Canada’s preliminary risk assessment pointed to “early pulmonary toxicity.” He sets out what is known about graphene toxicity in a careful, evidence-weighted way: - It is probably less hazardous than carbon nanotubes: it does not migrate to the pleura, it can be broken down in the lung, and mesothelioma is unlikely. - It is still not benign. Jagged edges can damage cells, catalytic metal impurities (iron, nickel) are present, and dose tracks particle number or surface area rather than mass.

His firm claim is procedural. Putting a material with unknown inhalation risk into a product that is all about inhalation, without asking the “basic no-brainer questions,” is irresponsible whatever the eventual risk turns out to be. He is careful not to claim the harm has been shown: “we don’t yet know if graphene particles are being released.”

Concepts and frameworks: - A four-question risk logic for materials: can it get into the body; can it behave in ways that cause harm; what harm, and by what mechanism; how much is needed. - “Irresponsible innovation”: a failure of process, judged independently of the outcome. - “Better safe than sorry” as the “universal rule” that applies when mechanisms are uncertain. This is a mild, practical precautionary stance, not a formal precautionary principle. - The irony of “hundreds of millions of dollars” spent on nano-risk research that is not being applied to real products: the research-to-practice gap.

Analogies and comparisons: Carbon nanotubes, taken literally as a toxicological comparator (fibre pathogenicity, catalyst metals). Occupational aerosol science. The comparison is literal and technical.

Governance and who decides: He puts the burden of proof on manufacturers and regulators, and says uncertainty “will persist until manufacturers and regulators provide data.” He questions what FDA “approval” or product-equivalency claims actually mean, and whether equivalency reviews look at a mask shedding nanoparticles. The regulatory claims look opaque to him.

Expertise and publics: He speaks from 20+ years of nanomaterial risk research. He worries that the episode will “undermine trust and confidence in mask use” and make responsible mask-wearers anxious. Harm to public trust counts as a risk in its own right.

What he criticises: Manufacturers and developers who ignore past risk research (“naivety and disdain for past risk research”). Unclear regulatory approval claims. The research community’s failure to connect its findings to real products.

AI: Not addressed.

Quotes: - “But safer doesn’t mean safe.” - “this is irresponsible innovation on a grand scale — even if the risks turn out to be negligible” - “But when it comes to the risks of using new technologies, hope alone is not good enough.”

2021-04-09 — bounded-infinities-quantum-tunneling-and-the-future-of-education-9a39f7db8812 — “Bounded Infinities, Quantum Tunneling, and the Future of Education” (HIGH)#

Provenance: His own lecture text, the first FRANKx (Frank Rhodes) lecture at ASU, March 2021, first published on the ASU College of Global Futures blog. It opens with a long quotation from Laurie Lee’s Cider with Rosie, which is not his. The header image is credited “ChatGPT/Dall-E” and was evidently added later; it is only an image. He calls the ideas “still very much a work in progress.”

Argument: “We’re approaching one of the most profound tipping points in human history,” and getting through it needs an equally profound change in education, higher education included. It is laid out in six parts. 1. The Solution Problem. Human progress runs on a “problem-innovation-solution-consequences-problem cycle.” Almost every current challenge is the consequence of an earlier solution; agriculture, from breeding to the green revolution to pesticides, is his example. Historically the consequences arrived more slowly than we could innovate fixes. That gave a regime of negative feedback, roughly linear cause and effect, and trial and error that worked. We are now entering the inverse regime, where consequences pile up faster than we can fix them. That brings positive feedback, deep complexity, non-linearity, and the failure of “simple models of innovation.” His examples are climate change, social media, the gap between technical capability and safe and ethical use, and political movements. Innovation is also coming apart from solutions: innovation is pursued as change for its own sake. He says openly that he offers “very little supporting evidence” for this claim of a regime change. 2. Base Coding. The driver of the change is our growing mastery of three “base codes”: digital (bits), biological (DNA, which he stresses is not equivalent to digital code) and material (atoms and molecules, through nanoscale science). The real “game-changer” is “transcoding” between them, “for instance through transcoding between bio and cyber,” with intelligent machines “through a merging of all three.” The danger is “bricking” an organism, an ecosystem or “our own biology” by editing a base code we do not understand. 3. The Paradox of Bounded Infinities. We have infinite possibilities, but not necessarily the ones we need. His illustrations are a universe of odd numbers when the future needs even ones, and the Mandelbrot set. 4. Metaphorical Quantum Tunneling. This is the way through barriers that are impassable under conventional thinking. He says explicitly “I am using this as a metaphor, no more.” Existing tunnelling enablers include faith traditions (koans), the arts and humanities, and “alternate ways of knowing.” Half a century of science, technology and education has weakened them. 5. The Art of Serendipity. Putting unrelated ideas side by side jolts people out of conventional thinking. Future Rising was designed as a “quantum tunneling enabler.” 6. The Future? He asks three questions for higher education. How do we educate “informed, effective and responsible ‘base coders’”? How do we produce “metaphorical quantum tunnelers”? How do we tunnel “beyond the walls of formal higher education” to empower everyone, moving away from “the elitism that still shrouds higher education”?

Concepts: The solution problem; the innovation-consequence timescale inversion; base coding and transcoding (a term he says he has used before); bounded infinities; metaphorical quantum tunneling; quantum tunneling enablers; the art of serendipity. He contrasts what we need to do with what we can do.

Analogies and comparisons: Agriculture, pesticides and the green revolution as a structural example of the solution-consequence cycle. Laurie Lee’s horse, “eight miles an hour,” as a metaphor: conventional technology, and now conventional thinking about science and technology, as a prison. The three base codes put digital, biotech and nanotech side by side as structurally parallel capabilities. Physics metaphors are flagged as metaphors.

AI: Mentioned only briefly: intelligent machines as a merging of all three base codes. AI is framed as one product of transcoding, not as a separate subject.

Risk: Risk is framed at civilisational and systemic scale: consequences outpacing correction. “Sciencing” our way out, good intentions and bright ideas are “no longer enough.”

Education and being human: Education is the main lever. He calls for depth in the disciplines and also for going beyond them, and he criticises the academy as a potential barrier and as elitist. Justice language recurs: a “more just, equitable, sustainable and vibrant” future.

Quotes: - “almost every challenge we currently face as a society has its roots in a previous solution to an earlier problem” - “we’ve substituted what we need to do to thrive, with what we can do” (emphasis on “need” and “can” in the original) - “we are changing the world faster than we understand how to deal with the consequences of our actions” - “If we don’t we risk becoming victims of our own inventiveness.”

2021-08-03 — we-need-to-get-more-innovative-in-how-we-navigate-the-potential-risks-and-benefits-of-artificial-67944f611981 — “We need to get more innovative in how we navigate the potential risks and benefits of artificial intelligence” (HIGH)#

Provenance: His own short prose summary of a talk at a symposium on AI/ML and the human condition, hosted by the National Academies, Los Alamos National Laboratory and the NNSA. The substance is in a video and slide deck that the text does not reproduce. The figure captions (slides) are presumably his.

Argument: There has been a surge in AI ethics guidelines and ethics papers, but “comparatively little work” on the risk implications of AI. A bibliometric check at first seemed to show plenty of AI-and-risk papers. On inspection, most were about using AI to assess non-AI risks, not about the risks AI itself presents. Once those were filtered out, AI risk research looked very thin. He calls these “easy-to-overlook risks presented by AI” and offers the ASU Risk Innovation Nexus’s “innovative approaches to risk” as a way to steer towards beneficial AI.

Concepts: A distinction between AI ethics and AI risk. He thinks ethics is well served and risk is neglected. His subtitle asks whether we are “so obsessed with the rights and wrongs of AI” that we lose sight of risk. Risk Innovation appears only in a slide caption: “navigate complex threats and grow/protect value” (risk as a threat to value).

AI risk: Framed as under-researched and “easy-to-overlook.” The specific risks are not listed in the text. The framing is that of a risk scientist: ethics guidance is no substitute for systematic risk thinking.

Engages: The speakers at the event included Stuart Russell and Fei-Fei Li. He mentions them but does not engage with their arguments.

Quotes: - “there’s been comparatively little work on how we address the risk-implications of artificial intelligence” - “the vast majority of papers focus on how to use AI in more effectively assessing and managing non-AI risks, rather than addressing the risks presented by AI”

2021-09-07 — should-we-be-worried-about-elon-musks-tesla-bot-58dd3aa3c3c5 — “Should we be worried about Elon Musk’s Tesla Bot?” (HIGH)#

Provenance: His own prose. An edited version appeared in The Conversation.

Argument: Fear of Terminator-style robots is the least of the problems. What matters more are: 1. Tangible risks. The Bot is built on Tesla’s full self-driving and Autopilot stack, which has a record of crashes (including failing to recognise parked emergency vehicles). The problems are made worse by human behaviour: drivers pushing the limits. 2. Orphan risks. Privacy, autonomy and identity; how people respond to humanoid robots; ideological mismatches; performance hacks; new forms of weaponisation; job security; bias (“an army of racist, sexist Tesla Bots”). He judges these “manageable.” 3. The deeper worry. The “audacious plan” behind the Bot. Musk is a “future-builder” whose ventures are steps towards a vision of technology as humanity’s saviour: cheap energy, merging with machines, becoming interplanetary, “transcending our evolutionary heritage.” The worry is that wealthy innovators with “the vision, the resources, and the power” impose futures that most people do not want, or that are “catastrophically flawed.”

His tone is balanced. The plan “makes sense” in a Musk-esque way, the technologies “could open up a future full of promise for billions,” and Musk should be able to “flex his future-building muscles.” His concern is about responsibility and who takes part.

Concepts and frameworks: - Orphan risks, defined here as “risks that are hard to quantify and easy to ignore, and yet inevitably end up tripping innovators up if overlooked.” He links to an earlier Medium piece on orphan risks for tech startups and their funders. The concept already existed by 2021 and sat inside his Risk Innovation work. - The Risk Innovation three questions: “What’s important to your enterprise, your investors, your customers, and the communities you touch? What orphan risks threaten these things of importance? And what small steps can you take” to navigate them. Risk is framed as a threat to what matters. - Just because we can, should we?: the could-versus-should framing. - “Whose future?”: who gets to imagine the future. - “Make tech more human”: his description of Musk’s approach, with technology designed for a “world built by humans for humans” as a stepping-stone to superhuman technology.

Nature of AI: AI is embodied (“an AI ‘brain’”). Humanoid form follows from a human-built world. It is an enabling “cross-cutting technology” in Musk’s “universal toolbox,” which also keeps the company “several steps ahead of the competition (and regulators).”

AI risk: Several layers: near-term safety and reliability failures; sociotechnical orphan risks; learned bias (Microsoft’s Tay is linked); and, most serious, a power and vision risk: the concentration of future-making in a few people. Sci-fi robot-uprising risk is played down.

Tech leaders and companies: Directly critical of how Musk and other “powerful entrepreneurs” “impose their visions of the future on us.” He is sceptical of the naming (“dubiously-named autopilot tech”) and of releasing technology “into the wild so soon.”

Governance and who decides: The answer is broad participation. We should all be asking the Risk Innovation questions. Regulators are mentioned only as something Musk stays ahead of.

Science fiction: Dystopian robot films are used conceptually. Their “abiding lesson” is to be reread as being about who gets to imagine the future.

Quotes: - “technological competency is often the least of a developer’s problems” - “we risk handing our futures to innovators whose vision exceeds their understanding” - “the far larger challenge of deciding who gets to imagine the future, and be a part of sharing in building it?”

2021-10-05 — butterflies-chaos-theory-jurassic-park-and-a-nobel-prize-in-physics-1a660630ca7d — “Butterflies, Chaos Theory, Jurassic Park, and a Nobel Prize in Physics” (MEDIUM)#

Provenance: A short new introduction and a new addendum (2021), wrapped around a long excerpt from chapter 2 of his sole-authored book Films from the Future (2018). The excerpt is his own prose, written around 2017 (it mentions Hurricane Harvey “as I’m writing this”). It is a republication of earlier thinking, prompted by the 2021 physics Nobel.

Argument: Chaos theory (Lorenz, the butterfly effect; the Mandelbrot set; Gleick) shows that in complex systems tiny changes can have large, unpredictable effects, so we cannot wield perfect control over complex technologies. It also shows bounds (“predictable chaos”) and points of stability. These separate “plausible futures from sheer fantasy” and suggest that some better futures can be reached, or wasted, depending on foresight. His real-world example is the Arkema organic peroxide plant in Houston during Hurricane Harvey: planned contingencies (flood, power loss) were overwhelmed by small, unplanned cascades (overflowing toilets, a snake). He also cites Perrow’s “normal accidents.” Finally, he says that power plays an outsized role in steering technology and its fallout. This leads into the next section of the chapter, which is not reproduced.

Concepts: The butterfly effect; bounded unpredictability; plausible versus fantastical futures; Perrow’s normal accidents; power as a driver of technology trajectories. He is critical of the film’s “hokum” (the idea that chaos theory lets you predict when chaos will occur).

Analogies and comparisons: The chemical industry (Arkema) as a literal case of a complex-system accident. Jurassic Park as a conceptual cautionary tale of “naïve human arrogance.” Structurally, chaos theory is set beside quantum physics as a second blow to the assumption of predictability.

AI: Not addressed.

Change signalled: In the addendum he regrets leaving out Bradbury’s “A Sound of Thunder.” This is a small self-correction about sources, not a change of view.

Quotes: - “the concept that we cannot wield perfect control over complex technologies within a complex world is nevertheless an important one” - “the way that power plays an oversized role in determining the trajectory of a new technology”

2022-02-10 — are-we-asking-the-right-standards-questions-about-advanced-materials-c2eb7fd72849 — “Are we asking the right standards questions about advanced materials?” (HIGH)#

Provenance: His own opening remarks to the ANSI Nanotechnology Standards Panel Advanced Materials workshop, 20 August 2020, posted in 2022. The footnotes cite his own papers: Nature 2011, “Regulators: Don’t define nanomaterials”; Maynard & Kuempel 2005; and Maynard, Warheit & Philbert 2011, “The New Toxicology of Sophisticated Materials.”

Argument: Standards work needs to separate terms of art (“norms, expectations, opinions and perceptions that are not necessarily grounded in evidence, but that nevertheless grease the wheels that the world runs on”) from terms of science (“evidence based… derivable… traceable… insulated from opinion”). Treating a term of art as a term of science produces rationalisation, like claiming a definitive answer to “how long is a piece of string?” - “Nanotechnology,” “nanomaterial” and “nanoparticle” are terms of art. The 100 nm cutoff “is a number of convenience, not of science.” - Nano health standards moved away from an older, evidence-based and outcome-based tradition in aerosol and occupational science going back to the 1950s. That tradition defined hazard by the ability to reach vulnerable parts of the body and by physical and chemical mode of action. Nano standards put the label first and then treated it as if it were science, which risks “a house of cards.” - “Advanced materials” is even more clearly a term of art: “no fundamental scientific basis… dependent on context… temporal.” - Standards should begin with purpose, and for health and the environment with function and behaviour: exposure and dispersion, then biological interaction. A behaviour-based approach also catches old materials used in new ways and avoids assuming that new means risky.

He also asks for evidence that nano standards have produced measurable good outcomes, and hopes that participants have it.

Concepts and frameworks: - Terms of art versus terms of science. - Behaviour-based (function-based) rather than definition-based risk governance. - “Sophisticated materials” (2010/2011). This was a deliberate new label chosen to escape the baggage of “nanomaterials,” and he jokes that it cost citations. It comes with five categories of materials likely to “slip under the conventional risk radar”: - abrupt scale-specific changes in behaviour; - crossing normally protected biological barriers; - active, context-responsive materials; - self-assembling materials; - materials whose mechanisms are not captured by conventional hazard assessment. - Purpose-first standards: why standards, how success will be known, and what happens without them.

Analogies and comparisons: A literal lineage from nanomaterials to advanced materials, grounded in aerosol and occupational inhalation science. The “piece of string” is a metaphor. He does not apply the argument to AI.

Governance: He supports standards where they are fit for purpose and evidence-based, and warns against building governance on definitional categories. This is an enduring position, anchored in his 2011 Nature piece.

Quotes: - “100 nm, no matter how useful it is, is a number of convenience, not of science.” - “nature doesn’t care what we call a material, it just cares about how it behaves.” - “there is no reason to assume that new materials, by default, present new risks”

2022-02-12 — scarlett-johanssons-amazon-alexa-super-bowl-ad-may-be-fun-but-it-s-also-scary-cc11d2913707 — “Scarlett Johansson’s Amazon Alexa Super Bowl Ad May Be Fun, But It Also Raises Serious Questions” (MEDIUM)#

Provenance: His own prose. The pull quotes are repeats of his own sentences.

Argument: Amazon’s joke about a mind-reading Alexa points to where the industry is actually going. Smart speakers are nodes in “an interconnected web of sensors and data” whose real purpose is to analyse, predict and meet needs in “a seeming-quest to make us all ever-more-dependent super-consumers.” - Evidence that combined data sources are powerful: fitness-tracker maps revealing military bases; a beer app tracking personnel; Palantir predictive policing at the LAPD; racial bias in criminal-justice algorithms. - The trajectory: Alexa Hunches; voice biomarkers for dementia, Parkinson’s, cardiovascular disease and PTSD; devices that may tell “others” about your health or mind; eventually a link to brain-computer interfaces such as Neuralink.

He is clear about how firm each part is. He says outright that “much of this is speculation.” He holds on to the gap between inferred and actual thought, and insists that the danger lies partly in the seduction of believing machines can read minds.

Concepts: Data fusion leading to predictive inference; the “super-consumer”; conditioning (“consumers are being conditioned to believe that they need these technologies”); the gap between machine inference and human inner life; competitive pressure as the driver (“maintaining a competitive edge is everything”; “there’s little to stop the increasingly invasive use”).

Nature of AI: AI appears as prediction and inference from data trails: statistical anticipation, not understanding (“a high-tech sleight of an AI’s hand”).

AI risk: Privacy and intrusion into inner states; dependence; flawed or biased prediction used in high-stakes decisions (policing); disclosure of health or mental states to third parties.

Companies and leaders: Critical of Amazon’s market logic, and he asks whether Bezos is “hiding his long-term plans in plain sight.” Neuralink and Musk appear as the next step.

Cognition and formation: An early version of later themes. Technology that reaches into mind and body, and consumers being conditioned into dependence. He does not use the word “manipulation,” but the logic of predictive nudging towards consumption is there.

Governance: Implied only, in the closing question of whether Amazon’s own judgement (“bad idea”) is “enough to keep Amazon from going there.”

Quotes: - “a seeming-quest to make us all ever-more-dependent super-consumers” - “there remains a gap between what machines think we’re thinking, and what we are actually thinking”

2022-06-13 — jurassic-park-dominion-may-fall-short-on-the-science-but-its-social-commentary-is-worth-heeding-fc48c9344e7d — “Jurassic Park: Dominion may fall short on the science, but it’s social commentary is worth heeding” (MEDIUM)#

Provenance: His own pre-editorial draft of a Conversation article, published explicitly to show how editing changes a piece. That is itself a reflection on writing.

Argument: For all its hype, the film carries the franchise’s “cautionary tale of technological hubris.” In the real world, genetic engineering has moved faster than Crichton imagined: the Human Genome Project; Endy’s engineering approach to DNA and synthetic biology; DIY bio and iGEM; CRISPR; gene drives; biomanufacturing; gain-of-function research; mRNA vaccines; de-extinction (Church’s mammoth). On responsibility the record is fairly good: ELSI built into the genome project, responsibility built into iGEM, and caution in DIY bio. But as the technology grows more powerful and accessible (“aided and abetted by advances in artificial intelligence”), “a community of well-meaning scientists and engineers are unlikely to be sufficient.” He is even-handed on the lab-leak question, calling the rumours “unsubstantiated” while noting that they renewed the gain-of-function debate.

Concepts: Could versus should (Ian Malcolm’s line). DNA as biological “source code” that can be digitised and edited, which continues his base-coding idea. Advances in responsibility have to keep pace with technical advances. Power-hungry people combined with powerful technology.

Analogies and comparisons: Biotech is the subject here, not an analogy. There is a structural link to AI: AI accelerates biotech and makes it more accessible. Viruses versus dinosaurs: size does not track harm.

Governance: Self-governance by scientific communities is praised, but will not be enough as access widens. He does not say what should replace or supplement it.

AI: Mentioned only as an accelerant of biotech.

Quotes: - “even with good intentions, bad things happen when you mix powerful technologies with power-hungry people” - “a community of well-meaning scientists and engineers are unlikely to be sufficient”

2022-09-16 — 56-stunning-ai-generated-images-inspired-by-the-future-of-being-human-6d3ef5cd6674 — “56 Stunning AI-Generated Images Inspired By The Future of Being Human” (MEDIUM)#

Provenance: The text is his own prose. All the images were generated by Midjourney from one prompt, “Will we be able to design and create synthetic consciousness in the future?”, taken from the ASU Future of Being Human initiative website. They were then iterated and curated by him. The images are not evidence of his thinking. His framing of the process is.

Argument: After weeks of experimenting, he is “blown away.” AI image tools “unleash human creativity,” especially for people “whose technical skills lie far behind our imagination.” The process is “co-creation”: he prompts the tool and is prompted by it in turn. He is not worried that AI will make human art redundant, because art’s value lies in human perception and meaning-making, and “art, to humans, will always have humans in the loop somewhere.” He notes that “the interpretation is all on my side here, not the AI’s.” He is tentative about whether the outputs are art (“debatable”).

Concepts: Human-AI co-creation; iterative curation; human-in-the-loop; AI as a creativity catalyst; “transformative AI-human collaborations.”

Nature of AI: A tool and collaborator that produces variations. Meaning and interpretation stay human.

AI risk: Only the risk to human creativity and artistry, which he mostly dismisses here. It is an early, enthusiastic stance that comes before the ChatGPT era.

Being human: The question of synthetic consciousness is the prompt, but he does not explore it in the text.

Quotes: - “these machine learning image creation platforms have an amazing ability to unleash human creativity” - “Art, to humans, will always have humans in the loop somewhere.”

2022-12-08 — i-asked-open-ais-chatgpt-about-responsible-innovation-this-is-what-i-got-c0f4bfe14776 — “I asked Open AI’s ChatGPT about responsible innovation. This is what I got” (HIGH)#

Provenance: Mixed. The four answers (definitions of responsible innovation and public interest technology; why study the societal consequences of technology; why companies should hire people with social, ethical and legal expertise) are ChatGPT-generated and excluded as evidence. The questions, the evaluation and the framing are his own prose. What he chose to publish is telling: he treats ChatGPT’s answers as clearer than most expert writing in his own field.

Argument: One week after ChatGPT’s release, he finds its “ability to synthesize information” and explain complex ideas accessibly “a potential game changer.” The answers contain “not new information here, and no novel insights.” But “you’d be hard pressed to find experts” who could write as clearly. This opens up “machine-augmented communication and engagement” for science communication, public engagement, informing policymakers, and even colleges explaining their courses.

He then adds a “meta-layer.” Machines that interact “in very human ways while seeming informed, engaged, and attentive” bring a “raft of risks that we’re just beginning to unpack”: unhealthy attachment, gullibility, algorithmic bias, misinformation, the propagation of disinformation, and hate speech. Handling them needs new thinking about developing such technologies responsibly and ensuring social as well as economic value. He calls it a “serendipitous irony” that AI can help explain the concepts (responsible innovation, public interest technology) needed to govern it, and says “it’s one we should be running with.” He ends: “As long as we do it responsibly!”

Concepts: Machine-augmented communication; human-ChatGPT partnerships; responsible innovation; public interest technology; AI as a communication amplifier.

Nature of AI: A fluent synthesiser and communicator. Its conversations “feel authentic” and it seems human, but it is not a source of new insight. What is new about it is clarity and accessibility at scale, plus human-seeming interaction.

AI risk: Four families of risk: relational and psychological (attachment, gullibility); bias; the information ecosystem (mis- and disinformation, hate speech); and the general need for responsible development. Serious, but framed as a landscape “we’re just beginning to unpack,” not as catastrophe.

Cognition, language and formation: Early seeds of later work. Human-seeming conversational machines create risks of attachment and credulity. At this stage the language capability is mostly celebrated as good for public understanding.

Expertise and publics: A mild criticism of experts, including his own field, for “convoluted” explanations. He sees AI as a way to widen engagement.

AI in scholarship and writing: This is his first public experiment with ChatGPT, publishing AI output as data in an experiment. The pattern recurs in his later work.

Quotes: - “you’d be hard pressed to find experts who could give responses that are as clear and understandable as these” - “the dangers of unhealthy attachments and gullibility” - “something of a serendipitous irony in AI being instrumental in providing pathways to overcoming the challenges it presents us with” - “a leap in machine-augmented communication and engagement”

Low / none#