Late Lessons, Jensen Huang and AI
Markdown version

The Late Lessons reports in context: institution, intellectual traditions, afterlife#

Strand A working note. Written 26 September 2026 from primary sources where possible. Scope: what kind of documents the two EEA Late lessons from early warnings reports are, where they came from, what they draw on, and how they have been used since.

Page conventions. “LL1” is EEA Environmental issue report No 22 (2001); report page = PDF page. “LL2” is EEA Report No 1/2013; printed page = PDF page − 2. Both were checked against copies from the EEA website (LL2 full report via the Internet Archive’s copy, labelled “(05-2013)”).

Disclosure. Andrew Maynard co-authored LL2 Chapter 22 on nanotechnology (LL2 p. 530), based partly on Hansen, Maynard, Baun & Tickner (2008). Section 3 cites it only as one example of scholarship built on LL1.

0. Short answer: what kind of documents are these?#

1. Institutional setting#

1.1 The EEA’s mandate#

The EEA was set up by Council Regulation (EEC) No 1210/90 of 7 May 1990. That regulation was amended by Regulation (EC) No 933/1999 and Regulation (EC) No 1641/2003, then codified as Regulation (EC) No 401/2009. LL1 describes the EEA as “an independent Agency of the European Community established in 1993 to provide objective information to the policy-making bodies of the EU and its Member States”. It cites “Council Regulations, 1210/90 and 993/99” (LL1 p. 11); “993/99” is a misprint for 933/1999. The EEA’s 2019 Drivers of change report marks its 25th year as “1994–2019”.

Under the codified regulation, the Agency’s objective is to give the Community and Member States “objective, reliable and comparable information at European level” so that they can protect the environment, judge whether measures work, and keep the public informed (Art. 1(2)(a)). Three of its tasks bear directly on the Late Lessons project:

A Scientific Committee assists the Management Board and the Executive Director (Art. 10). LL1 presents itself as “an example of the information needed to help the EU and EEA member countries to frame and identify sound and effective policies” (LL1 p. 11). With their focus on early warnings and the costs of action and inaction, both reports fit tasks (i) and (j) closely.

1.2 Origins of the project and the 2001 volume#

Initiator. In his preface, EEA Executive Director Domingo Jiménez-Beltrán thanks “David Gee, who initiated this report” (LL1 p. 5). The 2013 volume says: “Originator of the Late lessons from early warnings project at the EEA was David Gee” (LL2 p. 5). Gee’s biography reads: a graduate in economics and politics, working “at the interface of science, economics, production and policy-making, within occupational and environmental health, since 1974, for trade unions and non-governmental organisations”; a former Director of Friends of the Earth (England, Wales and Northern Ireland); at the EEA from 1995 on “emerging issues and scientific liaison” (LL1 p. 196; LL2 p. 689). He was EEA editor and a chapter author for both volumes.

Editorial team and governance (LL1). The editorial team was:

The lessons were “distilled into twelve ‘late lessons’ by the editorial team, under the guidance of the EEA Scientific Committee” (LL1 p. 3). The acknowledgements name 23 authors, about 30 peer reviewers (Joel Tickner among them) and five Scientific Committee members (LL1 p. 6).

Method. Fourteen case studies were “chosen from a range of well-known hazards … where sufficient is now known about their impacts to enable conclusions to be drawn”. They are arranged by the date of the first early warning. Authors were asked to judge by “the spirit of the times” and not “the luxury of hindsight” (LL1 p. 11). Each author answered four questions (LL1 p. 11):

  1. When was the first credible scientific early warning?
  2. What actions or inactions followed, and when?
  3. What were the costs and benefits of those actions or inactions, and how were they distributed?
  4. What lessons follow for future decisions?

Authors worked pro bono and were asked to be brief (LL1 p. 12).

Stated purposes. LL1 hopes for “improved transatlantic agreement about future decisions” (LL1 p. 3); three chapters have a North American focus and three cover EU–US conflicts (LL1 pp. 11–12). It also ties itself to the 2001 White Paper on European Governance and to “democratising scientific expertise” (LL1 p. 16).

A gap the editors acknowledged. Every case in LL1 is a “false negative”. The editors invited industry representatives to propose “false positives” and examined Facts versus fears (Lieberman & Kwon 1998), but “no suitable examples emerged” (LL1 pp. 12–13).

Publication. Copyrighted 2001 (ISBN 92-9167-323-4); the EEA web page dates it 9 January 2002. Earthscan published a trade edition in 2002 (Harremoës et al., eds). The preface promised “a separate EEA publication” on the policy implications (LL1 p. 3); this pass did not find it.

1.3 The 2013 volume#

Why a second volume. LL2 gives four reasons (LL2 p. 9): 1. to extend the approach to further long-known hazards (lead in petrol, mercury, tobacco smoke, DDT) and to more recent ones (feminisation of fish, insecticides and honeybees); 2. to fill “an acknowledged gap” by analysing false positives; 3. to cover emerging issues: mobile phones, GMOs, nanotechnologies and invasive alien species; 4. to show how precaution can help with fast-changing, systemic challenges and support “sustainable innovations” with public participation.

Process. - Topics were chosen “on the basis of advice from the editor, in collaboration with the editorial team and an advisory board, members of the EEA Scientific Committee and the Collegium Ramazzini” (LL2 p. 9). - Authors were given “seven structuring questions”, which the Introduction does not print (LL2 p. 10). - Case studies were “peer-reviewed by recognised experts … within a set of editorial guidelines provided by the EEA” (LL2 p. 10). Unlike LL1, LL2 does not name its reviewers (LL2 p. 5). - The editorial team was Gee, Philippe Grandjean, Steffen Foss Hansen, Sybille van den Hove, MacGarvin, Jock Martin, Gitte Nielsen, David Quist and David Stanners. The advisory board included Joan Martinez Alier, Silvio Funtowicz, Jerome Ravetz and Nicolas de Sadeleer. Stirling and Wynne are thanked as “continuing sources of inspiration” (LL2 p. 5).

Engaging with disagreement. Chapter 2 (Hansen & Tickner) reviews 88 claimed false positives. It classes four as genuine regulatory false positives: US swine flu, saccharin, food irradiation and Southern leaf corn blight (LL2 p. 17). It notes that Cox (2007) disputed the method (LL2 p. 33). The honeybee chapter is followed by a dissent panel from Bayer CropScience (Panel 16.1, pp. 401–402) and a reply from the authors (Panel 16.2, pp. 403–406).

Structure. Twenty new case studies (Parts A–C); costs, justice and corporate behaviour (Part D, Chs 23–25); science and governance (Part E: Grandjean, Ch. 26; Gee, “More or less precaution?”, Ch. 27; an unsigned conclusion, Ch. 28). Annexes summarise LL1 (pp. 701–716) and update its cases (pp. 717–737).

Publication details. The EEA web page dates it 22 January 2013. The full report carries ISBN 978-92-9213-356-6 and doi:10.2800/73322 (summary: doi:10.2800/70069). The widely circulated file is labelled “(05-2013)”; it was not compared with the January release.

1.4 How the two volumes relate#

LL1 (2001) LL2 (2013)
Subtitle frame “the precautionary principle 1896–2000” “science, precaution, innovation”
Cases 14 historical false negatives 20 new cases, plus a false-positives review; “34 case studies” across both volumes (LL2 p. 644)
Author brief 4 questions (p. 11) 7 structuring questions (p. 10)
Synthesis 12 late lessons (pp. 168–194) Barriers to precaution, an EEA “working definition” of the precautionary principle, strengths of evidence, costs, corporate behaviour, justice, public engagement (Chs 23–28)
Theory of knowledge risk / uncertainty / ignorance (Box 16.1 p. 170; Table 17.1 p. 192) adds ambiguity, variability, indeterminacy (Table 27.1 p. 656) and a “knowledge-to-ignorance ratio” (pp. 654–655)
Peer review reviewers named (p. 6) reviewers not named; EEA editorial guidelines (pp. 5, 10)

LL2 reprints the twelve lessons as its baseline (LL2 p. 11; the source line misprints “1986–2000”). It says its working definition was “produced and refined” over “the last decade of discussions arising out of Volume 1” (LL2 p. 649). LL2 is a continuation that widens the frame, from judging past regulatory failures to arguing about how innovation should be governed. Small inconsistencies remain: LL1 pairs “uncertainty” with “precautionary prevention” and dates asbestos as a known “risk” from 1965 (LL1 p. 192); LL2 pairs uncertainty with “precaution” and dates asbestos from 1930 (LL2 p. 656).

1.5 Later EEA follow-ups#

2. Intellectual traditions#

2.1 Risk, uncertainty, ambiguity and ignorance#

Background. Distinguishing risk (known probabilities) from uncertainty (unknown probabilities) goes back to Knight (1921). Two later developments shaped the Late Lessons project.

How LL1 uses them. - The ESTO project “(see, for example, Stirling, 1999) provided the initial framing of this analysis” (LL1 p. 168). - Box 16.1 defines risk, uncertainty and ignorance (LL1 p. 170), and Lesson 1 is “Acknowledge and respond to ignorance, as well as uncertainty and risk” (LL1 p. 168). - A fuller account would add “complexity, indeterminacy, ambiguity, and the nature of disagreement (Wynne, 2001; Stirling, 1999)” (LL1 p. 184). - Wynne (1992) supports the claim that public concern centres on “unacknowledged ignorance” and on who controls the response to surprises (LL1 p. 185). The reference list wrongly gives it as “Vol. 6”; it is Vol. 2 (LL1 p. 191). - Table 17.1 maps risk to “prevention”, uncertainty to “precautionary prevention” and ignorance to “precaution” (LL1 p. 192).

How LL2 extends them. Table 27.1 adds ambiguity (answered by “participatory precaution”), variability and indeterminacy (LL2 p. 656). LL2 also argues that the principle matters most where the “knowledge-to-ignorance ratio” is low, “as with emerging technologies” (LL2 pp. 643, 654–655).

A telling link. The Commission’s 2000 Communication cites the same ESTO project, expecting it to “give a comprehensive description of scientific uncertainty” (COM(2000) 1, p. 14). The Communication and LL1 drew on the same Commission-funded work but reached different conclusions (see 2.6).

2.2 Post-normal science#

Funtowicz & Ravetz (1993) described “post-normal science” for problems where uncertainty about the system and the stakes of the decision are both high. In such cases, they argued, the quality of science must be checked by an “extended peer community” that includes people outside science.

How much the reports use it. - Directly, not much. In LL1 it appears in only one case study. The Great Lakes chapter (Gilbertson) says applying post-normal science to the Great Lakes Water Quality Agreement “has recently resulted in a singular legitimacy for multi-causal ecological statements” (LL1 p. 129). - Through people, more. In LL2, Funtowicz and Ravetz both sat on the advisory board (LL2 p. 5). Funtowicz co-wrote Chapter 19 on GM crops, which draws on his work on value judgements within science (LL2 p. 470). Ravetz’s “The post-normal science of precaution” (2004) is cited in Chapter 25 (LL2 pp. 610, 618). Funtowicz & Ravetz (1992) is cited in Chapter 27’s references (LL2 p. 664).

Where the family resemblance shows. Both volumes want lay and local knowledge brought into appraisal (Lesson 8, LL1 p. 169; LL2 p. 671), and both see science under policy pressure as bound up with values.

2.3 The Collingridge dilemma and lock-in#

Collingridge (1980) described a double bind in controlling technology. While a technology is young and easy to change, its effects cannot be foreseen well enough to justify control. By the time the effects are clear, the technology is so entrenched that changing it is slow, costly and hard.

Neither report cites Collingridge. A full-text search of both volumes found no mention of him. Both analyse the problem through the idea of lock-in: - LL1 draws on constructive technology assessment. It notes that technological systems tend to “‘lock-in’ to particular configurations at a relatively early stage”, “foreclosing other options”, sometimes “for arbitrary reasons” (LL1 pp. 186–187). It treats “robust, diverse and adaptable technologies” as insurance against surprises (Lesson 7; LL1 p. 187). - LL2 links “technological monopolies” (lead in petrol, asbestos, CFCs, PCBs) to “technological ‘lock-in’ but also to institutional and ideological lock-ins”, and to market prices that are artificially low (LL2 p. 673). - LL2 names a tendency to delay: “by the time evidence of harm is confirmed, the technology has often changed, leading to assumptions that … today’s technology is now safe”. Large sunk investments create “de facto technological lock-ins” (LL2 p. 672).

The EEA’s later flagship report does make the Collingridge link explicit (SOER 2020, p. 407).

2.4 Constructive, reflexive and anticipatory governance, and responsible innovation#

What LL1 draws on. Constructive technology assessment (CTA; Rip, Misa & Schot, cited as 1996), which brings in “innovators, regulators, users and other stakeholders” “from the outset” rather than as “an (often adverse) downstream reaction” (LL1 p. 186); technology options analysis (Ashford; Tickner) and alternatives assessment (LL1 p. 187); consensus conferences, scenario workshops and multi-criteria mapping (LL1 pp. 183, 187). Governance is framed as “a learning process in public policy, industry, science and civil society” (LL1 p. 187), and the choice of measures as “an essentially political business” (LL1 p. 183).

What LL2 adds. Stirling’s “opening up” versus “closing down” of appraisal (LL2 p. 644); Jasanoff on regulatory science (LL2 p. 645); an extract from René von Schomberg on responsible research and innovation (RRI; Box 27.5, LL2 p. 661) and a pointer to Owen’s work on responsible innovation (LL2 p. 662). Ch. 28 laments a world “more static in terms of political reflexivity and adaptations in governance” (LL2 p. 670).

Close relatives the reports do not cite. Neither volume uses the labels “anticipatory governance” or “reflexive governance”. The relatives are reflexive governance (Voß, Bauknecht & Kemp 2006); real-time technology assessment and anticipatory governance (Guston & Sarewitz 2002; Guston 2014), which grew from the same technology-assessment lineage as CTA; and the responsible-innovation framework of anticipation, reflexivity, inclusion and responsiveness (Stilgoe, Owen & Macnaghten 2013; Owen co-authored LL2 Ch. 13). The link is shared lineage and shared people, not direct borrowing.

2.5 Other traditions the reports draw on#

2.6 The European precautionary principle#

Origins. LL1 traces the principle to the German Vorsorgeprinzip of the 1970s. It had several elements: research and monitoring, a general reduction of environmental burdens, “clean production” and innovation, proportionality, cooperation among stakeholders, and action before “full ‘proof’” (LL1 p. 13, citing Boehmer-Christiansen 1994). LL1 notes its spread through marine agreements such as the Third North Sea Conference of 1990 (LL1 p. 14, Table 1.2).

Rio Declaration, Principle 15 (1992). “In order to protect the environment, the precautionary approach shall be widely applied by States according to their capabilities. Where there are threats of serious or irreversible damage, lack of full scientific certainty shall not be used as a reason for postponing cost-effective measures to prevent environmental degradation.” (A/CONF.151/26/Rev.1 (Vol. I), Annex I). LL2 misquotes this: its Box 27.1 has “serous or irreversible harm” and “costs effective” (LL2 p. 648). LL2 puts its own definition in the positive, as against Rio’s “triple negatives” (LL2 p. 649).

Treaty basis. - The Maastricht Treaty (1992) wrote the principle into the environment title (LL1 p. 14 quotes it). - It is now Article 191(2) TFEU (ex Art. 174 TEC): Union environmental policy “shall aim at a high level of protection … It shall be based on the precautionary principle and on the principles that preventive action should be taken, that environmental damage should as a priority be rectified at source and that the polluter should pay.” - Article 191(3) requires account of “the potential benefits and costs of action or lack of action”. The reports’ “pros and cons of action and inaction” echo this wording.

The 2000 Commission Communication (COM(2000) 1 final, 2 February 2000): - Aims. Among them, to “avoid unwarranted recourse to the precautionary principle, as a disguised form of protectionism” (p. 2). - Scope. The Treaty mentions the principle only for the environment, but “in practice, its scope is much wider”: it applies where there are “reasonable grounds for concern” that effects “may be inconsistent with the high level of protection chosen” (pp. 2, 9). - Place in risk analysis. It is “particularly relevant to the management of risk”, within a sequence of risk assessment, management and communication (pp. 2, 12). Judging acceptable risk is “an eminently political responsibility” (p. 3). - Guidelines. Measures should be proportional, non-discriminatory, consistent, based on an examination of costs and benefits that is “not simply an economic cost-benefit analysis”, subject to review, and able to assign responsibility for producing evidence. Reversing the burden of proof “cannot be made a general rule” (pp. 3–4).

The Nice resolution (December 2000). The Council “endorses” the broad lines of the Communication (para. 1). It adds that “civil society must be involved and special attention must be paid to consulting all interested parties as early as possible” (para. 15).

Case law and secondary law. The reports cite Case C-157/96 (the BSE case, 1998) (LL2 p. 649). LL2’s Box 27.1 also lists secondary laws that invoke the principle (LL2 pp. 648–649): - Directive 2001/18/EC on GMOs, Art. 4(1); - Regulation 178/2002 on food law, Art. 7; - REACH (Regulation 1907/2006); - Regulation 1107/2009 on plant protection products, Art. 1(4).

2.7 Where LL1 stands relative to these frameworks#

Where they agree. - LL1 calls the Communication and the Nice decision “the most significant support” for the principle in Europe (LL1 p. 13). - LL1 takes “reasonable grounds for concern” as one level of proof (Table 16.1, p. 184) and uses the same words in Lesson 12 (p. 169). - LL1 agrees that choosing measures is political (p. 183).

Where they differ. 1. Timing of participation. The Communication places precaution in risk management. LL1 argues that stakeholder involvement must “begin at the beginning rather than being artificially confined to the later ‘risk management’ stages”, because the stages “are not sequential” (LL1 p. 186). It cites the Nice resolution in support. 2. Kind of uncertainty. The Communication’s triggers are about gaps in data (pp. 13–14). LL1 extends precaution to ignorance, meaning surprises no assessment could have predicted (LL1 pp. 169–170, 192). 3. Who owns uncertainty. LL1 holds that scientific uncertainty is “not just a private matter for scientific bodies to autonomously resolve” (LL1 pp. 185–186). 4. Innovation. LL1 revives the German idea that precaution includes stimulating innovation (LL1 pp. 13, 186). The Communication worries about misuse of the principle as a trade barrier. 5. Level of proof. LL1 treats the choice of level of proof as “a key political decision with profound ethical implications” (LL1 p. 193). The Communication mainly treats this as scientific evaluation followed by a political decision.

LL2 keeps these positions and sharpens them in its working definition and its “precautionary and participatory framework for risk analysis” (LL2 p. 643).

3. How the reports have been used since#

3.1 EU institutions#

European Parliament research service. The EPRS in-depth analysis The precautionary principle: definitions, applications and governance (Bourguignon, December 2015) opens with the two EEA reports as evidence that lessons were “only learned belatedly” (p. 4). It reproduces the EEA working definition as one of three institutional positions, alongside the Commission and UNESCO (p. 10), adapts a figure of the EEA cases (p. 14), and sets the EEA’s claim that precaution can boost innovation against the European Risk Forum’s proposed “innovation principle” (p. 19).

Commission, DG Environment. Science for Environment Policy Future Brief 18 (September 2017) cites Late Lessons chapters on BSE and TBT (pp. 23–24). It discusses the tension with the “innovation principle” in Council conclusions of 27 May 2016 (Box 2, p. 18).

Commission evaluations. The 2019 Fitness Check of the water and floods directives, SWD(2019) 439, lists LL2 Chapter 15 in its bibliography: a light-touch citation, but in routine evaluation work.

EEA flagship and foresight reports. SOER 2020 and Drivers of change 2019 (see 1.5) carry the Late Lessons framing of early warnings, lock-in and precaution under ignorance into the Agency’s main reporting.

Research funding. Horizon 2020 topic SwafS-18-2018, “Taking stock of the application of the precautionary principle in R&I”, funded RECIPES (grant 824665, 2019–2022, Maastricht University) to “reconcile science, innovation and precaution”. The CORDIS record does not mention Late Lessons; it shows only that LL2’s question became an explicit EU research topic.

3.2 National policy#

The UK Government Chief Scientific Adviser’s Annual Report 2014, Innovation: Managing Risk, Not Avoiding It, cites both volumes in its evidence volume. Chapter 4, “Making choices in the face of uncertainty: strengthening innovation democracy”, is by Andy Stirling, one of the LL1 editors. This is an example of the reports entering national science-advice discussion through their own authors.

3.3 Courts#

No court citation verified. This pass did not confirm any judgment, Advocate General opinion or national court decision that cites either report. The EU case law the reports rely on (C-157/96, C-180/96, T-13/99 Pfizer, T-70/99 Alpharma) is earlier than, or parallel to, the reports. That case law is summarised with the reports in EPRS 2015 (p. 10). Treat any claim that courts have relied on Late lessons as unverified. A targeted full-text search of CURIA/EUR-Lex and national databases is still to do; see section 5.

3.4 Scholarship#

Uses of the framework. - Stirling’s widely cited EMBO Reports article on risk, precaution and science cites LL1 as a basis for precautionary appraisal (Stirling 2007). - Hansen, Maynard, Baun & Tickner (2008) applied the twelve lessons to an emerging technology. That paper became the basis of LL2 Chapter 22 (LL2 p. 530). - Gee’s own follow-ups (2003, 2006) applied the lessons to endocrine disruptors and to science under ignorance.

The false-positives exchange. This is the most direct scholarly argument over the reports’ evidence. - Hansen, Krayer von Krauss & Tickner (2007) sorted alleged false positives into categories and found few genuine ones. - Cox (2007) challenged how cases were classified; Hansen et al. replied (2007). - LL2 Chapter 2 rests on this work and points to the dispute (LL2 p. 33).

The wider debate the reports speak to. - LL1’s transatlantic aim entered a literature that questions whether “European precaution” and “American risk assessment” really differ. Wiener & Rogers (2002) and Wiener et al. (2011) found no across-the-board difference: relative stringency varies by risk. - Critics of the precautionary principle as a decision rule include Majone (2002), Sunstein (2005) and Löfstedt (2014). LL2 itself cites the European Risk Forum’s critique (ERF 2011; LL2 p. 664). - This pass did not check whether each of these authors discusses Late lessons specifically. They are cited here as the debate in which the reports take a side, not as reviewers of them.

4. Observations for later use (technology-neutral)#

  1. The conceptual framework comes from Commission-funded STS work. LL1 took its core distinctions from the ESTO project, which the Commission’s Communication also cites (LL1 p. 168; COM(2000) 1 p. 14). The disagreement between them is about how far precaution reaches (ignorance, framing, early participation), not a clash between science and anti-science.
  2. The evidence is expert testimony. Authors were chosen because they took part in the histories (LL1 p. 12; LL2 pp. 9–10). The case chapters are well informed and peer reviewed, but written from particular positions.
  3. The selection question is only partly answered. LL1 covers false negatives only (pp. 12–13). LL2’s answer (4 genuine false positives out of 88) depends on disputed classification (Cox 2007; LL2 p. 33).
  4. The Collingridge problem is present without the label. Lock-in and the timing problem run through both volumes (LL1 pp. 186–187; LL2 pp. 672–673). Later EEA work names Collingridge (SOER 2020 p. 407).
  5. The reports move from precaution toward innovation governance. They go from judging when to act on warnings to arguing about the direction, diversity and democratic control of innovation (LL1 pp. 186–187; LL2 pp. 644, 661–662, 673). LL2 takes up RRI (Box 27.5) but not the “anticipatory governance” literature.

5. Open items and what was not verified#

References#

The reports and EEA sources#

Intellectual sources#