Beautiful Particulars — David Weinberger at Harvard Book Store
Table of Contents
1. Logistics
link was the harvard.com homepage; event page is /event/weinberger26. Free, no registration, no livestream mentioned. Conversation partner Martin Wattenberg was missing.
- Event
- Harvard Book Store author talk — David Weinberger, Beautiful Particulars
- In conversation with
- Martin Wattenberg (Google DeepMind; former Harvard professor)
- When
- Tuesday, 2026-11-10, 19:00–20:00 ET
- Where
- Harvard Book Store, 1256 Massachusetts Ave, Cambridge, MA 02138
- Admission
- free; signed copies available to pre-order
- Link
- Event page
2. Thesis
Machine learning works through particulars, not generalizations. The listing's claim is that this is already displacing laws and principles as the default shape of explanation, for morality, free will, creativity, causality and knowledge. Read from here, that is a claim about epistemics with a testable edge: a model that preserves "tiny truths" is also a model with no contract, and verification work depends on contracts.
The thesis line ("AI works through particulars, not generalizations") is danah boyd's blurb, not the author's.
3. Speakers people
3.1. David Weinberger
Library Innovation Lab: co-director (2010–14), not director. Franklin Fellow dates disputed (2009–11 vs 2010–12). Google ~4 yrs is two stints; PAIR portion 2018-06–2020-06. PhD year needs-citation.
- Researcher, Harvard metaLAB. Long-time affiliate of Berkman Klein (Fellow from 2004, later senior researcher).
- PhD in philosophy, University of Toronto (year not found). Taught philosophy at Stockton State College, 1980–86.
- Gag writer for the comic strip "Inside Woody Allen", 1976–83.
- Co-director, Harvard Library Innovation Lab, 2010–14.
- Journalism fellow at the Shorenstein Center, spring 2015.
- Franklin Fellow at the U.S. State Department, working on e-diplomacy. His own bio gives both 2009–11 and 2010–12.
- Writer-in-residence in Google AI groups for about four years over two stints. The PAIR stint was 2018-06 to 2020-06.
- Editor of the open-access Strong Ideas series at MIT Press since 2019.
3.1.1. Books
"six books" holds counting Cluetrain (co-authored) and Beautiful Particulars; World of Ends and New Clues are essays.
| Year | Title | Note |
|---|---|---|
| 2000 | The Cluetrain Manifesto | with Levine, Locke and Searls |
| 2002 | Small Pieces Loosely Joined | |
| 2007 | Everything Is Miscellaneous | |
| 2012 | Too Big to Know | |
| 2019 | Everyday Chaos | HBR Press |
| 2026 | Beautiful Particulars | MIT Press |
3.2. Martin Wattenberg
DeepMind and former Harvard professor are on the event page; the PAIR co-lead role and Othello-GPT co-authorship come from the agent's own knowledge and were not checked against a source in this run.
- Building and studying AI systems at Google DeepMind; former Harvard professor.
- Co-led Google PAIR, which overlaps with Weinberger's Google stint [unverified].
- Co-author of the Othello-GPT "world model" paper (Li et al., ICLR 2023) [unverified]. This is the interpretability counterweight to the book's thesis.
4. The book
MIT Press, 184 pp, hardcover $27.95, ISBN 9780262054867, US pub 2026-11-03 (PRH + harvard.com); AU listing 2026-11-24. No reviews or excerpts found yet.
Beautiful Particulars: How AI's Attention to the Smallest of Differences Is Reshaping Our Biggest Ideas. MIT Press, 2026-11-03, 184 pp., ISBN
4.1. Questions the listing poses
- What does a coin look like to AI?
- Why are an AI's reasons for "the earth is round" so different from ours?
- Why does an image generator produce smiling Aztec-warrior selfies it "believes" are historically unlikely?
- Could a bodiless conscious AI make good moral decisions?
- Why is the world better at being causal than AI could "understand"?
- Creativity as moments that feel right, not a spark.
- How do you take the temperature of a joke?
4.2. Blurbs
attributions correct as printed; list was incomplete --- adds David Edmonds and Esther Dyson.
- Steven Levy, author of Hackers, Crypto and Facebook: The Insider Story.
- Jonathan Zittrain, Harvard (law, CS, public policy).
- Bruce Schneier, co-author with Nathan E. Sanders of Rewiring Democracy (MIT Press, 2025).
- danah boyd, Cornell; author of Data Are Made, Not Found (Chicago, 2025).
- David Edmonds, author of Death in a Shallow Pond.
- Esther Dyson, author of In Defense of Limits (forthcoming).
4.3. Lineage
2017 essay title was given as "Alien Knowledge: When Machines Justify Knowledge"; published title is "Our Machines Now Have Knowledge We'll Never Understand". Philosophies 2024 paper added as the missing link.
The argument has been building for nearly a decade:
- Our Machines Now Have Knowledge We'll Never Understand (Backchannel/Wired, 2017-04). Models justify their outputs in ways humans can't follow, which strains the idea of knowledge as justified true belief.
- Everyday Chaos (2019). Laws give way to complex models; prediction improves while explanation drops away.
- "The Rise of Particulars: AI and the Ethics of Care", Philosophies 9(1):26, 2024-02, doi:10.3390/philosophies9010026. Argues that ML keeps the particulars ("particulars speak louder") and ties this to care ethics. The book reads as this paper at length, extended to free will, causality and creativity.
4.4. Outside lineage
bibliographic data from agent knowledge, not fetched this run; run citation-auditor before promoting.
- Breiman, "Statistical Modeling: The Two Cultures", Statistical Science 16(3):199–231, 2001. https://doi.org/10.1214/ss/1009213726. Prediction as the referee.
- Anderson, "The End of Theory", Wired 16.07, 2008-06-23. The crude version of the thesis that Weinberger has to distinguish himself from.
- Pearl and Mackenzie, The Book of Why, Basic Books, 2018 (ladder of causation, ch. 1); Pearl, Causality, 2nd ed., CUP, 2009.
- Dancy, Ethics Without Principles, Oxford, 2004. For the other side, McKeever and Ridge, Principled Ethics, Oxford, 2006.
- Borges, "Funes el memorioso" (1942; Ficciones, 1944). English: Irby, trans., Labyrinths, New Directions, 1962. "To think is to forget differences, generalize, make abstractions."
- Zhang et al., "Understanding deep learning requires rethinking generalization", ICLR 2017.
6. Claims to test refutation
- "AI works through particulars, not generalizations"
- falsified if interpretability keeps finding compact, law-like structure (world models, shared circuits) that does the work. Ask Wattenberg.
- "Preserving particulars is what makes ML succeed"
- in tension with Zhang et al. 2017 and with compression accounts of generalization. Memorizing the training set generalizes worse. Falsified if the book's examples work only on memorized cases.
- "The world is better at being causal than AI could understand"
- a claim about rung two or three of Pearl's ladder. Falsified for any system that intervenes and learns from the result, such as agents running experiments.
- Moral particularism as a model for AI ethics
- is not falsifiable as stated. An audit needs some generalization to check against.
7. Open questions / Q&A
- If knowledge lives in particulars, what is a specification? A spec is a "for all x". Is that form going away, or can it now only be checked after the fact, one case at a time?
- Funes couldn't think because he couldn't forget. Isn't compression, which throws particulars away, the thing that makes ML work?
- (Wattenberg) Othello-GPT found a world model inside the network. Are models quietly generalizers, or are circuits just more particulars?
- Pearl's ladder: does the book let structural causal models back in, or is causality left to the world?
- Without principles, what does an auditor check? Can you state fairness without generalizing, and can checking cases one at a time ever finish?
- Breiman made predictive accuracy the referee. Isn't a held-out test set itself a generalization?
- After four years inside Google's AI groups: did practitioners treat particulars as the knowledge, or as noise to regularize away?
8. Actions
[ ]Post-event: notes by topic; what Wattenberg pushed back on[ ]Read the Philosophies 2024 paper before the book[ ]citation-auditor pass on Outside lineage and Wattenberg before promoting[ ]Gap: a research note on particulars vs. specs (Breiman, Pearl, Funes)