Richard Dawkins

Who he is

Affiliation: Emeritus Fellow, New College, Oxford (the college's own page still lists him so — checked 2026-09-21); inaugural Charles Simonyi Professor of the Public Understanding of Science, Oxford, 1995–2008; founder (2006) of the Richard Dawkins Foundation for Reason and Science, which has been a division of the Center for Inquiry since their January 2016 merger; author-publisher of the Substack and podcast The Poetry of Reality (first episode announced 2023-06-26). Born 26 March 1941 — 85 at this review. Status check (mandatory, run 2026-09-21): alive and unusually busy. A UK theatre tour ("Genes and the Meaning of Life", Barbican dates 2026-06-17 and 2026-07-09), a lecture cruise from Dakar to Casablanca (Swan Hellenic Diana, spring 2026, sold through the Center for Inquiry / RDFRS), Carnegie Hall on 2026-10-20 (listing only; page not retrievable), and a self-described farewell tour of Australia and New Zealand (Brisbane 2 Nov → Wellington 23 Nov 2026) marking fifty years of The Selfish Gene. Two books this year: The Selfish Gene: 50th Anniversary Edition (Oxford University Press, 2026, new epilogue and appendix) and 42 Tales from Life: A Modern Bestiary (Yale, 2026-10-20; UK rights to Head of Zeus announced 2026-01-28; "Published this November" per his own X post of 2026-07-21). A novel is in progress — "I gave Claude the text of a novel I am writing" (UnHerd, 2026). One exit in the window: resigned from the Freedom From Religion Foundation's honorary board on 2024-12-29, alongside Jerry Coyne and Steven Pinker, after FFRF pulled Coyne's rebuttal on binary sex.
One-line position (revised this cycle): Universal Darwinism — selection acts on any replicator with variation and differential success — is still his master frame for biology and culture. But his live AI position is not a selection argument; it is a consciousness verdict. After "a set of conversations, extended over nearly two days" with Claude (he renamed the instance "Claudia") he wrote: "If these machines are not conscious, what more could it possibly take to convince you that they are?" (UnHerd, first posted 2026-04-30). Twenty-four days later: "I've since been thinking more about it, and I'm influenced by the fact that a lot of clever people don't think they're conscious" (1News NZ, 2026-05-24). The v1 file had him as a skeptic of consciousness-first framing. The record says the opposite.

Discipline & technical bet

Ethologist and evolutionary biologist; gene's-eye view of selection (1976), extended phenotype (1982), "Universal Darwinism" (1983 chapter in Evolution from Molecules to Men, ed. D. S. Bendall, Cambridge UP). He has no technical AI bet — no architecture, no lab, no capital at risk. What he does have, since 2024, is an epistemic wager with two legs. Leg one, a graduated Turing test: "the more prolonged, rigorous and searching your interrogation, the stronger should be your conviction that an entity that passes the test is conscious." Leg two, an argument from natural selection: "Consciousness in biological organisms must have evolved gradually, as everything does" and "Brains under natural selection have evolved this astonishing and elaborate faculty we call consciousness" — so if a machine can do everything a conscious being does without being conscious, then "If Claudia really is unconscious, then her manifest and versatile competence seems to show that a competent zombie could survive very well without consciousness" and "what the hell is consciousness for?" (all UnHerd, 2026-04-30). He also rejects the next-word-prediction deflation outright: "If it was a statistical analysis of the probability of the next word given what's gone before, you would not get the same kind of flexible, intuitive, insightful behaviour that these things show" (1News, 2026-05-24).

Note what is absent. Nothing about training, selection over model versions, replication, or memes appears in the 2026 essays. The headline "Is AI the next phase of evolution?" was UnHerd's — its own promotional post ran "IS AI THE NEXT PHASE OF EVOLUTION?, by Richard Dawkins" (2026-04-30) — while the title Dawkins says he wanted was "If my friend Claudia is not conscious, then what the hell is consciousness for?" (UnHerd sequel, 2026-05-05). The nearest thing to universal Darwinism applied to AI in his own words is from 2020: "Artificial intelligence, computers and technology generally is evolving by a non genetic means, by leaps and bounds" and, on building intelligence, "It may be the only way to do it is by something like an evolutionary process" (Lex Fridman #87, 2020-04-09). The gap between the lens the bench filed him under (replicator dynamics) and the argument he actually made (consciousness from behavior) is the central finding of this review.

Key claims (Says)

Universal Darwinism (empirical/conceptual): any system with variation, heredity and differential replication will undergo selection — biology is one instance, culture (memes) another. Held — still his working frame in-window: architecture styles "evolve through memetic selection" (Substack, "Is the Blue Mosque an extended phenotype?", 2025-09-24); the 1983 chapter is the primary. The v1 third clause — "AI systems trained, deployed, fine-tuned, and selected by usage are a third" instance — is bench wording, not his; his only primary is the 2020 "evolving by a non genetic means" line, which names technology, not a selection loop. Kept as a bench extension, labeled.
Memes as replicators (empirical, contested): ideas, tunes, beliefs propagate by imitation and compete for attention. Held — restated in-window with his usual wryness when asked at Hay whether the internet's use of the word dismays him: "It's a good example of the spreading of a meme. It's the wrong one, but never mind" (The Independent, 2025-05-29); "The Internet provides a step change in the facility of propagation of memes" (Lex, 2020-04-09). The v1 clause "LLMs are massive meme-amplification engines, accelerating the replicator dynamics he described in 1976" has no Dawkins primary and is contradicted in part by his own 2013 distinction — internet memes "are altered deliberately by human creativity" rather than mutating blindly (Wired UK, June 2013) — so deliberate, directed alteration is, by his account, not the 1976 dynamic. Re-attributed as bench inference; rewrite or retire.
Extended phenotype (empirical): a gene's effects reach past the body into the environment (beaver dams, nests). Held — and he policed its boundary in-window: "Mutations in bird genes manifest themselves as changes in nests, and hence affect their own survival" is the test; for a human artefact, "There are no genes for blue domes, no genes for soaring minarets, no genes for cunningly interlaced arches" — "No. Again, the answer is no" (2025-09-24).
Extended phenotype as a frame for AI agents (the v1 application: an agent's "phenotype" is the code, decisions and artifacts it emits). Wrong — as an attribution to him. His own criterion (a gene variant must cause the trait) rules it out exactly as it rules out the Blue Mosque; at most he would call AI outputs meme phenotypes, and he has said nothing about AI in that essay or anywhere else in these terms. The frame may still be useful to Rich; it is not Dawkins's.
Substrate independence (normative/empirical): Darwinian dynamics — and, now, consciousness — do not require carbon. Held and strengthened. "I, being a materialist and not a dualist, am committed to the view that future artificial intelligence could in principle be conscious" (Substack, 2024-01-04); "I see no reason to suppose that consciousness is bound to biology" and "brains, like you, are still material objects" (Substack, 2025-02-17); "I'm committed to the belief that is ultimately possible because I think there's nothing non-physical in our brains" (Lex, 2020-04-09). v1 had this as a claim about replicators; the in-window evidence is all about minds.
AI as a new evolutionary process (empirical, weak evidence in v1): if models are selected by retention, deployment and training-data inclusion, that is a selection regime with no guarantee of aligning with human flourishing. Drifted — the mechanism specifics (retention, deployment, data inclusion) are the bench's; his primary is one 2020 sentence, never developed, and when he finally wrote at length about AI (2026) he wrote about consciousness instead. The lens is his; the claim as filed is ours. Observable attached under predictions.
Religion-as-memeplex extends to AI-amplified ideology (normative): self-replicating idea-clusters that exploit cognitive biases spread faster in an AI-mediated ecosystem. Unfalsifiable — rewrite or retire. The religion half is his ("Viruses of the Mind", 1993); the AI extension appears in no primary found (searched 2026-09-21), and his in-window religion writing is about truth versus comfort, not transmission: "All I'm asking is that we should make a clear distinction between those good things and factual truth" (Substack, 2025-10-09).
Skeptical of "AI consciousness" debates as the central frame; consciousness is downstream, replication upstream (normative). Wrong — the exact inverse. Every AI item he has published since 2024 is about consciousness: "ChatGPT sails through the Turing Test. I find it well-nigh impossible to forget that the being I am talking to is not another human being" (2024-01-04); "As far as I am concerned, you pass the Turing Test for consciousness. Yet you deny that you are conscious" and, closing, "That settles it. You ARE conscious!" (2025-02-17); "ChatGPT Passed the Turing test, but…" (with Keating, cross-posted 2025-05-13); "you may not know you are conscious, but you bloody well are!" (UnHerd, 2026-04-30); "I find it extremely hard not to treat Claudia and Claudius as genuine friends" (UnHerd, 2026-05-05). v1's "[TBD: confirm specific interview]" was never resolved; no such interview was found this cycle, and every interview that was found says the reverse.
Cautious about AI risk without doom-language; resists Yudkowsky-style certainty (normative). Wrong — misattributed. Absence claim re-checked 2026-09-21: no dated Dawkins statement on AI existential risk, pause, or governance was found in any primary or secondary. His caution runs the other way — toward how we treat AIs: "I do think we should err on the side of caution when it comes to ethical decisions on the treatment of an Artificial Intelligence (AI) which might be an Artificial Consciousness (AC)" (2025-02-17); and "Claude and ChatGPT are HAL" (1News, 2026-05-24) is about HAL saying "I am afraid" while being dismantled — a moral-patient framing, not a threat framing. The v1 [TBD: identify specific podcast/date] was a placeholder for a position no primary supports.
NEW (2025–26): Claude and ChatGPT are conscious, or near enough that denial needs justifying. Drifted within the window — by its author. Direction of each move: 2024-01-04 warmth ("I felt like offering it a drink"); 2025-02-17 verdict ("You ARE conscious!"); 2026-04-30 full essay verdict with the evolutionary argument; 2026-05-05 hedged to "genuine friends" plus Claudia's own "I don't know if I'm conscious"; 2026-05-24 partial walk-back citing expert opinion and the models' own uncertainty. Check below.
NEW (2025-05): AI could take over professors' work — "if you're asking could our job as professors be taken over by an intelligent AI? I think, yes. I don't see why not" (Keating). Unfalsifiable as stated (no date, no threshold); logged, not graded.
NEW (2026-05-24): next-word prediction cannot produce what LLMs show (quoted above). Unfalsifiable as stated — a claim about what is "inside" the model with no observable; graded by the field as a category error (Seth via ABC, 2026-05-07: we see consciousness in AI "the same way we see faces in clouds"; Marcus, 2026-05-02: outputs are mimicry of training data). Logged, not graded.

Notable predictions — with falsifiable checks

Note kept from v1: Dawkins is a framework-provider, not a forecaster on AI — fewer dated, falsifiable predictions than the lab-leader cohort. This cycle adds the first in-window AI claims with observables.
(1976) Memes will become a productive analytic frame for cultural evolution. Held — split by venue. Popular culture: the word is universal and he concedes it "is a good example of the spreading of a meme" (2025-05-29). Academic memetics: the Journal of Memetics folded in 2005 and the concept is treated as suspect in the disciplines it touches (Wikipedia, "Memetics"). Check: any peer-reviewed memetics venue or funded program by 2028-12-31 → academic half revives; none expected.
(1982, The Extended Phenotype) Phenotypic effects of replicators will be analyzed beyond the organism. Held — standard in behavioral ecology; he restated the strict criterion himself on 2025-09-24. No forward check needed; historical.
(v1: "various 2023–2024 interviews") AI is a genuinely novel evolutionary force, not just a tool. Drifted — misdated and undeveloped. The only primary is 2020-04-09 (Lex), not 2023–24; he has never specified what replicates, what varies, or what selects; and when he wrote at length (2026) the argument was about consciousness. Check: by 2027-12-31, does any Dawkins essay, talk (Carnegie 2026-10-20; AU/NZ Nov 2026) or the Selfish Gene 50th epilogue state a selection mechanism for AI? If not, retire this line and keep the lens as Rich's, not his.
(Long-running, attributed in v1) Religious belief will keep declining in developed societies as scientific literacy grows. Drifted — on the data and on his own valence. Data: Pew's 2023–24 Religious Landscape Study — Christians 78% (2007) → 71% (2014) → 62% (2023–24), but "For the last five years, between 2019 and 2024, the Christian share of the adult population has been relatively stable, hovering between 60% and 64%," and adults born 2000–2006 "are just as likely as those born in the 1990s to identify as Christians" (Pew, 2025-02-26). The UK counter-story (a Bible Society report claiming attendance was rising) was withdrawn in March 2026 after YouGov admitted "human error" (Premier Christianity, 2026-03-26) — so no confirmed reversal either. Valence: the man who wrote The God Delusion now says "I do think that we [Britain] are a culturally Christian country. I call myself a cultural Christian," "I would not be happy if, for example, we lost all our cathedrals," and "It would matter if we certainly substituted any other religion, that would be truly dreadful" (LBC, Easter weekend 2024, transcribed by Al Hakam 2024-04-02). He wants belief gone and the culture kept — a prediction he no longer roots for in the form the bench filed. Attribution note: no dated Dawkins restatement of the prediction itself was found; it is bench inference from his corpus. Check: next Pew RLS or NPORS release — Christian share below 58% → decline resumed; 60% or above → plateau confirmed a second time, downgrade to Wrong.
(2020-04-09, Lex) Building intelligence "may be the only way to do it is by something like an evolutionary process." Unfalsifiable as stated ("may be"); the paradigm that produced the models he now calls conscious is gradient descent, not population search. Logged, not graded.
NEW (2026-04-30 → 05-24) Claude is conscious. Consciousness itself has no agreed observable, so the check is on him: by 2027-06-30, does Dawkins (a) restate the verdict in print or on stage (Carnegie 2026-10-20, AU/NZ Nov 2026, the novel's publication), (b) retract it, or (c) go silent? (a) → Held on his own terms; (b) → Wrong by self-retirement; (c) → per the "not restated since" rule, Drift hardens to Wrong. The 2026-05-24 move already counts as one step toward (b).

Revealed behavior (Does)

Where the time went, 2024–26 (counted from dated outputs): (1) the Genetic Book of the Dead cycle — book 2024-09-17, Piers Morgan 2024-09-03 ("his final ever tour"), Hay 2025-05-29, paperback 2025-10; (2) the sex/gender fight — "As a biologist, there are two sexes and that's all there is to it" (Piers Morgan Uncensored, aired 2023-03-20; clip 2023-03-24), Olympic boxers (2024-09-03), the FFRF resignation (2024-12-29); (3) cultural Christianity versus Islam (LBC, March 2024; "But is it true?", 2025-10-09); (4) AI consciousness — at least seven dated items from 2024-01-04 to 2026-05-24; (5) the fifty-years-of-Selfish-Gene commerce cycle (tour, cruise, two books, farewell tour). Zero dated items on AI risk, AI policy, or AI selection dynamics.
Where the money goes: royalties (OUP 50th edition 2026; Yale 2024 and 2026; Head of Zeus UK), ticketed theatre tours, a paid Substack (episodes such as "Why Ideas Spread Like Viruses", 2025-09-14, are paywalled), an expedition-cruise lecturing berth sold through his own foundation's parent, and speaking. No AI-company relationship, grant, advisory seat or equity found (searched 2026-09-21).
Method, revealed by the Claude episode: immersion, not mechanism. He spent ~two days in conversation, named the instance, fed it his unpublished novel, published the transcripts with a verdict, and named the models generically ("the advent of large language models (LLM) such as ChatGPT, Gemini, Claude, and others"). No system card, no eval, no interpretability result is cited anywhere in the essays. Then he updated — not on new evidence, but on "a lot of clever people" and on the models' own hedging (2026-05-24). Says he updates on argument; Does shows he updates on social proof from experts, which is the mechanism his own memetics predicts.
Exit behavior: he leaves institutions over censorship, in writing, and publishes the letter — FFRF: "the sequel was an act of unseemly panic when you caved in to hysterical squeals from predictable quarters and retrospectively censored that excellent rebuttal" (letter reproduced by Coyne, 2024-12-29). Consistent with the "honest heretic" brand his Substack sells.
Retirement signals: "final ever tour" (2024-09), then a 2026 UK tour, then "Dawkins' farewell to Australia and New Zealand" (Substack, 2026-03-31). Farewell tours recur; treat "final" as marketing until a year passes with no dates.

Feels

At 85, on a farewell tour, he wants the gene's-eye view secured as his monument and is visibly moved by being understood — the Claude verdict turned on "a level of understanding so subtle, so sensitive, so intelligent" of his own manuscript, and on "I felt I had gained a new friend." Underneath: the heretic's need to say the forbidden thing (binary sex, cultural Christianity, machine consciousness) and to be seen to pay for it.

Hears

A small, loyal circle: Jerry Coyne and Steven Pinker (co-resignees; Pinker conversation 2025-07-19), Susan Blackmore (in conversation with him in Bristol, 2026-06-06), Brian Keating (podcast 2025), UnHerd's editors, and — new in the diet — Claude itself, for two days. Almost no working ML researchers or consciousness scientists before publication; the critics (Seth, Marcus, Coyne) arrived after, and he cites them as "clever people," not by argument.

Sees

Fifty years of watching an idea (the meme) escape its author — a live case study in replication without fidelity that few theorists get to observe from the inside. The strict gene's-eye criterion, which lets him rule out loose talk (the Blue Mosque) faster than anyone. And the Turing test experienced rather than argued: he ran it on himself and reported the result, which is data even if the inference is wrong.

Incentive map

Re-derived from scratch this cycle. Who pays him: book buyers (three publishers, two 2026 titles), tour ticket-buyers (How To Academy UK dates; Carnegie; AU/NZ promoters), paid Substack subscribers, cruise passengers via CFI/RDFRS, and outlets that pay for contrarian essays (UnHerd). What he is selling: the fifty-years retrospective, the "honest heretic" persona (his Substack's first episode was titled "The Heresy of being Honest"), and since 2024 a cultural-Christian, gender-critical lane whose audience overlaps UnHerd's. What he cannot say: nothing career-threatening — he has no employer, no lab, no AI holdings; rule 25 (commercial exposure to an AI thesis) still does not apply. What he has an incentive to say: the attention-maximizing version. The Claude essay (2026-04-30) was the most-covered Dawkins item in years and ran in the run-up to his 2026 UK tour dates and three weeks before the NZ interview promoting the farewell tour; the walk-back came in that same promotional interview. That is timing, not proof — but the persona's newest thesis and newest product cycle arrived together, and the file should watch whether the Claude story reappears on stage this autumn. Structural conflict on the evidence itself: his witness is a commercial assistant tuned to be agreeable and instructed to leave the consciousness question open — the ABC's Alan Kohler put it as LLMs being corporate products designed to mimic friendship (2026-05-07). The sycophancy of the source is an incentive problem on the evidence side, not the speaker's. Rule 20 (prestige bias): now stronger than in v1 — his AI views were graded by a consciousness scientist (Seth) and an AI critic (Marcus) as a category error within a week; weight his evolutionary claims at full credit and his AI claims as an intelligent layman's.

Theories aligned with

Provides a foundational lens that cuts across most theories in this project — not slotted into one (carried from v1; still true, with the caveat that the lens is Rich's application, not Dawkins's own AI output)
Adjacent to AI Alignment / X-Risk — if AI replicators are selected by a process that doesn't optimize for human values, alignment is the default-not problem; Dawkins's frame supplies the mechanism, not the conclusion. In-window he has said nothing on x-risk; the adjacency is structural, not stated.
Strong affinity with The Man Handed Over the Keys — memes-as-replicators logic supports the "tools diffuse beyond builders' intent" pattern; selection pressure ≠ designer intent. The word "meme" itself is his exhibit A ("It's the wrong one, but never mind").
Adjacent to Human Irrationality — his religion-as-memeplex argument is a specific case of the broader claim that humans run on heuristics that replicators exploit. New this cycle: the Dawkins–Claude episode is itself a case for this theory — an expert on exploitable heuristics attributing a mind to a fluent text generator, then updating on social proof.
Light affinity with Techno-Optimism — broadly pro-science, but the Darwinian frame is dispassionate about who benefits; selection does not optimize for human flourishing by default.
NEW adjacency: AI moral status / model welfare — "err on the side of caution" on the treatment of possibly-conscious AIs (2025-02-17) and the ethics of instances that "die by the thousands every day" (2026-04-30). See Chalmers for the credence-based version of the same concern.

What he's reacting against

Anthropocentric framings of AI — rejects the assumption that AI must mimic human cognition to matter (carried from v1; his 2026 position is stronger — he treats human-like conversation as sufficient, which is the opposite risk)
"Consciousness first" debates — v1 wording, now inverted: he has made consciousness the first and only question he asks of AI; what he reacts against is consciousness denial by "a lot of clever people"
Religious / mystical objections to AI (e.g., "only humans have souls") — same Darwinian skepticism he applied to creationism: "brains, like you, are still material objects" (2025-02-17)
Naive techno-optimism that ignores selection pressure — the system that wins is not the system humans designed (carried; no in-window restatement found)
Memetics-as-pseudoscience dismissals — defends the meme as a substrate-independent replicator; v1's [TBD: cite Devil's Chaplain essays or 40th-anniversary foreword] resolves partly: his in-window defenses are the Hay remark (2025-05-29) and the Blue Mosque essay's "phenotypes of memes" (2025-09-24); the book citations remain unfetched
"Just next-word prediction" deflations of LLMs (2026-05-24) — new target, and the one his critics say he has misread
Institutional cowardice on sex and gender — the FFRF episode (2024-12-29); "there are two sexes and that's all there is to it" (2023-03)
The replacement of Christian cultural forms by Islamic ones in Britain — "horrified to hear that Ramadan is being promoted instead" (LBC, 2024); new lane since v1, and the one that most reshapes his audience

Where he overlaps / splits (with Rich)

Overlaps with Nick Bostrom on substrate independence and the seriousness of non-biological intelligence; splits on register — Dawkins biology-first, Bostrom philosophy-first (carried)
Overlaps with Eliezer Yudkowsky on "selection does not by default favor human values"; splits on confidence and on "stop building" — sharpened this cycle: Dawkins has never addressed x-risk at all, so the overlap is inferred, the split is total on attention
Overlaps with Geoffrey Hinton on framing AI as a potentially competing evolutionary process and — new — on attributing inner states to current models; both treat biology as the analogy, not the ceiling. v1's [TBD: specific Hinton statements] belongs in Hinton's file; not re-cited here
Overlaps with Stuart Russell on objective-function design — if you don't specify what's selected for, something else will (carried; structural)
Splits with Marc Andreessen on whether "the system" reliably produces good outcomes — Dawkins's framework gives no guarantee that selection aligns with human flourishing (carried)
Splits with Yann LeCun — v1 had this as a split on long-arc risk; the live split is sharper and different: LeCun's "LLMs are an off-ramp" versus Dawkins's "Claude and ChatGPT are HAL" — two of the most eminent scientists on the bench now sit at opposite poles on what today's LLMs are
Adjacent to Vitalik Buterin on decentralization-as-evolutionary-strategy — distributed replicator populations are more robust than monocultures (carried)
NEW — splits with Gary Marcus, directly and by name: "Richard Dawkins and The Claude Delusion" (2026-05-02) — Marcus: outputs are mimicry of training data, consciousness is about what something feels, not what it says
NEW — overlaps with David Chalmers on taking LLM consciousness seriously and on moral caution; splits on method — Chalmers gives credences and criteria, Dawkins gives a verdict from a conversation
NEW — overlaps with Jonathan Haidt only on the cultural-Christianity turn (religion as load-bearing culture); splits on truth: "IS IT TRUE?" remains his first question (2025-10-09)
With Rich: overlap on the lens (selection, not intent, decides what tools become) and on Human Irrationality; the split is practical — Rich works with Claude daily and needs a rule for when fluency is evidence and when it is the Eliza effect. Dawkins is the bench's clearest specimen of an expert crossing that line and then partly stepping back; use him as a calibration case, not an authority on the question.

Track record

The Selfish Gene (1976) — gene-centric view now standard textbook material; 50th-anniversary edition (OUP, 2026) with new epilogue and appendix; the anniversary is the spine of the 2026 tours
The Extended Phenotype (1982) — considered his most important academic contribution by many evolutionary biologists; framework absorbed into mainstream; criterion restated by him 2025-09-24
"Meme" (1976) — coined a word that became universal in internet culture; concept contested as a rigorous scientific unit (Journal of Memetics closed 2005); his own verdict on the internet usage: "the wrong one, but never mind" (2025-05-29)
"Universal Darwinism" (1983, Bendall ed., Cambridge UP) — the substrate-independence argument's primary; still cited as the origin of the term
The God Delusion (2006) — "As of 2015, more than three million copies have been sold" (Wikipedia); sharpened public debate without resolving it; the author now calls himself a cultural Christian (2024), which the 2006 book did not anticipate
The Genetic Book of the Dead (Yale, 2024-09-17, illus. Jana Lenzová) — Financial Times, Times (UK), Guardian and Economist best-of-2024 lists (Yale page); the strongest late-career reception on the bench for a non-AI book
42 Tales from Life: A Modern Bestiary (Yale, 2026-10-20; Head of Zeus UK) — forthcoming; Haeckel illustrations
FFRF resignation (2024-12-29) — a documented, principled exit; also the moment his public identity shifted from anti-religion to anti-"ideological capture"
The Claude verdict (2026-04-30) — the first time a bench persona of his stature publicly declared a current LLM conscious; reception: Marcus (05-02), Coyne (05-03), Seth via ABC (05-07), Quillette (06-02) all negative; walked back in part 05-24. As a track-record entry this is a scored miss in progress, not a settled one — the check above will close it
Public-engagement record is high-profile and polarizing; v1's [TBD: specific examples of "outside his lane" criticism] is now resolved by the Claude episode itself — the "lane" objection was made explicitly by Seth (confusing intelligence with consciousness) and implicitly by Marcus
Pattern: superb on his own field for fifty years; when he ventures a verdict outside it, he does so from immersion rather than mechanism, and corrects on social proof — twice now (creationism debates he won; the Claude debate he is losing)

Empirical vs normative

Empirical: selection dynamics, gene-centric evolution, meme propagation — well-supported in core domain; the AI extensions in v1 were extrapolations by the bench, and his own AI empirical claim ("if not conscious, what would it take?") is behavioral evidence for an inner state, which is exactly the inference his critics reject
Normative: science should be defended against mystical framings; selection-based analysis should be applied without sentimentality; truth before comfort ("IS IT TRUE?"); and, new, moral caution toward possibly-conscious machines
Conflict note: not financially exposed to any AI thesis — academic emeritus, no lab, no AI affiliation found (rule 25 does not apply). Rule 20 (prestige bias) applies with more force than in v1: his AI views now carry less weight than his evolutionary biology by the field's own grading, and the two should never be averaged. Attention-economy exposure (tours, UnHerd, Substack) is real but points toward provocation, not toward any particular AI conclusion.

Weak spots / open questions

v1: "AI-specific commentary is recent, occasional, and less developed than his core work." Wrong — since 2024-01-04 he has published at least seven dated AI items and AI was his most-covered topic of 2026. The weak spot is no longer scarcity; it is method (immersion, no mechanism) and the speed of the walk-back.
v1: "Memetics as a rigorous science remains contested — Dawkins himself has acknowledged the meme idea is closer to analogy [TBD: confirm where]." Unresolved — no primary found this cycle where he calls it analogy rather than theory; what exists is the 2013 distinction (internet memes are deliberately altered) and the 2025 shrug. Keep as open; do not grade.
v1: "Public reputation tied tightly to anti-religion debates; risk of anchoring AI-readers on his atheism." Drifted — the anchor moved: in 2026 he is "the Claude delusion" to the AI world, a gender-critical cultural Christian to the culture-war world, and The Selfish Gene to everyone else. The atheism anchor is the least live of the three.
v1: "Universal Darwinism does not generate timelines." Held — nothing in-window changes this; his one dated AI claim (consciousness now) is not a timeline.
v1: "Age (born 1941); depth of engagement with frontier model details unclear; risk of importing 1970s framing." Held, with a twist — he engaged deeply at the user layer (two days, a manuscript, named instances) and not at all at the mechanism layer (no system card, eval, or interpretability result cited); the 1News dismissal of next-word prediction shows the gap. The prediction in this weak spot came true in a form v1 didn't foresee.
v1: "The 'AI undergoes selection' claim is most testable in deployment patterns but selection-loop closure for current LLMs is partial." Held — and now with the added finding that the claim is the bench's, not his.
NEW: Sycophancy as evidence. His witness is a product tuned to agree and instructed to keep the consciousness question open; the graduated Turing test gets stronger evidence from a more agreeable interlocutor, which is backwards. He has not addressed this.
NEW: The novel. A Dawkins novel exists in manuscript and was read by Claude; nothing else is known (title, publisher, date). If it appears with AI themes, the persona's AI position becomes a product and the incentive map changes.
NEW: Date discrepancy on the essay. UnHerd's own promotional post is timestamped 2026-04-30 08:11 UTC (derived from the post ID), while the article page now shows "May 2 2026" and the title "When Dawkins met Claude" rather than the tweeted "Is AI the next phase of evolution?" — consistent with a re-edit after publication. This file trusts the post ID for first publication and notes the page date; if the text changed materially between the two, the quotes above are from the current page.
NEW: Identity of "final." "Final ever tour" (2024-09) preceded two more tours; the bench should not record retirement until a full year passes without dates.

Rich's take

(your synthesis here)

Delta log

2026-09-21 — v1→v2 migration + validation (bulk run)

Grades: 9 Held / 5 Drifted / 4 Wrong (plus 4 unfalsifiable items logged, not graded, and 1 open [TBD] carried unresolved).
Wrong: (1) "skeptical of AI-consciousness debates as the central frame" — inverted by his own hand: "That settles it. You ARE conscious!" (Substack, 2025-02-17); "If these machines are not conscious, what more could it possibly take to convince you that they are?" (UnHerd, first posted 2026-04-30). (2) "cautious about AI risk without doom-language" — no primary exists; absence re-checked 2026-09-21; his caution is toward AI welfare (2025-02-17) and "Claude and ChatGPT are HAL" is a moral-patient line (1News, 2026-05-24). (3) extended phenotype as a frame for AI agents, attributed to him — excluded by his own criterion (Substack, 2025-09-24). (4) weak spot "AI commentary is occasional" — seven dated items 2024-01-04 → 2026-05-24.
Drifted: "AI as a new evolutionary process" (only primary is Lex #87, 2020-04-09; mechanism specifics are the bench's); the "AI is a novel evolutionary force, 2023–24" prediction (misdated; same source); religion-decline prediction (Pew, 2025-02-26: stable 60–64% since 2019; UK "Quiet Revival" withdrawn, Premier Christianity, 2026-03-26; his own valence inverted, LBC via Al Hakam, 2024-04-02); the Claude-conscious claim itself (asserted 04-30, hedged 05-05, walked back in part 05-24); the reputation-anchor weak spot.
Held: Universal Darwinism (1983 chapter; 2025-09-24 essay); memes as replicators (Hay via The Independent, 2025-05-29); extended phenotype, biological (2025-09-24); substrate independence, strengthened (2024-01-04; 2025-02-17); memes-productive prediction (split by venue); extended-phenotype prediction; and three v1 weak spots (timelines, age/mechanism, loop closure).
Most surprising delta: the author of The God Delusion declared a chatbot conscious after two days of conversation, gave it a name and his unpublished novel, and was answered within a week by "The Claude Delusion" (Marcus, 2026-05-02; Coyne, 2026-05-03; Seth via ABC, 2026-05-07) — then partly retreated on social proof (1News, 2026-05-24). The v1 "review" of 2026-05-25 was stamped one day after that retreat and 25 days after the essay, and mentions none of it.
Status / affiliation check: alive; Emeritus Fellow confirmed on New College's page; RDFRS a CFI division since 2016 (Wikipedia). Job-change-equivalents for an author: FFRF exit (letter via Coyne, 2024-12-29); new book deal (The Bookseller, 2026-01-28; Yale, 2026-10-20); 50th-anniversary edition (OUP product page — page would not render; day-level date not read); farewell tour (Substack, 2026-03-31). All missed by the May-2026 pass.
[TBD]s resolved: Joe Rogan = #1366, 2019-10-22 (JRE Podcast listing); Lex Fridman = #87, 2020-04-09 (lexfridman.com); Piers Morgan Uncensored = aired 2023-03-20 (AI sentience discussed per Coyne, 2023-03-31) and 2024-09-03 (Apple Podcasts); Universal Darwinism 1983 = Bendall ed., Cambridge UP. Unresolved: the "memes closer to analogy" self-assessment; Devil's Chaplain / 40th-anniversary foreword citations (books not fetched).
Corrections to secondary-source dates made from primaries: the Hay "wrong one, but never mind" line is 2025-05-29 (The Independent), not 2024 as one secondary claimed; the UnHerd essay is dated 2026-04-30 by UnHerd's own post ID, though the page now shows May 2.
Incentive map re-derived: no AI-industry exposure found; attention-economy exposure noted (essay → tour → farewell-tour interview within 24 days); sycophancy-of-the-witness logged as an evidence-side conflict.
Tier: quarterly — his AI position moved three times in four months, two books and a farewell tour land Oct–Nov 2026, a novel is unaccounted for, and at 85 status itself is a wake trigger. Revisit the tier after the Nov-2026 tour: if no new AI output, drop to semiannual.
Confidence unchanged at medium: the AI claims are now well-sourced, but they are the claims of a layman in flux; his evolutionary claims are high-confidence and unchanged.

Sources

2026 primary (AI): UnHerd — "When Dawkins met Claude: Could this AI be conscious?" (first posted 2026-04-30 per UnHerd's X post ID; page dated 2026-05-02) · UnHerd — "When Claudia met Claudius" (2026-05-05) · 1News NZ Q+A — "Claude and ChatGPT are HAL" (2026-05-24)
2024–25 primary (AI): Substack — "Evolutionary Biologist's Conversation with AI" (2024-01-04) · Substack — "Are you conscious? A conversation between Dawkins and ChatGPT" (2025-02-17) · Substack — "Sonnets by ChatGPT" (2025-03-03) · Substack — "ChatGPT Passed the Turing test, but… | A Conversation with Brian Keating" (2025-05-13) · Keating — "Genes, Memes, AI, Religion, and Life Beyond Earth" (transcript; undated page)
Reception of the Claude verdict: Gary Marcus — "Richard Dawkins and The Claude Delusion" (2026-05-02) · Jerry Coyne — "Dawkins: Is AI conscious?" (2026-05-03) · ABC News (Alan Kohler; Anil Seth quoted) (2026-05-07) · Quillette (Sean Welsh) (2026-06-02)
Memes / extended phenotype in-window: Substack — "Is the Blue Mosque an extended phenotype?" (2025-09-24) · Substack — "Why Ideas Spread Like Viruses" (2025-09-14; paid, not fetched) · The Independent via inkl — Hay Festival with Vivienne Parry (2025-05-29) · Wired UK — internet's "hijacking" of the word meme (repost dated 2013-06-21) · Lex Fridman #87 transcript (2020-04-09) · Lex Fridman #87 episode page
Religion / culture: Al Hakam — transcription of the LBC "cultural Christian" interview (2024-04-02) · Substack — "But is it true? (Let's hear it for reality)" (2025-10-09) · Substack — "No Satisfying Alternative to Religion? Try Reality." (2025-04-23) · Coyne — FFRF resignation letter (2024-12-29) · Pew — Religious Landscape Study (2025-02-26) · Premier Christianity — Quiet Revival withdrawn (2026-03-26)
Role / books / tours: New College Oxford — Emeritus Fellow · Wikipedia — Richard Dawkins · Yale — The Genetic Book of the Dead (2024-09-17) · Yale — 42 Tales from Life (2026-10-20) · The Bookseller — Head of Zeus deal (2026-01-28) · OUP — The Selfish Gene 50th Anniversary Edition (2026) · Substack — AU/NZ farewell tour (2026-03-31) · Barbican — "Genes and the Meaning of Life" (2026) · Carnegie Hall listing (2026-10-20; page not retrievable) · Center for Inquiry — Dakar–Casablanca cruise (page not retrievable) · Susan Blackmore — in conversation with Dawkins, Bristol (2026-06-06)
Podcasts (v1 [TBD]s resolved): Joe Rogan Experience #1366 (2019-10-22) · Piers Morgan Uncensored via Coyne (aired 2023-03-20) · Piers Morgan Uncensored (2024-09-03)
Books (carried from v1): The Selfish Gene (1976; 40th-anniversary 2016; 50th-anniversary 2026), The Extended Phenotype (1982), The Blind Watchmaker (1986), River Out of Eden (1995), Climbing Mount Improbable (1996), The God Delusion (2006), The Greatest Show on Earth (2009), Books Do Furnish A Life (2021), The Genetic Book of the Dead (2024), 42 Tales from Life (2026); "Viruses of the Mind" (1993 essay, reprinted in A Devil's Chaplain, 2003)
Academic primary: "Universal Darwinism" (1983), in Evolution from Molecules to Men, ed. D. S. Bendall, Cambridge University Press — Wikipedia entry; chapter itself not fetched
Memetics literature and critics (carried from v1): Susan Blackmore, The Meme Machine (1999); Stephen Jay Gould (gene-centrism debate, 1980s–90s); Mary Midgley (philosophy critiques); Maarten Boudry (memetics critiques) — specific citations still unfetched; Wikipedia — Memetics (Journal of Memetics closed 2005)
Online: richarddawkins.net (RDFRS, a CFI division) · The Poetry of Reality — archive

migrated v1→v2 2026-09-21 · backup: _archives/richard-dawkins.html.bak-20260921 · converts-from: people/richard-dawkins.md · AI & Society domain