S3: testimony, institutions and the 2020science record (2006–2017, with a 2026 retrospective)#

Supplementary reading for 05-maynard-risk-and-ai-map.md. Share S3-testimony-institutions-2020science. Prepared 26 September 2026. This file covers Andrew Maynard’s own thinking only and makes no comparison with any other material. The map was read but not edited.

Scope and conventions#

Items read in full (all in Resources/maynard-papers/web/). Short names are used in the cross-cutting sections at the end.

Page numbers. - T2006: the file holds the printed hearing record. Record page = PDF page + 48 (record pp.50–64 = PDF pp.2–16). Cited as “rec. p.X”. - T2007 and T2008: cited by PDF page. The printed page number is PDF page − 1. - Handbook10: cited by printed page (pp.573–586 = PDF pp.3–16). - CETI10: cited by PDF page (printed Global Redesign Initiative pp.301–307 = PDF pp.1–7). - Rethinking17: cited by printed page (pp.193–201 = PDF pp.4–12). - Define11b: PDF pages. GI08a is one page; GI08b two. - Markdown files have no pages; quotations are located by section heading where useful.

Weighting. No item in this share is AI-written. Single-authored items are treated as his thinking. Co-written items (Handbook10, Seven11, Weighing09, Guardian17, CETI10, and the co-authored quotation in LateLessons08) are shared positions and weighted lower, although he chose to repost or summarise several of them under his own name, which is some evidence of endorsement. Worrying11’s recommendations are a workshop consensus; only his framing counts. Prehistory26 is his own but retrospective: it is good evidence of how he now reads his past and of who drafted what, weaker evidence of what he thought at the time.

Overlap with the corpus. None of these items is in the Substack corpus. Brain14 is a blog post about the 2014 Nature Nanotechnology commentary whose text was reposted in the corpus on 2023-12-03 (3d-artificial-brains-and-ai). The blog post contains lines on the singularity, machine rights and AI’s effect on human behaviour that the commentary text does not.

How the map is referenced. “C” numbers are the map’s §4 commitments, “§5.x” its concept tables, “tension n” its §8 tensions, and “T” its §6 threads.

A. Congressional testimony and the research-strategy years (2006–2008)#

A1. T2006: House Science Committee, 21 September 2006#

Provenance. His own oral statement and prepared written statement as Chief Science Advisor, Project on Emerging Nanotechnologies (PEN). He says his comments are “my own personal opinions” (rec. p.50). The biography (rec. pp.63–64) is standard hearing material.

The argument. Federal nanotechnology environment, health and safety (EHS) research is incoherent. It “only coincidentally give[s] the fleeting illusion of coherence” (rec. p.50), like a mansion built by twenty master builders with no architect. Nanotechnology is “far too complex for disjointed, bottom-up research agendas” (rec. p.50), so a top-down strategy is needed, with teeth, funding and mechanisms. He makes three recommendations: a top-down strategic risk-research framework; at least $100 million of targeted research over two years; and a joint government–industry research body modelled on the Health Effects Institute (HEI). His own analysis finds about $11 million a year of highly relevant research against the NNI’s claimed $38.5 million (rec. pp.51, 56). The portfolio is out of step with market reality: research concentrates on carbon materials and on lungs, while silver and metal oxides dominate products and ingestion has no projects. He asks: “Is the emphasis on lung impacts due to careful consideration of relative risks, or because pulmonary toxicologists are more active in this field?” (rec. p.55). “A list is not a research strategy” (rec. p.51).

Key concepts. - Strategy over lists. Research that is not tied to decision needs produces the appearance of progress. - The “information trap”: knowledge is “sufficient to cast doubt on the safety of some nano-industries and products” but lacks the credibility for industry to act (rec. pp.59–60). - Independent cooperative science. Four attributes: independence, transparency, review, communication (rec. p.60). Research contracts rather than grants, dissenting critiques published with reports (rec. p.60). - Who bears the risk. “it is ultimately the public—as workers or consumers, for instance—that may bear many of the potential risks” (rec. p.53), which is why project-level data should be public. - Risk of rejection. “If investors and consumers reject nanotechnology through fear and uncertainty, missed opportunities in areas like medical treatment and energy production could deal a severe blow to the quality of life” (rec. p.52). - Designing risk out. He asks how much is being spent “to design and engineer risks out of nanotechnology processes and products (rather than just addressing them after the fact)” (rec. p.63). - Foresight and convergence. “I see no evidence of foresight”, including of risks as nanotechnology “converges with biotechnology” (rec. p.53).

What this adds to the map. - Primary evidence for Phase 0. The map’s Formation phase (“to 2014”) rests on later accounts. This is the formative Maynard in his own words: a risk scientist who wants uncertainty reduced “through systematic scientific research” (rec. p.52), who counts dollars and projects, and who wants strategy, metrics and “hard priorities” (rec. p.57). - Earlier origins. Symmetric risk (the map dates it from 2015-01-30 and 2016-03-02) is already here as the risk of fear-driven rejection and missed benefits (rec. p.52). Justice as “who bears the harm” (C11; FFTF) is here as workers and consumers bearing the risk of publicly funded technology (rec. p.53). The disciplinary-interest question about lung research is an early form of the logic behind orphan risks: which risks get studied depends on who is doing the studying. - Mechanisms, not just principles. The map notes that his governance thinking is “thin on mechanisms” (C5; tension 9; gaps). In 2006 he was specific: dollar figures, time frames, a named institutional model, governance safeguards for independence. - Bearing on Andrew’s notes. Strong evidence for note 1: quantitative, science-based risk assessment is his professional foundation. Even here, numbers are subordinate to purpose: “numbers alone can be misleading: What is important is the research that those numbers represent” (rec. p.53).

A2. T2007: House Science and Technology Committee, 31 October 2007#

Provenance. His own 36-page written testimony, with an explicit statement that the opinions are his own (PDF p.3).

The argument. Federal action on nano-EHS research has been “slow, badly conceptualized, poorly directed, uncoordinated and underfunded” (PDF p.19). Five things are needed: acknowledging unconventional behaviour, leadership, a strategic framework, mechanisms and resources. Six recommendations follow (PDF pp.4–5): a strategy updated every two years; mechanisms and a federal advisory committee for stakeholder input; 10% of nano R&D for EHS research (at least $50 million targeted, the rest exploratory); a public–private partnership; a funded programme (about $1 million a year) of public engagement ensuring “two-way communication between the developers and users of these technologies” (PDF p.4); and a single accountable leader. “We cannot afford to drive blind into the nanotechnology future” (PDF p.5).

Key concepts. - Unconventional behaviour. “Assuming that new technologies will have conventional, predictable and manageable risks is a recipe for disaster” (PDF p.7). Twentieth-century approaches may not be “up to the task” (PDF p.6); physical form is “making a mockery of our chemicals-based view of risks and regulation” (PDF p.6). Hazard is always “tempered by the likelihood of exposure” (PDF p.8). - “science in the service of safety, and not science for its own sake” (PDF p.9; again p.33). - Known unknowns and unknown unknowns. Targeted research answers pressing questions; exploratory research is needed “to identify the questions we haven’t thought of yet” (PDF p.14). “Many aspects of nanotechnology are so new that we do not yet know what are the right questions to ask regarding potential risks” and “it is difficult to set milestones on discovering the unknown” (PDF p.33). - Method is not strategy. “we must not mistake methodology for strategy” (PDF p.21). A prioritisation process without vision is “a bureaucratic reaction to criticism” (PDF p.21). - Research that falls between the cracks. “important research is not being funded because it falls between the cracks, because it doesn’t fit within a particular agency’s mandate, or because adequate funding mechanisms do not exist” (PDF p.22). Portfolios are driven by “personal research interests, rather than overarching needs” (PDF p.27). - Promotion versus oversight. The NNI is “more attuned to stimulating exploratory science and developing technology applications than providing science in support of oversight” (PDF p.30). Resources flow to rich science agencies, while “it is the agencies without the resources to do the right research that have the clearest perspective on what needs to be done” (PDF p.29). - The pacing problem. “new technologies will always be one step ahead of our understanding of how they might cause harm” (PDF p.33), so reactive oversight is hard to justify. - Leadership without command. The model should be “not one of command and control, but of leadership, coordination and facilitation” (PDF p.28). - No excuses. “good intentions are not enough” (PDF p.16); “to claim ‘it’s difficult’ is a poor excuse for inaction” (PDF p.20); “committees and networks in and of themselves do not constitute leadership” (PDF p.8). - A concrete early warning. Carbon nanotubes sold by Cheap Tubes Inc. with a safety data sheet describing them as graphite: “how long will it take before someone is able to tell me how to open the package, extract the material, and use it—safely?” (PDF p.26).

What this adds to the map. - New concepts: “science in the service of safety”; “mistake methodology for strategy”; exploratory research for “the questions we haven’t thought of yet”; research that “falls between the cracks”. - Earlier origins. “good intentions are not enough” (C10, anchored in FFTF) appears in 2007, aimed at government. The map’s “It’s complicated” is not an excuse (C5) has a 2007 twin in “it’s difficult”. The pacing gap (§5.3, from 2016) is stated in 2007. Research that falls between institutional cracks, and portfolios shaped by researchers’ interests, are the logic of orphan risks (§5.1, named 2018) applied to research funding a decade earlier. His structural account of institutions (map: in his own prose from 2022) is here in the promotion-versus-oversight conflict of the NNI. - Qualification. The map places him in favour of adaptive, distributed, soft-law governance (C12). In the research-strategy domain in 2006–08 he argued for strong, central, authoritative coordination: one leader, “a NEHI group with teeth” (PDF p.24), dedicated budget shares and measurable goals, while insisting it should not be command and control. These are compatible, but the map’s picture is incomplete: he has long wanted strong central capacity for risk research, and flexibility in regulation. - Qualification (engagement). The map’s tension 9 says engagement is “thin as mechanism”. In 2007 he proposed a federal advisory committee for “transparent input and review” (PDF p.25) and a funded engagement programme with four named aims (PDF pp.4–5). Mechanisms were concrete in his formative years; they thinned later. - Bearing on Andrew’s notes. Both notes are supported. Note 1: the testimony is built on quantitative risk science and budget accounting. Note 2: in the same document he says we “do not yet know what are the right questions to ask”, and that methodology must not be mistaken for strategy. Humility about ignorance was built into his quantitative period, not added when AI arrived.

A3. T2008: House Science and Technology Committee, 16 April 2008 (NNI Amendments Act)#

Provenance. His own written testimony.

The argument. Reauthorisation must address risk: a top-down risk-research strategy “informed by stakeholders from industry, academia and citizen communities” with “measurable goals” (PDF p.2); at least 10% of nano R&D for EHS; a coordinator with authority; full transparency of research investment; and public–private partnerships. The executive summary opens with broader questions: “How can we learn to use such a powerful technology wisely?”, “Who will decide how it is used, and who will pay the cost?”, and how will “the supposed beneficiaries” be engaged (PDF p.2). The NNI’s $68 million claim for 2006 is really $13 million of highly relevant research (PDF p.12): “the hubris surrounding nanotechnology research and development (R&D) funding is giving way to a sobering reality” (PDF p.3).

Key concepts. - Innovation and harm-avoidance are complementary. “The two aims of stimulating innovation and avoiding harm need not be, nor should be, mutually exclusive” (PDF p.5). - Rules of safe use ahead of the game. “Everything has the potential to cause harm. If we are smart, we learn how to avoid harm. And if we are very smart, we work out the rules of safe use ahead of the game” (PDF p.7). “Ignoring the signs of adverse consequences will only result in poor decision-making” (PDF p.7). - Counter-intuitive hazards. The skillet and the knife: same chemistry, different shape, different rules of safe use. “Because we cannot see these intricate nano-shapes unaided, we forget that they are important” (PDF p.7). - Promoters and markets are not enough. “Neither will safe nanotechnologies emerge if the promoters of the technology are calling all the shots”, nor “through wishful thinking and ‘spin’” (PDF p.8). “Accepted mechanisms of technology development and transfer—including investigator-driven research, generation of intellectual property, knowledge diffusion and market-driven commercialization—will not ensure” safety “on their own” (PDF p.11). - Transparency as a precondition of trust. Any assessment “not backed up by publicly accessible project-specific data is worthless” (PDF p.12). - Warning signs. “we are skating on thin ice, and are in danger of missing the warning signs” (PDF p.14). - Public preparedness. Americans are “woefully unprepared for the nano-age”; too little has gone into “educating and engaging the public” (PDF p.15). He endorses nano education partnerships for schools and undergraduates (PDF p.15).

What this adds to the map. - Earlier origins. “Who decides … and who pays” (C5, C11; FFTF) is his framing in 2008. His critique of hubris (§5.2, FFTF 2018) is applied in 2008 to R&D promotion. The claim that markets and promoters cannot secure safety alone predates the map’s structural thread (2022) by fourteen years. - Qualification (engagement and literacy). His 2008 language mixes two-way engagement with awareness and education. The map tracks AI literacy as a remedy from 2023 that loses his confidence in 2025. Its roots are here: in 2008 public ignorance was “a significant failing” of a government programme (PDF p.15). This helps explain the later trajectory: literacy has been one of his instruments for nearly twenty years. - Bearing on Andrew’s notes. Quantitative accounting is used to puncture comfort: a $68 million figure that “has little credibility” (PDF p.12). Numbers are tools of accountability, not solace.

A4. Funding08: “U.S. nanotechnology risk research funding—separating fact from fiction” (2020science, 18 April 2008)#

Provenance. His own blog post following T2008.

The argument. He explains PEN’s four-level relevance classification of the NNI’s 246 projects and why $68 million becomes $13 million. Numbers can serve two purposes: to “justify past performance, or … inform future actions” (the NNI chose the former).

Key concepts. He writes: “I’m not a great fan of bean-counting. Evaluating research in terms of dollars invested (or Pounds or Euros) is a crude tool at the best of times. But when it comes to assessing investments and returns, the fact is that bottom-line figures count.” Using past spending “as a feel-good exercise is a disaster when it comes to future planning—because the assessment is invariably based on wishful thinking rather than reality.” His analogy: going to the doctor with a headache and being told about millions spent on cancer drugs, “when all you wanted was an aspirin!”

What this adds to the map. The clearest single statement in this share of his double attitude to numbers. They are “a crude tool”, yet they “count”; they can inform or they can comfort. This is direct evidence for both of Andrew’s notes: quantitative accounting is part of his foundation, and he distrusts numbers that serve as “a feel-good figure”.

A5. Bulletin08: “Setting the nanotech research agenda” (Bulletin of the Atomic Scientists, 14 January 2008)#

Provenance. His own column.

The argument. Reauthorisation of the 2003 Act must turn attention from creating nanotechnology to making it safe and sustainable. Early research raises “warning flags”. Beyond harm lie social, ethical and legal questions that “are beyond the scope of scientific research”. The US needs a funded safety-research strategy, public engagement, and foresight.

Key concepts. - “every technology has consequences–and the more innovative the technology, the more uncertain the consequences.” - “Look before you leap, and engage interested parties as early as possible.” - “It is one thing to say that we can change the world with nanotechnology, but it is something else to ask ‘should we?’” - “Who is reaping the benefits of new nanotech applications, and who is paying the price?” - The Centers for Nanotechnology in Society show that engagement is “as much about helping scientists consider a broader set of values in their work, as it is about enabling citizens to make informed decisions”. - The old philosophy, “take care of the benefits, and the risks will take care of themselves”, is “a largely false dichotomy between risks and benefits”.

What this adds to the map. Earlier origins for could versus should (§5.2, FFTF pp.36–39), justice as who benefits and who pays (C11), and the engage-early lesson that the map dates to 2023-05-17 (“the one big lesson”). His view that engagement changes scientists as well as publics, the core of his later anti-deficit position, is present in 2008, alongside the awareness language noted under A3.

B. Institutions: the WEF proposals and his 2026 account of them#

B1. GI08a and GI08b: “Global Institute on (for) Emerging Technology Policy” (WEF “breakthrough idea”, 9 and 12 December 2008)#

Provenance. His own drafts, headed “Breakthrough Idea Maynard Draft”. The 9 December version is bullet points; the 12 December version is prose.

The argument. Emerging technologies are enabling, cross-boundary and “do not conform to conventional paradigms for technology innovation and implementation” (GI08a). They need “New and innovative policies … at the international, national, corporate and institutional level” (GI08a). The proposed institute would support “long-term science-based decision-making that will ensure the greatest social and economic benefits of emerging technologies, while minimizing adverse impacts” (GI08a). It would be “Science-based”, “Non-advocacy” and “Non-partisan” (GI08a), jointly funded by government and industry but operating “independently of the funders” (GI08b p.2). Its three challenges are working across boundaries, technology transfer, and “predicting, assessing and avoiding adverse consequences” (GI08b p.1). Engagement is needed to secure “buy-in from an increasingly vocal and powerful global society” (GI08b p.2).

Key concepts. Innovative technologies “require innovative policies in order to grow and mature” (GI08b p.1); independence from funders; real and perceived risks as barriers (GI08b p.2).

What this adds to the map. - Primary source. The map knows these proposals only from later accounts (§7 Phase 0; T5). The documents themselves do not use the term “early warning”. They speak of prediction, assessment, avoidance and (in CETI10) horizon scanning. The “early-warning institution” label comes from the index and later accounts. - Earlier origin of risk innovation. The founding claim of C3, that conventional tools are not enough and risk thinking must itself innovate, is here in 2008 as a claim about policy: new technologies need new policies. The map’s inferred shift “2015→2016: citing responsible innovation as an outside framework → proposing his own risk-innovation framing” should be softened. The impulse to innovate in how risk is governed is his from at least 2008–09 (see also TenThings09); only the name comes in 2015.

B2. CETI10: “A New Global Centre for Emerging Technology Intelligence” (WEF, 2010)#

Provenance. Published as a Global Agenda Council product under the chair, Chris Murray, with a note that views do not reflect all members (PDF p.1). In Prehistory26 he says “the conceptual work and drafting was done by Tim and myself with relatively little direct input from the rest of the council.” Treated as co-drafted with Tim Harper.

The argument. Decision-makers “are foundering in a world dominated by rapid and unprecedented social and technological developments” (PDF p.1). The GM-food experience shows how easily emerging technologies are mishandled. Concerns “were grounded in a backlash against corporate control that cut consumers out of the decision-making process”, and people said no “not because of the science and technology, but because of the way they were handled” (PDF p.1). Hence: “hierarchical, evidence-based decision-making is not sufficient on its own to ensure the success of new technologies” (PDF p.2). A new paradigm must predict and avoid hurdles, work with multiple stakeholders, address health and environmental impacts “before they occur” and respond rapidly (PDF p.2). The Centre would be “neutral, transparent and authoritative” (PDF p.1), “immune to any political or business interests” (PDF p.5), and would do horizon scanning, publish assessments and convene an annual retreat. Debate is polarised between promoters and defenders of the status quo, with technologies presented as “either having the ability to usher in a techno utopia or the potential to destroy the world” (PDF p.3). The Centre would help decisions “be based on science fact not science fiction” (PDF p.3). Its first focus: nanotechnologies and synthetic biology (PDF p.5).

What this adds to the map. - Earlier origin of backlash as a risk. The map dates this to 2024-02-18. Here, in 2010, GM foods are the lesson that exclusion produces rejection that “severely retarded the implementation of a technology that could save and improve millions of lives” (PDF p.1). - Bearing on note 1. “Evidence-based decision-making is not sufficient on its own” states his position exactly: the evidence base stays, and more is added. - Refusing the utopia–apocalypse binary (§7 constant 7) is here in 2010. - A qualification about fiction. In 2010 “science fiction” is a pejorative for poorly informed opinion. By 2012 (Speculated12) and 2014 (Brain14) he was using speculative design and fiction deliberately, and by 2018 films were his main lens. The map treats stories as tools as constant from 2018. The record shows a turn between 2010 and 2012. - No AI. The 2010 focus is nanotechnology and synthetic biology. This fits the map’s claim that AI arrived late.

B3. WEF10: “Emerging technologies at the World Economic Forum – rethinking integrative approaches to global risks” (2020science, 30 November 2010)#

Provenance. His own post, written while he chaired the Council on Emerging Technologies; “these are my own views”.

The argument. A high-level debate on resource scarcity produced a “technology count” of zero. Either technology is seen as too complex, or there is “a naive assumption” that scientists will pull a rabbit from the hat. Emerging technologies play three roles in relation to global risk: tools for monitoring, potential solutions, and “agents of change which may lead to a dramatically altered risk-landscape”. Without integration, “when we most need a technology ‘white rabbit’, the hat will be empty!”

What this adds to the map. An early statement of technologies as both risk-reducers and risk-creators within one landscape (symmetric risk, C6), and of complexity as the reason integration matters: “in systems where associations between cause and effect are complex, you ignore synergistic inter-relationships between factors at your peril.” The complexity premise (C8) predates the 2015 and 2018 sources the map cites.

B4. Building11: “Building a sustainable future” (2020science, 19 January 2011)#

Provenance. His own post about a WEF white paper co-written with Tim Harper.

The argument. “‘Technology doesn’t just happen’”: people wrongly assume “bolt-on answers to pressing problems as and when they are needed”. Three questions drove him: how to have solutions available when needed; how to invest proactively; and how to avoid new risks while using technology to reduce old ones.

Key concepts. “I am constantly surprised at the blind faith many people have in science and technology”. Start “with the problem, not the solution”. Learn from failure: “it is where we have failed to cure a disease, or to relieve poverty and hunger, or to increase someone’s quality of life, that we have the most to learn.”

What this adds to the map. An early critique of techno-solutionism, paired with the conviction that technology is needed. The map has “technological foreshortening” (2023) and the “fix” frame (2024) but little on solutionism before FFTF. “Technology doesn’t just happen” is an early statement of non-determinism (map: “Technology is not deterministic”, 2025-03-30), which supports the map’s reading of tension 5: inevitability applies to the overall trajectory, choice to the path.

B5. Prehistory26: “Before the Fourth Industrial Revolution: Notes on an Institutional Prehistory” (andrewmaynard.net, 8 April 2026)#

Provenance. His own retrospective essay. The file carries no AI-authorship note and the index gives none. It is used as evidence of his 2026 self-understanding and of drafting credit.

The argument. The Fourth Industrial Revolution rests on eight years of forgotten WEF council work (2008–2015) on governing emerging technologies. He traces the 2008 one-pager, the 2010 CETI proposal (drafted with Harper), the 2011 white paper, the Top Ten Emerging Technologies list (an idea of Javier Garcia-Martinez’s; first list 2012), his December 2015 Nature Nanotechnology piece on “Navigating the fourth industrial revolution”, and the Global Future Council on Agile Governance (from 2016; report 2018). The council kept seeing a widening “governance gap between what new technologies could do and how societies were prepared to handle them” (section “2008”). The 2008 bottom line “could appear unchanged in almost any serious AI governance document being written today.” The persistent question is “how do you build governance capacity for technologies whose pace and scope outrun the institutions responsible for them?” (section “2016 and after”).

Key concepts. - Policy versus intelligence. “‘Policy’ implied an organization that would prescribe. In contrast, ‘intelligence’ implied an organization that would scan, analyze, and inform.” The change “narrowed the original ambition in ways I still have mixed feelings about.” - Lessons for AI governance: “Institutional experiments that fail are still formative”; “The ideas keep working after you stop pushing them”; bylines misattribute intellectual labour; “conditions for possibility” matter more than who plants the flag. - Continuity. “The language had changed; the underlying diagnosis had not” (on the 2018 agile-governance report and CETI).

What this adds to the map. - It is his own statement that his 2008 governance questions are today’s AI governance questions. That is direct support for the map’s claim that he reasons from past technologies by process and institution (C13) rather than by hazard analogy. - “Governance gap”, a 2026 term in the map (universities), is here the council’s 2008 diagnosis. - A qualification to the honest-broker reading (tension 12). His “mixed feelings” about moving from “policy” to “intelligence” suggest that he wanted institutions able to prescribe, not only to inform. The honest-broker role was never his whole view of institutions. - Drafting credit: CETI10 can be treated as substantially his and Harper’s.

C. Regulating nanomaterials (2009–2011)#

C1. Weighing09: “Nanotechnology: Weighing the risks of regulation” (2020science, 8 July 2009)#

Provenance. An early draft of the Nature commentary “Too small to overlook”, written jointly with David Rejeski. The introduction and closing note are his. Shared position.

The argument. Voluntary reporting schemes for nanomaterials in the UK and the US failed: 13 UK submissions in two years, and an EPA estimate that about 90% of materials likely in commerce were not reported. Regulators are “grappling with ensuring the safety of unknown quantities of unknown materials, being used in unknown ways.” Canada and France moving to mandatory reporting is welcome, since, “Given the reticence of industry to volunteer information”, it “will enable regulators to make decisions based on reality rather than speculation.” Firms themselves report a “lack of sufficient data to quantify risks” (quoting Lindberg and Quinn), and insurers rank nanotechnology as a top emerging risk.

What this adds to the map. Direct, early evidence of his view that voluntary industry disclosure does not deliver the information that oversight needs. The map’s claim that “self-governance fails” (T5) rests mainly on 2023–26 AI sources and a co-written 2019 chapter; here is a 2009 empirical case. His closing note (“a maturity of thought that is lacking in the draft above”) is an early instance of public self-correction. The photograph caption “Me handling multi-walled carbon nanotubes some years ago” is an early touch of self-implication.

C2. Handbook10: “Conclusions: triggers, gaps, risks and trust” (Maynard, Bowman and Hodge, 2010)#

Provenance. Co-authored concluding chapter of the International Handbook on Regulating Nanotechnologies, which the three co-edited; he is first author. He posted it himself and reproduced most of it in Seven11 under his name. Shared position, strongly endorsed.

The argument. Nano-regulation is a “wicked” problem, “the nanotech equivalent of the Tower of Babel” (p.573). “Effective regulation cannot afford to be based on imagined futures that are not connected to scientific, social and economic reality” (p.573), yet “misplaced speculative regulation” can cause harm in both directions, “ranging from unnecessary barriers to economic growth to inadvertent harm to the public and the environment” (p.574). The risk discussion must be “decoupled from speculative visions of future technologies and informed by plausible emerging risks”, and “grounded in established approaches to identifying, assessing and managing risks” (p.575). Seven challenges follow: the language game; filling science gaps (with open-ended research to find new gaps); standards and metrology; regulatory gaps (regulators shoehorn “new challenges into old regulatory frameworks”, p.577, but should not shift “with every technological whim”); balancing innovation and safety; moving forward with caution; and transparency and trust.

Key concepts. - The limits of quantitative risk assessment. “Over the past few decades, regulation of materials and products has typically been built on quantitative risk assessment – the purview of invisible experts – and quietly modulated by political and economic interests. The result has been a science-based regulatory approach that, while both professional and competent, nonetheless has tended to deal retrospectively with well-established risks” (p.582). - Two dangers, one balance. “a blinkered adherence to a science-driven and hierarchical decision-making approach will ignore these values”, and “perceptions that are counter to current scientific understanding may have an undue influence on regulatory decisions” (p.583). “It is a question of achieving a new balance” (p.583). - Citizens and experts. “the challenge will be how to empower people to be an effective part of the decision-making process, rather than how to make decisions on their behalf. At the same time, the details of how regulations are crafted and enacted will of necessity remain the responsibility of a small number of experts” (p.583). - Structural conflicts. Governments and industry hold “dual roles of promotion and oversight” (p.579). “the people likely to take the brunt of technology missteps are not necessarily those who the developers and implementers answer to directly” (p.579). “the risk conversation cannot afford to be only driven by the developers and promoters of the technology” (p.580). “in the absence of business being willing to be more transparent and properly self regulate, government will step in” (p.581). - Ethics and pragmatics of engagement. “Ethically, it is questionable to deny citizens the opportunity to be a part of the process of technology innovation where it potentially impacts on their lives and livelihoods” (p.580). Pragmatically, GM foods showed that ignoring citizens is “a serious mistake” (p.580). - Comparative risk. No confirmed deaths attributed to nanomaterials, set against 34,017 US road deaths in 2008 and 26,000 child deaths a day from poverty-related causes (p.581). - Looking ahead. “increasingly revolutionary technologies are waiting in the wings – active nanomaterials, smart nano-devices, synthetic biology, advanced robotics and information technology … The only question is, are we prepared?” (p.583). - Brand. Nanotechnology as a brand, “wonderfully ambiguous”, aiming “to evoke images and emotions rather than precision” (p.584, note 1).

What this adds to the map. - The most direct evidence in this share for Andrew’s note 1. Quantitative risk assessment is described as “professional and competent” (p.582), and new approaches are to be “grounded in established approaches” (p.575). What is criticised is not the method but its use: invisible experts, political modulation, and a backward-looking bias towards “well-established risks”. That last point is also the clearest early statement of why quantitative methods are weak for emerging technologies (note 2). - Earlier origin: plausibility. “plausible emerging risks” (p.575) and “grounded in current realities and probable developments” (p.577) put the plausible-versus-imaginable discipline (C7, formalised in FFTF 2018) in 2010, in co-authored form. - Earlier origin: the 2026 split between public standing and expert action. The map treats the lecture’s “you cannot hand a problem of this magnitude over to everyday people” as a single-source 2026 statement [mixed] (C5; T5; tension 9). The same structure appears here in 2010: people should be empowered, and “the details … will of necessity remain the responsibility of a small number of experts” (p.583). The 2026 position is not new. It is an old position restated, which strengthens the map’s evidence for it. - Earlier origin: structural accounts (C10; §5.2 “structural incentives”, from 2022 in his own prose). Promotion–oversight conflicts and accountability mismatch are set out in 2010. - Earlier origin: brand-nano (map: 2018-02-21), in co-authored form. - Earlier origin: comparative risk against a baseline (§5.1, “occasional”, 2021–25). - Two-sided engagement. The “two dangers” passage is balanced in a way the map’s C5 and C9 do not fully capture: he warns against technocratic disregard of values and against perception overriding evidence. The map should present his engagement commitment as bounded by evidence, not only as a commitment to publics.

C3. Seven11: “Seven challenges to regulating ‘sophisticated materials’” (2020science, 22 July 2011)#

Provenance. His post. The body summarises the Handbook10 chapter and the co-authored Nature Materials commentary (Maynard, Bowman and Hodge), largely in the chapter’s words. The text is truncated in the middle of “Transparency and Trust”.

The argument. As Handbook10, framed by his introduction: regulations were built for “simple” materials; researchers now design materials “from the ground up” with “few analogs”; regulators have so far stretched existing frameworks, but sophisticated materials “will transcend current regulatory mindsets”.

What this adds to the map. Mainly confirmation that he endorsed the Handbook10 positions under his own name, including the passage on quantitative risk assessment and “invisible experts”. His introduction also places “sophisticated materials” (§7 Phase 0) in 2011 in his own prose.

C4. Define11a and Define11b: “Don’t define nanomaterials” (early and penultimate drafts, July 2011)#

Provenance. His own drafts, posted because the published Nature commentary is paywalled. He notes that the published version is more focused but “I’m not sure it has so much of my ‘voice’ in the phrasing and nuances” (Define11a, end notes). The drafts are the better evidence of his thinking.

The argument. “Five years ago, I was a strong proponent of developing a regulatory definition of engineered nanomaterials. Today I am not” (Define11a). In Define11b: “I have changed my mind” (PDF p.1). The science shows no bright line between nano and non-nano risk; risk depends on size, shape, porosity, surface and bulk chemistry, and context. A policy-made definition would be “a term of art, not of science” (Define11a; Define11b p.2), and would force policymakers “to use science to support an assumption, rather than modify their assumptions based on the evidence” (Define11b p.3). The alternative is to identify “materials and products that raise plausible and specific concerns – irrespective of what they are called” (Define11a), through “evidence-informed trigger points that indicate property-driven courses of regulatory action” (Define11b p.3). “Materials need to be regulated by the potential risks they present, and not by the technological labels that come attached to them” (Define11b p.3).

What this adds to the map. - The earliest signalled change of mind in the record. The map’s table of changes “signalled by him” begins in 2020. This one is from 2011 and is announced in the first two sentences. It also shows how he updates (§7): a change driven by accumulating evidence about mechanism, stated plainly. - Earlier origin: behaviour, not labels (C13; map anchor 2022-02-10). It is fully formed in 2011, and first appears in TenThings09. - Earlier origin: plausibility (“plausible and specific concerns”). - A method he could carry over. Trigger points based on properties that indicate risk, rather than category definitions, is a regulatory design idea the map does not record. It fits his later scepticism about technology-specific hard law. - Consistency across apparent reversals. In T2007 (PDF p.18) he criticised EPA for treating nano and bulk forms as the same chemical because they share a “molecular identity”. In 2011 he opposes a nano definition. Both apply one rule: regulate by behaviour, not by name.

D. The 2020science risk record (2008–2016)#

D1. LateLessons08: “Late lessons from early warnings” (2020science, 20 July 2008; first on the SAFENANO blog)#

Provenance. His framing, a quoted conclusion from the co-authored Nature Nanotechnology commentary (Hansen, Maynard, Baun and Tickner 2008), and the EEA’s list of twelve lessons (the EEA’s words, reproduced). Only his framing and, with lower weight, the co-authored quotation count. Per the brief, the EEA material is not analysed here.

The argument. His framing: some lessons “have begun to sink in”, but “a refresher course in responsible nanotechnology wouldn’t go amiss”; “Nanotechnology is all about the future. But it seems an occasional glance back in history is needed”. The co-authored conclusion: nanotechnology has learnt “new tricks” (asking critical questions early, collaboration, engagement), but “The question seems not to be whether we have learnt the lessons, but whether we are applying them effectively enough”. Nanotechnology “is being overseen by the same government organizations that promote it”, and slogans such as “‘risk research jeopardizes innovation’ or ‘regulation is bad for business’” cloud the waters.

What this adds to the map. It confirms his direct engagement with the EEA framework in 2008 (the map cites the reports only via a 2018 post). The distinction between learning lessons and applying them is his own recurring theme in this share (see D4 and D11). The promotion–oversight conflict appears again, in co-authored form.

D2. TenThings09: “Ten things everyone should know about nanotechnology safety” (2020science, 29 August 2009)#

Provenance. His own talk turned into a post; “a rather more personal perspective” and “my own reflections”.

The argument and key concepts (in his order, 10 to 1). - 10. “There’s no such thing as ‘nanotechnology safety’”. The NNI definition is “one of expedience, not of science”. Generalising about nanotechnology leads to “rationality by-pass”. Nanotechnology “—like most technologies—is safety-neutral. It isn’t the technology so much as what is done with it that is important.” - 9. “We’re living in a post-chemistry world”. Physical form matters as well as chemistry. “No exposure—no harm.” - 8. Knowledge has “more holes than a Swiss cheese”. - 7. Nanomaterials are “shape-shifters” whose hazard changes over the life cycle. - 6. “The technology’s new, but that doesn’t make old safety practices redundant”. Established occupational hygiene works “even if hard data on a new material’s toxicity are lacking”; “old tricks may work with new technologies, but probably only up to a point”; “there is a world of difference between safe and safer.” - 5. “Lower exposures mean lower risks”. “less stuff means lower risk”, as a rule of thumb for acting without toxicity data. - 4. “Measurement without meaning is like a car without an engine”. “Numbers—hard data—can be comforting. But without a clear idea of their relevance, they can also be misleading.” In the absence of limits, BSI’s “rough and ready” benchmarks give meaning to measurements, “as long at the working benchmark levels do not become set in stone”. - 3. “When the data run out – innovate!”. Unknowns “pull the rug out from under conventional approaches to quantifying and managing risks”. “More than ever in the future, we will have to rely on new and innovative approaches to managing risks; ones that enable decisions to be made in the absence of hard data.” He endorses “soft” (qualitative) approaches such as expert judgement and control banding. A science-based understanding of risk “looks increasingly like a Swiss cheese, no matter how hard we try.” - 2. “It’s good to talk”. “Safety shouldn’t be a competitive issue.” - 1. “People matter”. “the primary focus of risk research should be the people it ultimately impacts”; “It isn’t about the buzz of new discovery. It isn’t about getting rich and famous. It isn’t about making a profit. And it isn’t about sustaining ideologies.” He worries that newcomers to nano-risk research lack the public-health culture “of putting others first”. “This is a very personal perspective, and I may be wrong.”

What this adds to the map. - The single best source in this share for Andrew’s two notes. In 2009 his foundation is the quantitative occupational-health toolkit: exposure, dose–response, exposure limits, measurement. He says plainly that “old safety practices” are not redundant (note 1). He also says that when data run out, the right response is to innovate, not to hang onto numbers, because numbers “can be comforting” but “misleading” (note 2). The structure of his AI position is already here: where hazard and exposure cannot yet be characterised, use rules of thumb, benchmarks, precautionary exposure reduction, qualitative judgement and shared knowledge, and keep asking what matters to people. - Earlier origins. The phrase “It’s good to talk” (map: FFTF pp.226–229) is here in 2009, applied to sharing safety knowledge among firms. “Behaviour, not labels” (C13) begins here. “I may be wrong” (map method note: FFTF p.170) appears in 2009. The risk-innovation impulse (C3) is present six years before the name. - His own statement of the “regulate use, not technology” rule. The map records that in 2023 he doubted the nano-era rule of regulating “what people do with the technology” for general-purpose AI (§5.3; changes of mind table). This post shows the rule was his own (“safety-neutral”; “what is done with it”). The 2023 doubt is therefore a revision of his own earlier principle, not only of his field’s, and the map could say so. - “People matter” as an early statement of purpose. It is an early form of C1 (the point of risk thinking is people thriving) and of T11, stated as professional ethics.

D3. Deepwater10: “Beyond the obvious – lessons from the Deepwater Horizon Oil Spill” (2020science, 25 October 2010)#

Provenance. His own draft of a piece for the University of Michigan School of Public Health magazine.

The argument. The spill shows “what can go wrong when we trust in technology without investing sufficiently in the future.” Innovation failed on three counts: consequences of an unproven technology were not explored; there was too little upstream investment in understanding and mitigating risk; and there was no foresight in developing technologies to manage failure. “more realistic scenario planning would have helped prepare for low probability but high impact risks.” It is “naïve” to assume “that technology-based solutions will present themselves as and when needed.” “as emerging technologies become increasingly complex and powerful, the consequences of mis-steps on public health and the environment will only become more catastrophic.” He calls for “a new paradigm that places a science-based understanding of risk at the center of sustainable development”. Otherwise we gamble, and “the house always wins – eventually.”

What this adds to the map. - Earlier origin: taking tails seriously. The map’s tension 2 (plausibility versus tails) presents readiness to consider low-probability scenarios as growing from 2025. Here in 2010 he calls for scenario planning for “low probability but high impact risks”. Together with Brain14, this supports the map’s view that the change is one of emphasis, and moves the evidence for the pairing back to 2010–14. - Earlier origin: rising consequence and irreversibility (C8; FFTF pp.166–167), stated in 2010. - Bearing on note 1. “a science-based understanding of risk at the center” is the foundation he builds on.

D4. Worrying11: “What was worrying us about nanotechnology safety seven years ago?” (2020science, 9 August 2011)#

Provenance. A short framing by him, then the full research and regulatory recommendations of the 2004 Buxton symposium (the first in a series he co-chaired, per the T2006 biography). The recommendations are a workshop product; only his framing counts.

The argument. Rereading the 2004 recommendations: “Disturbingly, they look remarkably similar to recommendations still being made.” “So are we making progress, or are we simply going round in circles?”

What this adds to the map. A first-hand record of slow uptake of early warnings in his own field, in his words, seven years on. The 2004 workshop’s line that “there is no need for a new risk management paradigm … but there is a need for new tools” is not his statement, but it shows the consensus he worked within in 2004. By 2015 he was calling for “a radical new approach to risk” (Parallel15). The shift from “new tools within the paradigm” to “a new domain” is a real development, and it was additive.

D5. Speculated12: “Exploring speculated catastrophe and mundane reality” (2020science, 4 February 2012)#

Provenance. His own short post about a class he taught with the speculative designer James King at Michigan.

The argument. Science and public-health students used creative work to capture “the tension between the catastrophic consequences often imagined to arise from human endeavors, and the mundane reality that often develops.” Released from “the rigid limitations of their science education”, they told “science-grounded stories that connected with people on a far deeper level than just the facts would allow.”

What this adds to the map. The earliest evidence in this share of his use of speculative art and story in teaching about risk (map: stories as tools from FFTF 2018; teaching with film 2019–22). It marks the turn from CETI10’s “science fact not science fiction”. The catastrophe-versus-mundane frame is an early form of “far more mundane–but no less serious” (2020-11-12).

D6. Decade14: “A decade of uncertainty in nanoscale science and engineering” (Nature Nanotechnology column, March 2014; reposted November 2014)#

Provenance. His own single-authored column.

The argument. The 2004 Royal Society–Royal Academy of Engineering report turned grey-goo speculation into “a solid foundation of an evidence-based approach to plausible nanoscale material risks”. Research output rose tenfold. But the legacy has a “darker side”: driven by “fears of novel and unique behaviour”, millions went into materials that “don’t always seem to hold the dramatic risk surprises that some thought they would”, encouraging researchers “to dig ever-deeper to discover the assumed-to-exist evidence of novel risk”. “The speculation of possible risk has developed into an assumption of as-yet-to-be-discovered risk.” The field has “worn a rut” and may have created “a new, metaphorical grey goo”. He suggests using “speculation as a lever” to move towards evidence-based approaches to the next generation of materials.

What this adds to the map. - Humility turned on risk science itself. The map’s C7 applies plausibility to “hype and doom alike”. This column applies it to precaution and risk research: a risk community can lock itself into assumed hazards and miss new ones. It is an important counterweight for any reading of his work as simply precautionary. - Bearing on note 2. Hubris in risk assessment cuts both ways: taking comfort in methods and numbers, and over-investing in assumed risks. Humility here means keeping research tied to evidence and staying open to what has not been looked at.

D7. Brain14: “Is 3D printing an artificial brain plausible? And what are the risks?” (2020science, 11 December 2014)#

Provenance. His own post introducing his Nature Nanotechnology commentary and a Slate piece. The commentary text is in the corpus (2023-12-03); the lines below are only in this post.

The argument. Additive manufacturing makes complexity cheap, which could allow three-dimensional neuromorphic substrates and eventually an artificial mind. On AI risk: “I personally don’t buy this vision of an AI ‘singularity’ — being an academic surrounded by brilliant minds leaves you a little jaded as to what brilliant minds can actually achieve!” The more plausible risks are “artificial minds that challenge our very notions of humanity”, with “moral risks”: “To what extent does an intelligent machine have rights?”; “Could withholding moral rights from machines corrode moral codes within human communities”; “Will prolonged interactions with intelligent machine change human behavior in potentially harmful ways?” Also, a more likely danger than malevolent superintelligence: “demonstrating the dangers of relying on intelligence to do the smart thing by — with the best intentions in the world — catastrophically messing up the planet we live on.” These are “incredibly speculative and certainly not empirically testable risks. But technology innovation has a habit of turning on a dime”, so “it would seem foolish not to use this and similar speculative scenarios” to prepare. “This is a part of risk science that needs the freedom to dream, and the realism to anchor those dreams in plausible outcomes.”

What this adds to the map. - The earliest AI-risk statements in the record. The map dates AI’s entry to 2015-01-30 and superintelligence agnosticism to FFTF (2018). This post from December 2014 already has superintelligence scepticism; AI’s moral status as a risk to human moral codes (map: FFTF p.58; 2023–24); AI changing human behaviour through prolonged interaction (the core of C14, which the map dates from 2016 for neurotechnology and 2018 for AI); intelligence not being goodness (FFTF p.108); and well-meaning intelligence causing catastrophe (myopic benevolence). It also ties AI to “our very notions of humanity” (C16). - Tension 2 was there from the start. In one paragraph he holds both halves of the map’s tension: untestable speculative risks are worth preparing for because innovation “turn[s] on a dime”, and dreams must be anchored “in plausible outcomes”. This supports the map’s reading that the later change is one of emphasis. - Bearing on note 2. A clear early statement that some AI risks are “certainly not empirically testable”, and that the right response is prepared, humble speculation, not quantification.

D8. Novelty14: “Is novelty in nanomaterials overrated when it comes to risk?” (Nature Nanotechnology column, 2014; reposted 2 February 2015)#

Provenance. His own single-authored column, reposted in full.

The argument. Linking novel properties to novel risk made sense early on, but “novelty” is subjective and transient: materials change when combined into products and released. “Novelty as a result is a subjective, transient, and consequently a rather unreliable indicator of potential risk. It tends to obscure the reality that conventional behaviour can sometimes lead to harm, and that mundane risks are still risks. And it favours the interesting (and possibly the headline-grabbing) over the important.” Better to map “plausible domains of risk” through life-cycle exposure for realistic products, “filtering out plausible modes of harm from the merely speculative and ‘novel’”. Harm rests on “the hard reality of material–biology interactions, and not on an arbitrary determination of novelty”. Materials will increasingly be “defined by what they do rather than what they are called”.

What this adds to the map. - A new, transferable concept: novelty is an unreliable sentinel of risk, and the interesting crowds out the important. It is a precursor of his later account of how attention selects risks (orphan risks, 2018; risk selection, 2026 [mixed]). - Earlier origins: behaviour, not labels (C13), and plausibility (C7), both in 2014. - A nuance for “defies analogy”. In 2014 he warned that fixation on novelty hides “mundane” risks. It is worth setting beside his 2024–26 emphasis on AI’s discontinuity (tension 1). His earlier self would ask which of AI’s risks are novel and which are ordinary risks in new clothes. My reading: this is not a contradiction, but it is a check on over-reading “defies analogy”.

D9. Aitken15: “Characterizing nanoparticles in the 1880’s” (2020science, 5 July 2015)#

Provenance. His own post introducing his column “Learning from the past”.

The argument. John Aitken counted airborne nanoparticles in 1889 (52,000 per cubic centimetre atop the Eiffel Tower) with an instrument whose principles underlie modern ones. This challenges “a myth … that engineered nanoparticles are so new and novel that we don’t have the means to characterize them, or measure exposures.”

What this adds to the map. Deeper evidence for note 1 and for C13: a new technology does not make old knowledge and instruments obsolete. The map cites the same lesson from a 2024 repost (“seemingly novel challenges don’t always demand novel solutions”); this is the 2015 original.

D10. Parallel15: “For tech innovation to succeed, we need parallel innovation in how we think about risk” (2020science, 7 September 2015)#

Provenance. His excerpt from his column “Why we need risk innovation” (Nature Nanotechnology, published online 3 September 2015).

The argument. Google’s proposed nanoparticle diagnostic pill faces health risks, but “the probability of causing harm is not the only risk”: outmoded regulation, investor ambivalence, consumer suspicion and social-media backlash also threaten it. He notes the Future of Life Institute’s AI-safety grants and CRISPR embryo research as signs of the same problem: “a growing disconnect between the rate at which we are innovating, and our ability to assess and manage the adverse consequences”. “Important as evidence-based health and environmental risk assessment and management are, they fail to capture the full panoply of personal, social, environmental, technological, economic, political and corporate risks”. So “we need a radical new approach to risk … parallel innovation in how we conceptualize risk … risk innovation”.

What this adds to the map. It fixes the date of the column behind the map’s “2016-01-11 (from a 2015 column)”. Its key sentence supports note 1 directly: evidence-based risk assessment is “important”; the problem is what it fails to capture. AI appears in 2015 as one example among several emerging technologies, as the map says.

D11. CNT16: “What’s the latest on carbon nanotube safety?” (2020science, 15 June 2016)#

Provenance. His own post introducing his column “Are we ready for spray-on carbon nanotubes?”

The argument. The science now shows that inhaled carbon nanotubes can cause inflammation, fibrosis and possibly mesothelioma, and can promote cancer. In 2007 he had questioned treating them as graphite (T2007). NIOSH’s 2013 recommended limit was “a thousand times lower” than suppliers’ figures, yet the supplier’s safety data sheet still treated nanotubes as nuisance dust. “Clearly, despite the science moving on, not a lot has.”

What this adds to the map. A compact, first-hand case of an early warning (his own, 2007), confirmed by science (2013) and still not acted on by a supplier (2016). It gives concrete content to the map’s “attention decay” point (2016-02-01) and to his refrain that learning lessons is not the same as applying them (LateLessons08; Worrying11).

D12. Navigating16: “Navigating the nanotechnology risk landscape – pointers for early career scientists” (2020science, 28 July 2016)#

Provenance. His own summary of his June 2016 Nature Nanotechnology column.

The argument (seven guideposts). (1) “Risk starts with something that is worth protecting”: “risk concerns threats to something you or others value”, including “security, friendships, social acceptance, and our sense of personal and cultural identity”. (2) “‘Nanotechnology’ is an unreliable indicator of risk.” (3) “We live in a post-chemicals world”: “Treating nanomaterials as if they are precisely-defined chemicals can lead to serious errors of judgment where risk is concerned.” (4) “Benchmarking is important”: it helps “place plausible boundaries around the potential impacts” and “avoid over-speculation”. (5) We have co-evolved with nanoscale materials. (6) “How you think about nanotechnology risk is probably incomplete”: researchers must “constantly re-assess their assumptions” and have “the humility to recognize and respect expertise outside of their domain”. (7) “We need to be quick to question, and slow to respond”: premature action on “potentially misleading immature science” risks “hard-to-rescind decisions”, “Yet we also need to ability [sic] to respond proactively where early warnings of potential harm do begin to emerge – even before the science is mature.” He closes by calling these his “admittedly limited and probably occasionally blinkered experiences”.

What this adds to the map. - Threat to value carried into his home discipline in 2016. The map traces the frame from startups outwards. Here it is applied to nanomaterials research, alongside the conventional grammar, which shows the two definitions of risk sitting side by side (tension 3). - “Quick to question, slow to respond” is a new concept and his most precise statement on acting on early warnings. It balances the cost of acting on immature science against the need to act before the science matures. The map’s account of precaution (§5.1: proportionate, participatory, scaled to irreversibility) gains a timing rule here. - Bearing on both notes. Benchmarking (a quantitative practice) is valued because it bounds speculation (note 1). Treating complex things as precisely defined leads to “serious errors of judgment”, and every discipline’s picture “is probably incomplete” (note 2).

E. 2017#

E1. Guardian17: “It’s time for some messy, democratic discussions about the future of AI” (The Guardian, 1 February 2017)#

Provenance. Co-written with Jack Stilgoe. Shared position. It is in line with his single-authored statements of the time (Brain14; FFTF ch.8).

The argument. The 1975 Asilomar meeting is “still held up as a beacon of scientific responsibility”, but it was driven by scientists wanting to head off regulation through self-governance, and was blind to corporate futures (Genentech followed a year later). The 2017 Asilomar AI meeting included Google, Facebook and Tesla but had “no public and their concerns, no journalists, and few experts in the responsible development of new technologies.” Its principles are “all rather Motherhood and Apple Pie: comforting and hard to argue against, but lacking substance” and “short on accountability”. “Scientists are more inclined to guess at what the public are worried about than to ask them”. Superintelligence talk is “an AI echo chamber, in which speculative discussions have taken on a moral significance that far exceeds their social importance”, “despite rather long odds on the viability of this scenario.” “AI is already a thing in the world, enabling and constraining our lives in ways that we barely understand.” Governance must “get beyond follies and trollies”; “the decisions can’t just be taken by a narrow group of experts”; citizens probably care more about livelihoods, security and distribution. The principles “should be treated as hypotheses … They now need to be democratically tested.”

What this adds to the map. - The earliest published critique of AI industry self-governance in the record, with named companies, a year before FFTF. The map’s Asilomar references are 2023–25 and largely approving of 1975 as precedent (T2; 2025-02-23). This co-written piece is sharper about 1975, and the later approval should be read alongside it. - Earlier origin: operationalised ethics (§5.3; map anchor 2019-04-15): principles “lacking substance” and “short on accountability”. - Earlier origin: the ethics-to-risk reframe (map 2019–23): trolley-problem ethics as convenient to engineers, versus livelihood risks. - A small point for tension 9. The complaint that there were “few experts in the responsible development of new technologies” fits the map’s reading that his remedy for technical-expert monopoly often adds his own kind of expertise. - Bearing on note 2. “in ways that we barely understand” is a 2017 statement of the ignorance that the map’s AI chapter attributes mainly to 2023–26.

E2. Rethinking17: “Rethinking Risk” (in Visions, Ventures, Escape Velocities, ASU Center for Science and the Imagination, 2017, pp.193–201)#

Provenance. His own single-authored chapter responding to two short stories.

The argument. “I’m not sure I buy the idea of ‘risk aversion’” (p.193): the concept hides “what is at risk, and what the consequences of failure or loss are” (p.193). “Risk—at least in the analytical sense—depends on numbers” (p.193). Probability “is a powerful way of making trade-offs between different choices” and “takes some (but not all) of the unpredictability out of decisions. Yet numbers can be deceptive” (p.194). A 99% chance of success still means failure one time in a hundred, “possibly more, if there are incalculable uncertainties involved” (p.194). “Risk calculations are also highly dependent on what is considered important, as well as who decides what’s important” (p.194), and financial or political risk numbers “will be meaningless to people who may stand to lose their health, livelihood, or dignity” (p.194). Heuristics are “a great evolutionary response to staying alive” but unreliable for “risks we haven’t evolved to handle every day” (p.194). His confession: facing a one-in-a-million risk from a CT contrast dye, “As a physicist, I’m expected to be good with numbers” (p.195), yet he signed “not because I’d done the math and it made sense, but because that was what I was expected to do”, and the value that decided was avoiding embarrassment (p.195). Organisations too have “institutional heuristics” and can rationally refuse to risk “identity death” (p.196). Threat to value “extends far beyond conventional metrics of risk” and costs something (“imagine a regulator including interpersonal relationships in risk assessments—it’s hardly likely to make the process any easier”, p.197). It lifts risk talk beyond “simplistic ‘go/no-go’ options” (p.197). His conclusion: “an evolution of the old black-and-white mathematics of risk” (p.200).

What this adds to the map. - The clearest statement anywhere in this share of how the two definitions of risk relate (tension 3). Probability is the analytical core and a powerful trade-off tool; value decides what the numbers are about and who they mean anything to. He does not give an integration rule, but he admits the cost of the broader frame, which the map does not record. - Direct evidence for both notes. The word he chooses is “evolution”, not replacement (note 1). “numbers can be deceptive” and “incalculable uncertainties” (note 2). - Earlier origins. Evolved heuristics misfiring in a technological world (2017) prefigures the evolutionary-mismatch argument of 2026-01-10 (§5.8). The CT-scan confession is self-implication a year before FFTF. His rejection of “risk aversion” as an explanation supports the map’s claim that resisters are protecting what they value (T7; C9). - On “whose value?” (tension 4). The chapter says the value that matters “isn’t always universally shared”, that what employees value “may differ from what’s of value to the organization they work for” (p.197), and that decisions should protect what is valuable “not just to corporations and governments, but also to individuals and the communities they are a part of” (p.200). This answers part of the map’s worry that the frame is enterprise-facing.

F. Cross-cutting findings#

F1. Earlier origins of ideas the map dates later#

F2. Corrections and qualifications to the map#

1. Phase 0 now has primary sources. The formative Maynard (2006–2011) was a forceful, quantitatively literate advocate for strategic, well-funded, transparent risk science, with a strong central coordinating role. The map’s Phase 0 row (“known from later accounts”) can be replaced with these documents.
2. The earliest signalled change of mind is 2011 (regulatory definitions of nanomaterials), not 2020.
3. The honest-broker trajectory is U-shaped, not linear (T8; tension 12). My reading: he testified to Congress as a forceful advocate in 2006–08 (“out of touch with reality”, T2007 PDF p.21), while his institutions described themselves as “non-advocacy” (GI08a; T2006 rec. p.52). He adopted Pielke’s honest-broker language in 2018 and moved back towards advocacy from 2024. His 2026 regret that “policy” became “intelligence” (Prehistory26) suggests he never fully accepted a purely informational role.
4. Central capacity as well as adaptive governance. Alongside the adaptive, multi-stakeholder, soft-law portfolio (C12), he has argued for strong, authoritative central coordination of risk research, dedicated budget shares (10% of R&D), measurable goals, full transparency, and independent jointly funded research bodies on the HEI model (independence, transparency, review, communication, relevance). The map’s list of his governance instruments is incomplete without these.
5. Engagement mechanisms were concrete early, and mixed with awareness-raising. In 2007–08 he proposed a federal advisory committee and a funded engagement programme, and he treated low public awareness as a failure of government. The map’s claim that his engagement is thin on mechanism (tension 9) is true of 2018–26, not of his formative years. His literacy remedy (map: 2023, declining 2025) has 2008 roots.
6. The 2026 split between public standing and expert action is old. Handbook10 (p.583, co-authored) makes the same split in 2010. This reduces the map’s reliance on the [mixed] lecture for that claim.
7. The “regulate use, not technology” rule was his own (TenThings09). His 2023 doubt about it for AI is a revision of his own principle.
8. Fiction came late as a method. In 2010 “science fiction” was a pejorative (CETI10; Handbook10 on “imagined futures”). The turn to speculative design and story is visible from 2012–14.
9. Humility applies to precaution too. Decade14 and Navigating16 show him warning that risk research can rut into assumed hazards, and that premature action on immature science carries its own risks. The map’s plausibility thread should include this self-critique of the risk enterprise.
10. “Early warning” is not his term in the WEF proposals. The documents speak of predicting, assessing and avoiding adverse consequences, and of horizon scanning and intelligence.

F3. Evidence bearing on Andrew’s two notes#

Note 1: quantitative risk assessment remains a foundation, built on rather than abandoned. The evidence in this share is consistent and strong. - His formative practice is quantitative: exposure measurement, dose–response, occupational exposure limits, budget and relevance accounting (T2006; T2007; T2008; Funding08; TenThings09). - He repeatedly says the old tools remain: “old safety practices” are not “redundant” (TenThings09); new approaches should be “grounded in established approaches to identifying, assessing and managing risks” (Handbook10 p.575); evidence-based risk assessment is “important” but incomplete (Parallel15); evidence-based decision-making is “not sufficient on its own” (CETI10 PDF p.2); risk innovation is “an evolution of the old black-and-white mathematics of risk” (Rethinking17 p.200); the past offers instruments and knowledge that novelty-talk ignores (Aitken15). - Quantitative risk assessment is called “professional and competent” (Handbook10 p.582). What is criticised is its use by “invisible experts”, its political modulation, and its tendency “to deal retrospectively with well-established risks”. - Implication for the map: C2 (“extends conventional thinking rather than replacing it”, 2018) and C4 (“identity-defining”) are right. The map’s language on “impatience with” risk science (§2) should be balanced by this record of loyalty and additive development.

Note 2: the lack of quantitative AI methods reflects humility about ignorance, not neglect. There is strong support, with some evidence that complicates it. - Support. He has distrusted numbers as comfort since at least 2008: “a feel-good figure” (Funding08); “Numbers—hard data—can be comforting. But without a clear idea of their relevance, they can also be misleading” (TenThings09); “we must not mistake methodology for strategy” (T2007 PDF p.21); “numbers can be deceptive” and may hide “incalculable uncertainties” (Rethinking17 p.194). He has acknowledged ignorance even in his most quantitative documents: “we do not yet know what are the right questions to ask” (T2007 PDF p.33); science-based understanding “looks increasingly like a Swiss cheese, no matter how hard we try” (TenThings09); “How you think about nanotechnology risk is probably incomplete” (Navigating16); AI shapes lives “in ways that we barely understand” (Guardian17); some AI risks are “certainly not empirically testable” (Brain14). His stated rule for such conditions is 2009’s: “When the data run out – innovate!”, meaning decisions “in the absence of hard data” through rules of thumb, benchmarks, “soft” approaches, precautionary exposure reduction, shared knowledge and attention to what people value. - Complicating evidence. Where data can be had, he demands them and uses numbers hard: measurable goals and metrics (T2006 rec. p.57; T2008 PDF p.2), project-level transparency without which assessments are “worthless” (T2008 PDF p.12), mandatory data reporting to replace “speculation” with “reality” (Weighing09, co-written), benchmarking to “avoid over-speculation” (Navigating16), comparative risk figures (Handbook10 p.581), and evidence-based “trigger points” (Define11b). Humility for him has never meant avoiding numbers. It means refusing numbers that answer the wrong question or that stand in for understanding. My reading: the early record suggests quantitative tools he could still bring to AI without hubris, such as accounting for how much risk research is funded and how relevant it is, and demanding disclosure. The map’s “missing methods” gap should be reframed. Its absence reflects a deliberate judgement about AI hazard and exposure (consistent with “no framework yet exists”, 2023-11-26), not a break with his quantitative past, and there is quantitative work he has done before that the judgement leaves room for. - Humility in both directions. Decade14 and Navigating16 apply the same humility to precaution: assumed risks can become a “new, metaphorical grey goo”, and acting on immature science can do harm. So his humility does not lean only towards alarm.

Digest: the most important additions from S3#

This share gives the map what it lacked: primary evidence for Andrew Maynard’s formative years (2006–2011) and the 2014–17 bridge into his risk-innovation work. It is almost all his own writing. The exceptions are the co-authored 2010 Handbook conclusions (reposted under his name), the WEF CETI proposal (drafted with Tim Harper), a draft co-written with David Rejeski, and a 2017 Guardian op-ed co-written with Jack Stilgoe.

1. Both of Andrew’s notes are borne out, with documentary roots going back almost twenty years. The early Maynard is a quantitative risk scientist. He counts research dollars and relevance, demands metrics and exposure data, and wants uncertainty reduced “through systematic scientific research” (2006). He never treats that foundation as disposable. In his words, “old safety practices” are not “redundant” (2009). New approaches should be “grounded in established approaches” (2010). Evidence-based assessment is “important” but incomplete (2015). Risk innovation is “an evolution of the old black-and-white mathematics of risk” (2017). Alongside this, from the start, he distrusts numbers as comfort: “Numbers—hard data—can be comforting. But … they can also be misleading” (2009). He warns against “a feel-good figure” (2008) and against mistaking “methodology for strategy” (2007). Even in his most quantitative testimony he admits “we do not yet know what are the right questions to ask” (2007). His 2009 rule, “When the data run out – innovate!”, is the template for his AI stance: decide without hard data by using rules of thumb, benchmarks, precautionary exposure reduction, qualitative judgement and engagement. The humility runs both ways. In 2014 he warned that risk research itself can rut into assumed hazards (a “new, metaphorical grey goo”), and in 2016 he wrote “quick to question, and slow to respond”, while still acting on early warnings “before the science is mature”.

2. Many ideas the map dates to 2018 or later are much older. These include symmetric risk and the risk of public rejection (2006), who decides and who pays (2008), could versus should (2008), “good intentions are not enough” (2007), behaviour not labels (2009–11), plausibility (2010–11), taking low-probability, high-impact scenarios seriously (2010), the pacing gap (2007), orphan-risk logic in research funding (2006–07), structural conflicts between promotion and oversight (2007–10), and “It’s good to talk” (2009). His first documented AI-risk writing, from December 2014, already rejects the singularity. It asks whether “prolonged interactions with intelligent machine[s]” might change human behaviour for the worse, and raises machine rights as a risk to human moral codes. It also holds both halves of the map’s plausibility-versus-tails tension in one paragraph.

3. Corrections and qualifications. - His first signalled change of mind is in 2011, on regulatory definitions of nanomaterials, not in 2020. - He was a forceful advocate before Congress years before adopting the honest-broker label, so that trajectory is U-shaped. - He has long wanted strong central capacity for risk research: a single accountable leader, 10% of R&D, and independent bodies modelled on the Health Effects Institute, built on independence, transparency, review, communication and relevance. This sits alongside adaptive, soft-law regulation. - His engagement mechanisms were concrete in 2007–08 (an advisory committee and a funded engagement programme) and were mixed with awareness-raising, which is where his later literacy remedy comes from. - The 2026 lecture’s split between the public’s standing and experts’ crafting of rules appears in 2010 (“the details … will of necessity remain the responsibility of a small number of experts”). - The nano-era rule of regulating use rather than technology was his own (2009). - His turn to fiction as a lens came after 2010, when “science fiction” was still a pejorative for him.

4. Institutions and his own late lessons. The WEF proposals (2008 and 2010) called for an independent, jointly funded but funder-independent body for anticipatory intelligence on emerging technologies. They do not use the term “early warning”. In 2026 he says that renaming “policy” as “intelligence” narrowed the ambition “in ways I still have mixed feelings about”, and that his 2008 questions are today’s AI governance questions. His own record also documents how early warnings are acted on slowly. Carbon nanotubes were questioned in 2007, NIOSH set a limit a thousand times lower in 2013, and a supplier’s safety data sheet was unchanged in 2016: “despite the science moving on, not a lot has.” In 2011 he found the research recommendations of a 2004 workshop still being repeated.

5. How he reads technology promoters, in general terms. Innovation and harm-avoidance “need not be, nor should be, mutually exclusive”. The risk–benefit split is “a largely false dichotomy”. Safety will not emerge “if the promoters of the technology are calling all the shots”, nor from market mechanisms “on their own”. Voluntary industry disclosure has failed (2009, co-written). And AI principles written without the public are “Motherhood and Apple Pie” that should be “democratically tested”.