B10 notes: 2023-07-25 to 2023-09-01 (18 posts)#

Reading notes on Andrew Maynard’s Substack posts in batch B10. Only his own prose is treated as evidence. Quotes are exact, including his original typos and curly punctuation.

Batch context. ChatGPT is about eight months old. Maynard has just finished teaching his first online undergraduate course on prompt engineering (all GPT-4), and a new academic year is about to start. He is also serialising The Moviegoer’s Guide to the Future, a weekly podcast in which he reads chapters of his 2018 book Films from the Future (episodes 4 to 9 fall in this batch). The mirror holds only his short written introductions to these episodes, not the audio. The batch contains no Modem Futura podcast posts, so the user’s instruction to skip those removed nothing here.

Most of the batch is about AI in education and learning. Alongside that run a thread on AI consciousness and personhood, a reflection on how ideas form in human heads and in LLMs, and a direct AI–nuclear comparison.

Relevance summary:

HIGH#

2023-08-14 — chatgpt-stimulates-creativity-critical-thinking — “If you’re obsessed with ChatGPT’s accuracy, you’re missing the point”#

Provenance. His own prose. It includes one block quote from Ethan Mollick (One Useful Thing) and links to Gary Marcus, Analytics India and Rolling Stone in a coda. This is his first-hand report on the course described in 2023-07-16 chatgpt-created-my-course.

Argument in his terms. - The evidence base is his own reading. Over six weeks he read “over 2,000 conversations” between ChatGPT (GPT-4) and 72 undergraduates, each completing 30 graded assignments, in his online basic prompt-engineering course. He calls it a “ChatGPT baptism of fire”. - He teaches the limits explicitly. “ChatGPT makes stuff up, it gets things wrong, it’s inconsistent, and sometimes it’s frustratingly hard to get it to do what you want it to.” - Central claim: the flaws help. ChatGPT is “a profoundly effective catalyst for engaged and creative thinking”, and in some cases effective “because of its limitations”. Because the output cannot be trusted, users have to engage their critical thinking, “at least if they understand what they are doing”. - What he saw in students. They thought more deeply and creatively. They became curious and got better at asking questions and “test the answers”. They took deep dives, made “serendipitous connections”, used ChatGPT “as a sounding board”, and learned in self-directed ways. Students who struggle with conventional education told him it created a learning environment “uniquely responsive to them”. - A caveat he states. The course was deliberately designed to stimulate thinking. What surprised him was how much ChatGPT amplified that. - Mechanism. LLMs “tap into deep wells of human knowledge and insight — far more effectively than any individual can”. They are articulate and adjust to the user’s level. Two features matter: long, coherent conversations, and explanations pitched in accessible language at the right level. - The educational frame. This is learning how to think, not what to think, which he calls “a central tenet of modern education”. He attacks the “deluge” of articles faulting ChatGPT on facts, and the “rather unimaginative idea” that a 2 + 2 error makes it “a failed technology”. - Conditions. It needs AI literacy for instructors and students, since “there are plenty of “not smart” ways of using the technology!” - Hedging. He jokes that he may be “suffering from an overdose of ChatGPT”, then says “But I don’t think so.” He senses “something unexpected and emergent about generative AI that shakes the cage of conventional thinking”. - Coda. He cites Gary Marcus’s essay (“What if Generative AI turned out to be a Dud?”; he calls it “worth-reading”), an Analytics India piece predicting OpenAI’s bankruptcy, and Rolling Stone’s feature on five women who warned about AI: Timnit Gebru, Joy Buolamwini, Safiya Noble, Rumman Chowdhury and Seeta Peña Gangadharan. The landscape is “increasingly complex” and delivery “by no means certain”. He stays optimistic: - hallucinations are not “a limiting factor”, though “greater accuracy and trustworthiness would be good”; - even an economic hiatus would not stop the technology, since “we’ve already stepped over the threshold of understanding”; - still, “there are risks here — moral as well as social and economic”.

How firmly. Strongly and enthusiastically, based on his own classroom observation, not systematic study. He hedges with self-irony and takes the critics seriously enough to report them.

Concepts. AI as catalyst for creative and critical thinking. Flaws as a feature for learning. The “knowledgeable and articulate yet imperfect AI companion”. AI literacy. Learning how to think, not what to think. Tailored learning at scale (“how education is scaled beyond it’s current institutional limits”). Access: “who gets access to teaching that’s tailored to their needs”.

Comparisons. None to past technologies.

On AI (what kind of thing it is). A “brilliant but flawed” system that draws on collective human knowledge and is able to “connect with them on a very human level”. Emergent and cage-shaking. He treats it as a conversational partner, not an information source.

On AI risk. Accuracy is explicitly not the main issue for learning. He names moral, social and economic risks without detailing them. He reports concerns about bias and “morally questionable processes” without dismissing them. He also names economic viability (OpenAI’s finances).

On companies. Only indirectly, through the uncertainty about OpenAI’s viability.

On cognition and formation. This is his strongest positive claim so far about AI and human cognition: well-used AI “switches students’ minds on”. His worry is about unskilled use, not about AI weakening thinking. It is a useful benchmark for comparison with later batches.

Criticises / engages. Accuracy-obsessed commentators; “outmoded perspectives on education”; “narrow minded assumptions of what AI should do”. He engages Mollick (agreeing), Marcus and the Rolling Stone critics (respectfully).

Change of view. None signalled. His enthusiasm grows out of the course experience.

Quotes. - “ChatGPT is a profoundly effective catalyst for engaged and creative thinking” - “what generative AI is really good at isn’t about accuracy, but about how it enables us to think and learn differently.” - “brilliant but flawed AI systems that are able to connect with them on a very human level.” - “a commitment to ensuring the technology serves society, rather than the other way round.”

2023-08-21 — the-messiness-of-the-provenance-of-ideas — “The incomparable messiness of the provenance of ideas”#

Provenance. Mostly his own prose, with three non-Maynard elements: - a block quote from Mark Daley (then CIFAR Vice President for Research); - a list of terms ChatGPT produced when he asked it for alternatives to “provenance of ideas” (Ideation Tapestry, Mental Mélange, Cognitive Archipelago, Brain Brew and Ideas Rhizome); - ChatGPT’s definition of the “Ideas Rhizome”, which he quotes as ChatGPT’s.

The critique of Deleuze and Guattari, the fungal-mycelium metaphor and the application to LLMs are his.

Argument in his terms. - The prompt. He is “appallingly bad” at recalling how his thinking evolved and who influenced it. Meanwhile debate is growing about the provenance of generative AI outputs (he cites Alex Reisner in The Atlantic on pirated books in training data). He says there are “important questions that need to be addressed when it comes to using people’s work without permission”. “Yet on some level”, LLMs’ assimilation of vast information “feels very close to what we all do”. - His own case. In 2021 he helped CIFAR frame a call around “The Future of Being Human”, and a year later launched ASU’s Future of Being Human initiative and this Substack. He worried he had “inadvertently coopted” CIFAR’s framing and checked with Daley. Then, recording the Ghost in the Shell chapter for the podcast, he found he had subtitled it “Being Human in an Augmented Future” in 2017–18 and had forgotten. His books, the initiative, the Substack and even the CIFAR call are “merely the visible fruits of a messy and largely hidden network of influences”. - Attribution has limits. He is uneasy with the phrase “provenance of ideas” (philosophical “baggage”; it implies attribution is possible). Attribution works where ideas are built “brick by brick” and codified. That is “the bedrock of academic scholarship, intellectual property, and ownership over ideas”, which his academic writing respects. It fails for “chance meetings, random inspirations, serendipitous encounters, and the echoes of long-forgotten conversations”. His own thinking is “something of a black box — even to me.” He asks whether attribution is “a rather outmoded idea”, as a question, not a claim, especially since generative AI is now “part of the mix of influences that inform and modulate our thinking”. - Using ChatGPT. He turned to ChatGPT “in a very self-conscious twist of irony”. It was “helpful as a sounding board” but “somewhat limited intellectually”. - Deleuze and Guattari. He likes the rhizome idea but criticises how it “extends a biological metaphor in overly confident yet intellectually convoluted ways to the social construction of meaning”. He recalls the 1990s critique of postmodernist appropriation of natural science. - His own metaphor: fungal mycelia. Threads that are “probing, connecting, interrogating, assimilating, and ultimately transforming”. Coherent ideas occasionally fruit from a largely impenetrable web. The image emphasises “collective and intertwined systems” in the creation of knowledge. - Applied to LLMs. GPT-4 is exposed to myriad artifacts that form “a mycelium-like mat of associations, connections, and inferences”. Its outputs are the fruits of this “ideas mycelium”, with “very little direct traceability”. Provenance lies in the matrix of “stuff”, not in particular sources. - The governance implication. Who owns the “stuff” and who authorises its use matter, as do its breadth and diversity. Just as his own exposures bias his thinking, “the composition of the “stuff” that generative AI is built on leads to biases and influences in its outputs”. So “great care” is needed to understand and, “where necessary, oversee” what goes in. - Conclusion (tentative). “maybe we need to accept the increasing messiness of the provenance of ideas”. What matters is the “shape, structure, and composition” of the ideas mycelium, “rather than increasingly hard to identify bright lines between input and output.”

How firmly. Exploratory and self-deprecating. The claims are posed as analogies and questions (“I wonder”, “maybe”).

Concepts. Provenance of ideas (a term he dislikes). The “ideas mycelium”. The human mind as a black box. Composition of training inputs as the source of bias. Attribution as possibly outmoded in non-linear idea formation. AI as part of the “mix of influences” on human thought.

Comparisons. A human–LLM analogy for idea formation, used conceptually. He limits it himself: generative AI “seems to mirror this — at least in part”. There is a biological metaphor (mycelium), and he admits it goes “farther than is probably wise”. No past-technology comparisons.

On AI. An LLM is an associative matrix built from human artifacts. It does not “regurgitate” sources traceably; it fruits from a mat of associations. Implicitly, it is closer to human idea formation than to a database.

On AI risk and governance. Bias from the composition of training data. Copyright and consent over training data are acknowledged as real. Oversight of inputs is preferred to input-to-output attribution. The approach looks at the whole system, not at individual sources.

On cognition, language and formation. A key early statement that generative AI is now one of the influences shaping human thinking, and that human idea formation is itself largely opaque, collective and non-linear. This connects his being-human thread to his later interest in how AI mediates the formation of ideas.

On scholarship. Academic attribution is valued, but is shown to rest on the codified, traceable part of how ideas form.

Criticises / engages. Deleuze and Guattari’s metaphor-stretching. He engages Reisner (Atlantic) and Daley (CIFAR).

Change of view. A personal revision of history: his “future of being human” thinking predates the 2021 CIFAR framing, going back to the 2017–18 Ghost in the Shell chapter.

Quotes. - “the assimilation and use of vast amounts of information feels very close to what we all do” - “the process of how I got to where I am in my thinking is something of a black box — even to me.” - “especially as generative AI is increasingly a part of the mix of influences that inform and modulate our thinking.” - “the composition of the “stuff” that generative AI is built on leads to biases and influences in its outputs.”

2023-08-23 — could-we-build-conscious-ais-in-the-future — “Could we build conscious AIs in the near future? Quite possibly”#

Provenance. His own prose. Much of it summarises the arXiv preprint by Patrick Butlin, Robert Long and colleagues, including Yoshua Bengio, Stephen Fleming and Megan Peters (arXiv 2308.08708). It also cites Thomas Metzinger’s 2021 call for a global moratorium on “Synthetic Phenomenology”. His own evaluations and extensions are identified below.

The paper, as he reports it. No current AI is likely to be conscious, but conscious AI could be built soon with existing or near-term technology. There are three assumptions: - computational functionalism; - scientific theories of consciousness can describe the relevant features; - consciousness can be assessed against theory-derived conditions.

The paper offers 14 indicators. It warns about both over-attribution (muddying the waters, social disruption, the temptation to assume that “systems like text-based generative AI “feel” human”) and under-attribution, and is more worried about under-attribution.

His own claims. - The paper is not “flights of fantasy around machine self-awareness, sentience, or existential risk” but “rigorously reasoned”. He deliberately separates it from x-risk discourse. - The prospect is “profound”. It is “a step toward the artificial construction of entities that were previously believed to be the sole domain of evolution-driven biology.” - It raises two kinds of challenge: to “our concepts of what it means to be human”, and to “the moral responsibilities that come with being the creators” of conscious machines. - Computational functionalism: “This is quite a bold assumption, but it’s one that stands up to scrutiny.” It implies substrate independence: biological, silicon, “lab-grown biological, bio-digital, or even quantum” systems could in principle be conscious. He endorses the assumption; he does not just report it. - Where he goes beyond the paper: - He names how “tensions begin to emerge between the the economic expediency of denying consciousness, and the moral responsibility to not inflict suffering”. - He widens suffering beyond “biological exceptionalism” and physical pain to “lack of freedom, restriction of agency, loss of sense of self, marginalization, cognitive dysfunction”. - Irony. The rubric for detecting AI consciousness “could also be used as a roadmap for achieving this”. - Forecast. There is “a high likelihood that this is an avenue of development that will continue to grow”, whatever the moral concerns. - Governance. The paper calls for research; “I’d probably go further and suggest that we need more than just research here.” Society must prepare “moral, ethical, and practical norms”, “policies and regulations”, and ways to “make sense of what it is to be human when we can create non-human consciousness”. - Bottom line. “The one thing we cannot afford to do is to deny that this is a possibility or, worse, devise ways of enslaving AIs that are able to understand and be impacted by what this means — under the claim that they are “just machines”.”

How firmly. He is cautious on feasibility (“Quite possibly”) and firm on the moral conclusion.

Concepts. Computational functionalism and substrate independence. Over- and under-attribution of consciousness. Economic expediency versus moral responsibility. A broadened concept of suffering. AI enslavement. Rubric as roadmap. Societal preparedness beyond research.

Comparisons. Biology and evolution as the former sole source of consciousness. There is an implicit analogy between withholding consciousness recognition and “marginalization”. No past-technology comparisons.

On AI. AI may become a moral patient: a non-human, not necessarily human-like consciousness. Consciousness does not imply free will, intelligence or emotions.

On AI risk. Here the risk runs toward the machines as much as toward us: moral harm done to conscious AI for commercial convenience. He also names risks to human self-understanding and social disruption from over-attribution. Explicitly not existential risk.

On companies. Implicit only: “economic expediency” points to commercial incentives to deny AI consciousness. No company is named.

On governance. Norms, policies and regulations, prepared in advance. He cites Metzinger’s moratorium without endorsing or rejecting it.

On being human. Creating non-human consciousness forces a redefinition of what it is to be human.

Change of view (signalled by the sequence). On 2023-08-18 he wrote that “we’re still a long way from machines that have consciousness and self-awareness”. Five days later he calls the near-term arguments “compelling”. Two days after that (2023-08-25) he writes of getting “ever-closer to developing machines that have awareness and agency”. The paper seems to have moved him within a week.

There is also a tension with the 2018 Ex Machina chapter he republished in April 2023 (see B08). That chapter judged human-like machine intelligence implausible because it would need radically different computing substrates. Here he accepts substrate independence for consciousness. The topics differ (consciousness versus superintelligence), but the substrate reasoning has changed. He does not comment on it.

There is continuity with the 2018 worry (republished 2023-04-16) about “what rights they in turn may have”, now expressed more strongly.

Quotes. - “This is quite a bold assumption, but it’s one that stands up to scrutiny.” - “tensions begin to emerge between the the economic expediency of denying consciousness, and the moral responsibility to not inflict suffering.” - “I’d probably go further and suggest that we need more than just research here.” - “devise ways of enslaving AIs that are able to understand and be impacted by what this means — under the claim that they are “just machines”.”

2023-07-25 — oppenheimer-and-ai — “Oppenheimer is as relevant to the future of AI as it is nuclear weapons”#

Provenance. His own prose. He links Christopher Nolan’s remark about AI’s “Openheimer moment” [sic] (Guardian), The Conversation on AI and elections, the Bulletin on a US–China AI arms race, and ASU Threatcasting 2017 on AI weaponisation. The afterword links to the updated Films from the Future resources.

Argument in his terms. - The film shows that “the trajectories that powerful technologies take are deeply intertwined with very human power dynamics, politics, and personal beliefs.” That is what carries over to AI. - He plays out AI scenarios: power plays, “fights for truth and democracy in a world where artificial intelligence potentially undermines both”, and the geopolitics of mastering “one of the most powerful technologies to emerge in decades”. - Where the analogy breaks. Nuclear development was “contained” (at least as the film portrays it). AI is “hidden, dispersed, readily accessible”, and developed by a complex, diverse set of players. Its risks are “equally hidden, dispersed”, driven by “shadowy actors and naive developers”. - “Nuclear weapons represent a more tangible and immediate risk than artificial intelligence.” That “shouldn’t diminish the challenges” of AI safety. He names three current concerns: - generative AI subverting democratic processes; - a divisive US–China “arms race”; - long-standing worries about AI weaponisation. - Much of the risk landscape’s complexity comes from “the personalities, agendas, and power-wielding of humans in the system”. Films open “a window” onto that. - A three-hour biopic won’t make anyone an expert in responsible innovation, “and neither should it”. But he wants people and organisations to use such platforms to move toward nuanced conversations and “away from stances and statements that have all the sophistication of a bumper sticker”.

How firmly. Moderately. It is a reflective piece built around a film.

Concepts. Power dynamics as part of technology trajectories. A risk landscape that is hidden, dispersed and accessible (in contrast to contained nuclear capability). Film as a window for responsible-innovation conversation. Rejection of “bumper sticker” positions.

Analogies. Nuclear weapons, used both literally and structurally: - literally, to compare risk profiles: nuclear is more tangible and immediate, AI more diffuse and hidden; - structurally: both technologies’ trajectories are driven by human power, politics and belief, and both have geopolitical arms-race dynamics.

He stresses the differences at least as much as the parallels.

On AI. “one of the most powerful technologies to emerge in decades”. Diffuse, widely accessible, developed by many actors.

On AI risk. Catastrophic potential is acknowledged (“the potential catastrophic risks of artificial intelligence are very different”). The concrete risks he emphasises are democracy and truth, geopolitical arms races, weaponisation, and misuse by “shadowy actors” as well as “naive developers”. He ranks AI below nuclear weapons on tangibility and immediacy.

On governance. Implicit: diffuse technologies can’t be governed like a contained programme. He calls for nuanced public conversation.

On leaders. Human personalities and agendas are central to risk, but no AI leaders are named.

Criticises. Bumper-sticker stances on AI (unnamed, probably aimed at both the doom and the dismissive camps).

Quotes. - “the trajectories that powerful technologies take are deeply intertwined with very human power dynamics, politics, and personal beliefs.” - “the risks it potentially represents are equally hidden, dispersed, and driven by a complex and diverse community of shadowy actors and naive developers.” - “Nuclear weapons represent a more tangible and immediate risk than artificial intelligence.” - “away from stances and statements that have all the sophistication of a bumper sticker.”

MEDIUM#

2023-07-27 — chatgpt-and-college-applications — “ASU allows ChatGPT to be used in law school applications”#

Provenance. His own prose. He discloses that he is at ASU and was formerly at the University of Michigan (which banned ChatGPT in law applications).

Argument. - The college essay “has evolved into an art form that is less about who you are and more about where you come from”, rewarding coaching and parents’ ability to pay. It is a system that “deeply disadvantages” those without insider knowledge. - Generative AI is “a potential game changer” for equity, “if used smartly”. Used “as a collaborative partner”, it can shift power toward students who don’t know “the social etiquette of college essays”. - AI as translator. It can take “the complex, fractured and jumbled thoughts and ideas of a young person” and “translate these into prose that authentically capture the possibilities they represent”. Essays should never be tests of “unwritten rules” or “mastery of prose, rhetoric, spin, and use of English language.” - Pitfalls. “saccharine-laced boilerplate”, fabrication, and his own cynicism that templated coaching already homogenises essays. - Kickers. - The digital divide: he argues for access for all, not for keeping AI out to protect the privileged. - AI literacy: “every high school in the country should be ensuring that their students have the AI skills and savvy they need to succeed.”

Firmness. Strong advocacy, conditional on smart use.

Concepts. AI as an equaliser. AI as “translator” of thought into prose. The digital divide. AI literacy as universal secondary-school provision.

On cognition and formation. He treats AI’s articulation of a student’s half-formed thoughts as authentic self-expression, not substitution: “They just need to be themselves.” There is no worry here about outsourcing self-formation to AI, which is worth noting for later comparison.

On governance. He welcomes institutional permission with a certification-of-accuracy condition (ASU), and implicitly criticises prohibition (Michigan).

Quotes. - “the college essay has evolved into an art form that is less about who you are and more about where you come from” - “This is where tools like ChatGPT can be highly effective “translators” if used appropriately.”

2023-07-31 — ai-and-the-future-of-being-human — “Artificial intelligence themes, narratives, and the future of being human”#

Provenance. His own prose introducing an image: his AI “mind map” (version 7, downloadable PDF). The map’s contents are not in the text mirror. It grew out of a 2021 ASU article (linked, not read here).

Argument. The map is “idiosyncratic” and “rather unreliable”, reflecting his own interests. It tells a story “that spans the earliest history of humanity to the furthest reaches of the futures” people aspire to. He was especially interested in how immediate opportunities and challenges fit a larger “historic, behavioral, social, and speculative picture of what it might mean to be human in an AI-mediated future”. The larger story “transcends any single discipline or area of expertise”.

Why it matters. It frames AI in terms of deep history and being human, not as a single-technology or single-discipline issue, and it hints at his transdisciplinary stance (“develop, use, and even constrain, AI”).

Quotes. - “what it might mean to be human in an AI-mediated future” - “both transcends any single discipline or area of expertise, and spans every aspect of humanity’s journey from past to future.”

2023-08-02 — fifteen-questions-about-generativeai — “Fifteen questions every college professor should be asking about ChatGPT and other generative AI”#

Provenance. His own prose. The fifteen questions were “compiled with the help of a few colleagues”.

Argument. - This is the first semester where generative AI is “a fact of life from the get-go”. Yet “many educators remain blissfully unaware” of what it is, how students use it, and what their policies are. - The questions cover: - basic information and institutional updates; - syllabus statements; - whether non-use is acceptable (Q4) and academic freedom to set the level of use (Q5); - cheating responses and AI detectors (Q7–8); - AI-assisted grading and disclosure (Q10); - writing-dependent courses; - redesigning assignments; - integrating AI; - rethinking teaching; - AI literacy (Q15). - He deliberately gives no answers. They “really should be the responsibility of instructors and their schools and colleges to research, discuss, and agree on”. Without clear answers the semester will be “messy”, and students will “suffer as they struggle to divine what they can and cannot do”.

Concepts. Institutional responsibility and preparedness. The legitimacy of opting out (Q4) sits alongside AI literacy (Q15).

On governance. A distributed, institution-level model. Responsibility lies with instructors and their organisations, not with a single authority. The emphasis is on preparing now, while there is “still a window of opportunity”.

Quote. - “these really should be the responsibility of instructors and their schools and colleges to research, discuss, and agree on.”

2023-08-08 — thinking-differently-about-ai-and-risk — “Thinking differently about AI and risk”#

Provenance. A short piece of his own prose reposting a 20-minute 2021 video talk to a National Academies-sponsored symposium series. The slide-deck filename reads “AI-Human-Condition-Maynard-7-29-21”. The talk’s content is not in the text and was not viewed.

Argument. He is “surprised at how prescient it was”. The talk’s core is “how to frame novel risks associated with AI, and how the risk innovation framework we have developed here at ASU could be applied to good effect”. Although it predates widespread LLM use, the ideas are “perhaps more relevant than ever in an age of generative AI”. He points to the Risk Innovation Nexus website.

Why it matters. It is an explicit 2023 re-endorsement of risk innovation, his ASU framework (developed in earlier batches), as the right lens for generative AI risk. It shows continuity from pre-ChatGPT risk thinking into the LLM era. The linked slide title also ties AI risk to the human condition.

Quote. - “At the heart of the talk is a discussion of how to frame novel risks associated with AI, and how the risk innovation framework we have developed here at ASU could be applied to good effect.”

2023-08-18 — being-human-in-an-augmented-future — “Being Human in an Augmented Future and the movie Ghost in the Shell. The Moviegoer’s Guide to the Future Episode 7”#

Provenance. His own introduction to his narration of chapter 7 of Films from the Future (written late 2017 to early 2018), plus the standard series boilerplate (see the series note below). The chapter audio is not in the mirror. Listed chapter sections: body hacking, “More that “Human”?”, “Plugged In, Hacked Out”, “Your Corporate Body”.

Argument. - The film “wraps deep reflections of the meaning of personhood” in an anime action movie. He watches it with students at least once a year. - The chapter shows “the seeds” of his thinking about personhood and being human “when technology is increasingly a part of who we are”. He wrote it four years before launching ASU’s Future of Being Human initiative. - New 2023 claim. It is relevant to AI and AGI. “While we’re still a long way from machines that have consciousness and self-awareness”, we must think now about how they might be treated. That treatment “reflects and defines our own humanity”. Above all we must avoid “dehumanizing them” by denying they are or could be “human” in order to justify controlling and using them. The film helps move “from very conventional ideas of humanity to more sophisticated concepts of personhood”.

Concepts. Personhood beyond the human. Dehumanisation of AI as a moral hazard. Treatment of AI as a mirror of our humanity.

Change of view. See the consciousness post: his “a long way” here becomes “compelling” near-term arguments five days later.

Quotes. - “how we avoid dehumanizing them by claiming that they are not and never will be “human” to justify how we control and use them.” - “While we’re still a long way from machines that have consciousness and self-awareness”

2023-08-25 — ai-platos-cave — “AI and the Art of Manipulation and the movie Ex Machina. The Moviegoer’s Guide to the Future Episode 8”#

Provenance. His own introduction to his narration of chapter 8 (the chapter itself was republished as text on 2023-04-16; see B08), plus series boilerplate. The epigraph is a line from the film.

Argument. - “In some ways everything about artificial intelligence has changed over the past 12 months. In others, little has changed.” The chapter predates LLMs, yet “the underlying ideas remain profoundly important — probably more so now”. - The film was prescient on three things: “the right to experiment and develop AI without permission” (permissionless innovation), “the nature of intelligence”, and “the risks of us being manipulated by the machines we create”. - An updated reading of Plato’s Cave. We are getting “ever-closer to developing machines that have awareness and agency”, machines not bound by “our biologically grounded and very human ideas of right and wrong”. These become “the “enlightened”” who can choose to bring humans out of the cave “or leave us there, to be manipulated, used, and enslaved.” - A wry complaint that ideas buried in “a book that most dismiss as simply being about movies” struggle to be seen.

Concepts. Permissionless innovation. Artificial manipulation. Plato’s Cave. Machine awareness and agency. Machines outside human moral frameworks.

Note the symmetry. Two days earlier he warned against humans enslaving conscious AIs. Here AIs might leave humans “enslaved”. He holds both directions of domination in view.

Change of view. He re-endorses the 2018 ideas. His 2023 wording (“ever-closer”, “awareness and agency”) goes further than the 2018 chapter’s scepticism about machine minds.

Quotes. - “In some ways everything about artificial intelligence has changed over the past 12 months. In others, little has changed.” - “as we get ever-closer to developing machines that have awareness and agency, and that are not bound by our biologically grounded and very human ideas of right and wrong”

2023-08-28 — dear-mr-musk-can-i-have-my-book-back — “Dear Mr. Musk, Can I Have My Book Back Please?”#

Provenance. His own prose. It includes the August 2021 cover letter written by his Moviegoer’s Guide to the Future class and signed by 54 students (class-authored, not his alone). One student quote: “have they not seen any movies?”

Argument (anecdote). - After Tesla announced the Tesla Bot in August 2021 (a “friendly” humanoid robot to “eliminate dangerous, repetitive, and boring tasks”), his class was discussing “the fine line between what entrepreneurs can do, as opposed to what they should do.” Student views ranged from “amazing” to alarm. Most felt “curiosity and excitement” threaded with concern that unless the team took a ““how to innovate responsibly” crash course, this could end very badly.” - The class sent 27 copies of Films from the Future to the Tesla Bot team, with a supportive letter hoping for “a little less I Robot, and a little more Bicentennial Man”. Maynard hid his own signed copy of Iain M. Banks’ Look to Windward in the box, addressed to Musk (a Banks fan). There was never any reply. - He is “invariably disappointed” when students’ efforts land “on deaf ears”. He asks the team to read the books and to “work a little harder to create a robot that exemplifies what they should do, and not just what they can do.”

Why it matters. It shows his stance toward tech leaders: supportive of ambitious innovation, but insisting on “should” as well as “can”. It also shows his view of responsible innovation as something engineers can learn (via science fiction and his book). There is a mild, humorous critique of corporate non-responsiveness to publics (students). He treats the class as a space for developing publics’ views of responsible innovation.

Analogies. Science fiction (I, Robot, Bicentennial Man) as cautionary and aspirational templates, used conceptually.

Quotes. - “the fine line between what entrepreneurs can do, as opposed to what they should do.” - “work a little harder to create a robot that exemplifies what they should do, and not just what they can do.”

2023-08-29 — chatgpt-enterprise-game-changer — “ChatGPT Enterprise could be a game changer for universities”#

Provenance. His own prose.

Argument. - Equity drives the access question. ChatGPT Plus costs $20 a month, which disadvantages students “on the margins”. Universities have been blocked from providing access by security and data-privacy issues. - Enterprise solves this. In ChatGPT Enterprise, data is not used for training and remains the institution’s property. He reports “anecdotes” of researchers “naively” uploading human-subjects data, paper drafts and grant proposals, “oblivious to the reality that they are giving away protected data and their intellectual property to OpenAI”. He notes Michigan’s in-house U-M GPT, but says ChatGPT “still has the edge”. - A forcing function. Universal access would make AI use “a default option in learning” and a “forcing function in how we teach”. It would force instructors “to re-evaluate how they teach, and even the purpose and nature of education and learning”, which he does not think “such a bad thing”. “I’m a huge proponent of this”. - Benefits. Real-world skills employers demand. Transformative for students with “jagged profiles” who don’t fit “the standard model of an average learner” (linking Punya Mishra). Cost is acknowledged (“Whether this is justified by the opportunities it opens up remains to be seen”).

Firmness. Strongly positive.

On companies. Treats OpenAI’s product as the solution to institutional barriers. The earlier data risk is framed as users’ naivety, not as a question of company conduct. My observation (not his): the post raises no concern about vendor dependence or the power of a single company in education.

On governance. Institutional adoption as an equity and privacy measure.

Quotes. - “universal access to ChatGPT within universities is likely to become a forcing function in how we teach.” - “oblivious to the reality that they are giving away protected data and their intellectual property to OpenAI.”

Series note: The Moviegoer’s Guide to the Future boilerplate (all six podcast episodes in this batch)#

Each episode post ends with the same passage in his own prose, “About Films from the Future”. It is his clearest recurring statement of method: - He began the book in 2017 to explore “the deeply complex landscape around emerging technologies, the future, and socially responsible innovation”. - Most books on technology and the future “take a polarized stance”: doom, or technology saving the world. These sell but do not help in a landscape with “few right and wrong answers”. - One must weave together insights from many fields, “including the arts and humanities”. “dialogue and discussion are far more important than preaching.” - The book is “not a book about science fiction movies, or about specific technologies, but about how all of us can think differently about our roles in ensuring the future we’re building is better than the past we leave behind.”

Quote. “where dialogue and discussion are far more important than preaching.”

LOW / NONE#

• 2023-07-28 — minority-report-predicting-criminal (LOW): Podcast episode 4 (chapter 4 of Films from the Future). Short intro on predictive policing and “the slippery slope of trying to predict criminal behavior”. Listed chapter sections include machine-learning-based precognition and “Big Brother, Meet Big Data”. Plus series boilerplate. Not a Modem Futura post.
• 2023-07-29 — the-unimagined-preposterousness-of (NONE): A 33-word cross-post stub. The original redirects to Mark Daley’s Substack (Noetic Engines), where Daley discusses whether LLMs are Dennett’s “counterfeit people” or could be conscious. No Maynard prose. It shows only that he chose to share Daley’s piece three weeks before his own consciousness post.
• 2023-08-04 — the-moviegoers-guide-to-the-future-episode-5 (LOW): Podcast episode 5 (Limitless, pharmaceutical cognitive enhancement). The intro notes that social norms around “chemicals-based augmentation” differ from those around physical augmentation. It adds that “an obsession with improving “intelligence” often reflects our biases and preconceptions around what intelligence actually is”, which he says matters for AI too. Plus boilerplate.
• 2023-08-11 — social-inequity-elysium (LOW): Podcast episode 6 (Elysium). The intro frames technology as able to exacerbate wealth inequity “if we’re not careful”. A cut draft section attacked “the sheer immorality around how we’ve monetized health and wellbeing” in US healthcare. The chapter’s workplace-safety narrative reflects “having worked in occupational health for over a decade in my early career”. Bioprinting raises the question of who gets access. This is a biographical marker of his occupational-health roots in risk.
• 2023-08-16 — riding-the-ai-tiger (LOW; guest post by Brad Allenby, not evidence of Maynard’s views):
• Allenby, his ASU colleague, argues that AI regulation assumes a simple system when AI is a complex adaptive system. He likens current efforts to building “a Gosplan for AI” and uses asbestos as the simple-system example.
• He names three problems: cycle time, knowledge, and the scope of effective regulation.
• His remedies: perception-only networked working groups, agile adaptive capability (FDA/EMA as partial models), soft law, and skunk works.
• Maynard wrote none of it. Hosting it shows sympathy for complexity-based, agile governance framings, which his own earlier posts also use.
• 2023-09-01 — welcome-to-the-singularity (LOW): Podcast episode 9 (Transcendence; sections include technological convergence, neo-Luddites, techno-terrorism, “Exponential Extrapolation”). The intro dismisses the film’s nanobots as fantasy: “No future exists within this universe where nanotechnology, for instance, enables “nanobots” to recreate everything”. This is consistent with his nanotech scepticism. He adds that “the more powerful our technologies are, the more important it is for us to collectively take their potential impacts on society seriously”, and that the consequences we expect from powerful technologies “rarely are” what happens. Plus boilerplate.