# AI Has Exposed the Biggest Lie About Humanity ![[images/AI-Biggest-Lie-Weinberger.webp]] **Creator:** Infotechrg (@InfoTechRG, shown on YouTube as "Infotech"), the *Digital Disruption with Geoff Nielson* podcast · **Guest:** David Weinberger · **Published:** 2026-11-09 · **Length:** 1:13:07 · [Watch on YouTube](https://www.youtube.com/watch?v=4GNsmfBuwLM) > Caption source: the uploader's own (manually created) English subtitles, fetched with yt-dlp (android_vr client) on the first try. The subtitles have some phonetic errors, which are corrected in this note: "Jeff Nielsen" is **Geoff Nielson**; "a Lee researcher" is **Fei-Fei Li**; "chicken sectors/fixer" means **chicken sexers**; "Peter Crocker" is **Peter Drucker**; "interleave" is **Interleaf**; "McLennan Companies" is **McLean & Company**; "Di" means **DEI**; and "a pie driven advisory" is the ad's "API-driven advisory". Timestamps come from the chapter list and the subtitles. Screenshots are frames from the video. The video was added from the user's "Summarize" playlist. ## Outline ### Cold open and introduction (0:00 to 1:05) - [0:01](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1s) A pre-show moment is left in: Nielson tells Weinberger he wants to cover how to conceptualize AI ("some of that kind of Heidegger influenced stuff"), "beautiful particulars, your upcoming work", and "everyday chaos" and its business implications. ![[images/AI-Biggest-Lie-Weinberger-t0020.webp]] *Host Geoff Nielson at the Digital Disruption desk in the cold open* - [0:35](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=35s) Nielson's intro: "AI is everywhere, rewiring everything. But maybe its most fascinating magic trick is how it's making us rethink our world and ourselves." He introduces Weinberger as a "veteran technologist, author, and philosopher". ![[images/AI-Biggest-Lie-Weinberger-t0050.webp]] *Guest card: David Weinberger* ### How technology shapes our understanding of the world (1:06 to 4:44) - [1:27](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=87s) Weinberger's premise: in the West, "our dominant major technology has shaped how we think about ourselves in the world." - [1:41](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=101s) The Enlightenment: clocks and watches were the peak technology, and "We got the clockwork universe", made of simple, exquisitely engineered pieces that combine into something complex. - [2:15](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=135s) Computers in the early 1950s made "everything" look "like information being processed": "I can't process it", the brain as computer, DNA as information. - [2:49](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=169s) Then the web, when "everything starts to look like a messy web". Now "it's hard to deny that we're in the AI age", though "We're like two minutes into this new era." ![[images/AI-Biggest-Lie-Weinberger-t0130.webp]] *Weinberger on the clockwork universe and the computer age* - [3:14](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=194s) With a philosophy background and 40 years in and writing about tech, he cares as a citizen about jobs, bias, inaccuracy and "AI laser bearing robots", but his intellectual interest is "how AI may be shaping our ideas about ourselves". The universe and our lives are chaotic, "Little changes can have gigantic effects", so he won't predict what will happen. ### How AI changes our thinking about intelligence (4:45 to 7:18) - [4:50](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=290s) Nielson: in 30 to 50 years "we'll have a better rearview mirror than we do windshield right now." - [5:34](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=334s) Anecdotally, more people say **optimize**. Optimizing admits that the result won't be perfect, that there are multiple values, and that you must balance them. That differs from the computer era's "input output unambiguous" thinking: "No we don't. We get choices. We get a possibility space." - [7:01](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=421s) "The world is not set up so that there is a perfect that satisfies all needs." We have to choose what to optimize "and have a serious discussion about our values." He calls that "a great step forward". ### What learning language reveals about humans and AI (7:19 to 9:28) - [7:22](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=442s) He watched his four-year-old granddaughter acquire words like "yet" from "an AI point of view, a large language point of view". Nobody taught her a definition; she learned from lots of examples, as connections formed between the word, reality and other words: "she's learning like a machine learning system did." - [8:44](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=524s) The caveat: that does not mean an LLM works exactly the way she does ("or that it's nearly as cute as she is"). - [9:11](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=551s) In business, this changes how we think knowledge "arises, develops and is handled", and it changes ideas about rule-based systems, "especially including rule based management." - [9:29](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=569s) Ad: Info-Tech's CIO Analytics, "a single platform to run the IT department". ### Is AI a new way to understand reality? (10:28 to 14:20) - [10:28](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=628s) Nielson (who has a five-year-old and a two-year-old) asks whether the new self-image is "underlying truth" or merely useful, since 200 to 300 years ago we'd have said the brain and universe are mechanical, "all cogs". - [11:51](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=711s) "That is a great question, which is how podcast guests gain a little time to think about an answer." He makes the case that this change is different. - [12:22](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=742s) Earlier technology-driven worldviews "were reductive": they found laws that "are the same for everybody", like Newton's law of gravitation applying to everything with mass, or one program handling all data by reducing it to what it was programmed to process. Aristotle, by contrast, thought things interacted based on their differing properties. - [13:29](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=809s) The web's "messiness is the opposite of reduction": everything depends on "messy relationships that cannot be pinned down". - [14:06](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=846s) "our brain seems to like those generalizations. They simplify thinking". ### Why data changed the direction of AI (14:21 to 15:44) - [14:21](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=861s) Machine learning is the kind of AI he means. Around 2007, during another lull ("type of AI winter"), researcher **Fei-Fei Li** suggested that the genius algorithm-writers weren't paying attention to the data: "Maybe we should start with the data, not with the algorithms." She also built a dataset that could easily be used (this is ImageNet, not named on air). ### Machine learning explained: patterns versus rules (15:45 to 19:40) - [16:12](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=972s) The classic example is handwritten numerals on messy checks. The rules approach says a one is a straight line and an eight is two circles, "what your second grade teacher" taught. But in real samples, an upright one "is rare" and a textbook eight "Almost never" happens. ![[images/AI-Biggest-Lie-Weinberger-t1000.webp]] *Weinberger explains the handwritten-numeral example* - [17:03](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1023s) Instead, have the computer find the pixel patterns that distinguish a three from an eight. It "doesn't care if they make sense or not". - [17:49](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1069s) His key point: "machine learning is looking for what makes differences." Rules look for what eights have in common. "we've all been brought up thinking that truth is in the similarities", in laws, categories and words. - [19:04](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1144s) "We've been bound by our limitation of our minds, which are tiny, you know, 2 pounds or whatever it is of brain". Looking at differences brings "remarkable breakthroughs like we've never ever seen before". He ends with "Thank you and good night." ### What generalizations leave out (19:41 to 23:33) - [19:41](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1181s) Nielson asks whether similarities and differences are two sides of the same coin. Weinberger grants you can flip it, but "what makes something what it is, is it's differences from everything else" goes against at least 2,500 years of Western culture's search for essences, which goes back to the ancient Greeks. - [21:48](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1308s) Newton, "One of the two greatest geniuses in Western history" (the other, he says, is Aristotle), "nails it", but treating things as mass reduces them to what every physical thing has in common. The law is "vastly simplifying, extremely useful", and is never precisely accurate anywhere in a universe full of mass. - [23:08](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1388s) Machines "can look at the particulars and details of individual things and use that to advance knowledge at the expense of human understanding, because it doesn't care whether we understand the patterns." - [23:34](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1414s) Ad: McLean & Company's Human-Centric AI Program. ### AI and the limits of human understanding (24:14 to 27:22) - [24:44](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1484s) He agrees that the shift is seismic and bigger than earlier ones, because there is empirical evidence that machines that "don't know anything" and aren't conscious, handling "inconceivable amounts of particulars", "routinely come up with better knowledge than we do", for example in protein folding and weather patterns. - [26:06](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1566s) "in some ways I think it may be a great lifting of a burden." Science should continue and get more support ("stop being such jerks about it"). But acknowledging "the finitude of our of our minds" relieves us of "thinking that we are the rational animals, our essence" and that "Our destiny is to understand the universe." ### Can we trust answers we don't understand? (27:23 to 29:25) - [27:23](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1643s) Nielson finds a sadness in this and compares it to *The Hitchhiker's Guide to the Galaxy*'s answer machine ("42"): are we just feeding data to a machine and trusting its answer, "somewhere between scary and dangerous"? ![[images/AI-Biggest-Lie-Weinberger-t1700.webp]] *Nielson raises the Hitchhiker's Guide "42" comparison* - [28:39](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1719s) Weinberger: we will keep trying to understand the universe in human terms, which requires reducing things, but should recognize we can't do it all: "the world is essentially chaotic and unpredictable by human minds." ### Rethinking fairness through individual differences (29:26 to 32:38) - [29:26](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1766s) He hopes the shift will change our minds about **fairness**, a moral idea so simple ("You can explain it to a four year old") that Western philosophy has rarely taken it seriously. - [29:57](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1797s) The cake example: at the eight-year-old's birthday party, the four-year-old asks why her slice is small. You answer that her sister is bigger and it's her birthday, giving two relevant differences. Abstractly, fairness "means treating people exactly the same. Unless there's a relevant difference." - [31:25](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1885s) In real life (he cites 1960s laws on fairness that we still argue over, and DEI), the argument is about the particulars of the case, such as a hire. If we accept that, we see "fairness as a conversation about what you want to optimize for", which "gives validity to the particulars". ### AI bias: who decides what to optimize for? (32:39 to 35:21) - [32:39](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1959s) Nielson: "relevance is now doing an awful lot of heavy lifting". Who decides what is relevant and what to optimize for, and is that surmountable? - [33:55](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2035s) The bad outcome is to feed morality, company vision or strategy into the machine. Computers in the 1950s were feared and ridiculed for this, as in the film *Desk Set* with Katharine Hepburn (he can't recall her co-star): a computer expert arrives at a company with a team of women librarians, the argument against him is that his computer has "no humanity", and at the end the computer spews paper chaotically "And librarians win." ![[images/AI-Biggest-Lie-Weinberger-t2070.webp]] *Weinberger recalls the 1957 film Desk Set* ### Why human values require human conversation (35:22 to 37:50) - [35:41](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2141s) Example: if an AI résumé screener is passing "all white guys" to interviews, you look at what it's missing, stop it using proxies for gender, and decide, for instance, that you care about innovative thinking more than grad school. Then "you're having the right conversation." - [36:23](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2183s) Likewise for a car with 15 contradictory goals (speed, minimal ecological impact, safety). "these are human values, and that conversation is a deeply human one, and it needs to happen among people. It can't happen among machines." - [37:20](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2240s) "fairness ultimately is an a good faith conversational negotiation." Saying only "it's unfair" is a single-dimensional view, "Then everybody loses." ### Where AI can help business decision-making (37:51 to 40:06) - [38:15](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2295s) It is easiest when the outcome is clear, as in supply chain management (assuming an ethical supply chain without, for example, slave-sourced raw materials). - [39:05](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2345s) A job-application sifter, after careful testing for systemic bias, might do a first pass, though he'd "still be a little nervous". AI helps most where the value aimed for is clear and the human values are "relatively simple or settled", as in much of medicine. ### Separating AI capabilities from human responsibility (40:07 to 42:03) - [40:07](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2407s) Nielson, deliberately facetious: AI is great for things that don't require human values, "And by the way, everything requires human values." - [40:36](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2436s) Weinberger: protein folding's value (advancing medicine) is clear. If it is later misused, "It's a human problem", just like a carmaker's design trade-offs or a hiring decision. "That's not a computer problem. That's a human problem." ### AI writing tools and the importance of human review (42:04 to 43:15) - [42:06](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2526s) Overworked managers who must write reviews for too many reports speak their views to an AI, get a rough draft and edit it: "it's hard for me to be against that." It can go bad if someone never reads the output (as with homework), but "the human ultimately has complete control over what it says." ### Human expertise we can't fully explain (43:16 to 44:55) - [43:16](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2596s) **Chicken sexers**: a centuries-old trade whose experts can tell a chick's sex "and they can't tell you how". Trainees learn by guessing next to an expert who says yes or no, over and over, "in a machine learning sort of way", and good ones reach about 98% accuracy. (He mentions he is vegetarian and finds the culling of male chicks upsetting.) ![[images/AI-Biggest-Lie-Weinberger-t2650.webp]] *The chicken-sexer example of expertise that can't be explained* - [43:58](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2638s) If AI gets 99%, "AI will do that, and we won't complain", though he feels bad for a "long, noble profession", which amuses Nielson. ### Is the world becoming more chaotic? (44:56 to 48:33) - [45:00](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2700s) Nielson: is the world getting more chaotic, or is our understanding of it? - [45:52](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2752s) Weinberger: the world is objectively getting more complex, "more people, more stuff, more interactions, more moving parts and more interdependencies", which is the heart of chaos. - [46:50](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2810s) His forthcoming book (*Beautiful Particulars*) "explicitly takes an optimistic view" of how AI may be changing our idea of how the world works, "in part because I didn't want to be depressed for the five years that I was writing it." - [47:17](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2837s) Rules and classes are a "great shortcut" and "insanely useful" (skyscrapers are largely rule-based), "But we also know that virtually everything in our life is chaotic." The computer age made everything look like information, "which is helpful, but misleading because it's not. It's stuff." ### Why particular details matter to us (48:34 to 50:08) - [48:49](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2929s) Particulars are what matter to us: he gets "a little dopamine hit" thinking of his granddaughter because she's "not a generic granddaughter". Mattering is "at the essence of consciousness": "you can't be conscious that nothing matters to you." - [49:50](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2990s) Less happily, the particular nail in your tire. ### Information overload and the pressure on leaders (50:09 to 54:11) - [50:09](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3009s) Nielson: farmers and cobblers once did one thing for life; then products were refreshed every ten years; now everything is faster. The leaders he works with feel stress that is "palpable" and "unsustainable", and the five-year strategy "is out the window". Is there a message of hope? - [52:38](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3158s) Weinberger: "everything you say is true." His generation hasn't "quite absorbed the internet yet", and younger generations may learn from those raised on AI. "Reality has always been like this. We just have coping methods for blocking out particulars." The same fear was voiced about computers, and we got used to them. ### Black swans and the limits of five-year plans (54:12 to 56:22) - [54:32](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3272s) The "black swan idea" (Nassim Nicholas Taleb's *The Black Swan*, not named on air) "was a bestseller", but "why did we need to hear that? Doesn't everybody know that?" Besides big black swans, "every day is series of little tiny black swans." ![[images/AI-Biggest-Lie-Weinberger-t3300.webp]] *Weinberger on black swans and five-year plans* - [55:21](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3321s) The problem with a five-year plan isn't only the faster pace: "It's always a bad idea." Strategy in war wasn't a thing until the 18th century, "when suddenly, mainly Germans wrote books" about it. - [55:59](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3359s) Peter Drucker had to change the title of a book in the early 1960s because it used "business strategy", and his publisher said nobody would go for it. (This is *Managing for Results*, 1964, originally titled *Business Strategies*.) - [56:19](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3379s) "Strategy is a utopian idea." You do better "if you stay open to the developing particulars", which changes how leadership functions and where ideas come from. ### How open conversations uncover business ideas (56:23 to 58:08) - [56:23](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3383s) Ideas should come from all ranks and from customers talking "unfettered", not always about business, with discussions filtered up to management "with permission". AI can help with the filtering. An open internal network is "where new ideas come from." - [57:37](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3457s) Decision-making "feels bad" because it reduces possibilities, which goes against how humans "have thrived by increasing possibilities." ### Why leadership needs more than one perspective (58:09 to 59:43) - [58:08](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3488s) Nielson: is it a skill to keep options open while filtering noise? Weinberger prefers to think in terms of conversation and social organization rather than individual skill: "Leaving it up to a single leader just seems to me to be nuts." Even his favorite president should be "a networker". - [59:26](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3566s) "in days of AI, I will say it's about accepting the particular nature of life and cherishing that, cherishing the differences." ![[images/AI-Biggest-Lie-Weinberger-t3560.webp]] *Leadership as conversation, not a single decider* ### The risks of relying on a single leader (59:44 to 1:02:47) - [59:44](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3584s) Nielson: culture currently worships the lone founder, and "decision by committee" is denigrated as slow. - [1:00:23](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3623s) Weinberger: committees "can be a pointless, time wasting circus", and someone must be empowered to decide, but one supreme leader with "miraculous powers" is "insanely risky". We have the technical infrastructure for networked input, but not yet good processes to turn it into decisions ("Not going to be by voting"). - [1:01:51](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3711s) Hierarchies "infested with implicit threat to your livelihood" squash real information. "good decisions are not made in small possibility spaces". ### Collaboration and innovation: Interleaf and Google (1:02:48 to 1:07:15) - [1:02:48](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3768s) Nielson: the enemy is ego and power. Has Weinberger seen organizations that work this way? - [1:03:18](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3798s) **Interleaf**, his first business job (he says "mid 90s", then "1986"), made an early integrated word processor and graphics system and had an internal discussion network open to any topic. When one of its image processors for printers failed, the engineers in the gaming discussion group, who knew graphics hardware as a hobby, "figured it out instantly. This was because of side conversation." ![[images/AI-Biggest-Lie-Weinberger-t3830.webp]] *Weinberger on Interleaf's open discussion network* - [1:04:52](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3892s) He was writer-in-residence at Google for four years, in the People + AI Research group, which was prolific because it was "a completely open discussion space". Nothing "was not ever viewed as a waste of time". - [1:05:56](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3956s) In a hallway, a developer working on embedding spaces called a colleague over: "it just figured out prefixes... I didn't tell it anything about prefixes." They both laughed with joy at the discovery. "humans depend upon and get deep pleasure from living in possibility spaces where new things happen." ![[images/AI-Biggest-Lie-Weinberger-t3960.webp]] *The embedding-space story from Google's AI research group* ### Parting advice for business and tech leaders (1:07:16 to 1:11:29) - [1:07:16](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4036s) Nielson asks for parting advice. Weinberger jokes: "other than day drinking. Bring back the three martini lunch." - [1:08:14](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4094s) Machine learning works despite being "often inaccurate because probabilistic". It works because "it doesn't know about the rules" but knows "impossibly large numbers of particulars computed at enormous expense and environmental damage". That tells us "particulars are in many ways constitutive of the world, particulars in their differences." - [1:09:43](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4183s) Examples: choosing that particular donut; a particular unexpected note in a song that lifts you; humor as the difference between setup and punchline ("If they're too close... It's not funny... too far, it's absurd"). ![[images/AI-Biggest-Lie-Weinberger-t4250.webp]] *Weinberger on humor as a matter of difference* - [1:10:59](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4259s) "It's liberating": being relieved "of the idea that we are the one creature in the universe who can understand it" ("I'm not saying AI can understand it, it doesn't understand anything"). "And if we can give up on that, then we'll see the world more clearly." ### Closing (1:11:30 to end) - [1:11:30](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4290s) Nielson thanks him: the idea is "simple, but it's really profound". - [1:12:18](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4338s) Outro and Info-Tech Research Group ad. ## Key takeaways - Each dominant technology has supplied a metaphor for the world (clockwork, information processing, the web). Weinberger's claim is that AI's metaphor is different because it is not reductive. - Rules, laws and essences work by discarding differences. Machine learning works by finding the patterns that distinguish particular things, and it gets results humans often can't understand. - Accepting the limits of human understanding can be a relief rather than a loss: we are not obliged to be the rational animal that understands everything. - Fairness and AI bias are questions of which particulars are relevant and what to optimize for. Those are value questions for human conversation, not for the machine. - AI helps most where the goal is clear and the values settled, and where a human reviews and controls the output (for example, first drafts of reviews). - Five-year plans and single-leader decision-making ignore how chaotic the world is. Open, unfettered conversation across a company surfaces ideas and problems that hierarchies squash. ## Notable quotes - "our dominant major technology has shaped how we think about ourselves in the world." ([1:31](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=91s)) - "machine learning is looking for what makes differences." ([18:09](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1089s)) - "we've all been brought up thinking that truth is in the similarities." ([18:33](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1113s)) - "to advance knowledge at the expense of human understanding, because it doesn't care whether we understand the patterns." ([23:16](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1396s)) - "in some ways I think it may be a great lifting of a burden." ([26:05](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1565s)) - "fairness ultimately is an a good faith conversational negotiation." ([37:22](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2242s)) - "That's not a computer problem. That's a human problem." ([41:47](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2507s)) - "Strategy is a utopian idea." ([56:19](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3379s)) - "Leaving it up to a single leader just seems to me to be nuts." ([58:49](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3529s)) - "good decisions are not made in small possibility spaces" ([1:02:15](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3735s)) ## Claims check | Claim | Verdict | Notes / source | | --- | --- | --- | | The Enlightenment's clocks and watches produced the "clockwork universe" picture ([1:41](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=101s)) | Broadly accurate | The clockwork metaphor is associated with 17th-century mechanical philosophy (Descartes, Boyle) and became widespread with Newtonian physics in the 1700s. | | Around 2007, a researcher (Fei-Fei Li) proposed starting with data rather than algorithms and built an easy-to-use dataset ([14:32](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=872s)) | Accurate | Fei-Fei Li began ImageNet around 2006 to 2007; it was presented in 2009, and its 2012 competition result (AlexNet) launched the deep-learning boom. | | Handwritten numerals on checks were a classic machine-learning problem ([16:12](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=972s)) | Accurate | Yann LeCun's convolutional networks read handwritten check amounts in the 1990s; the MNIST digit dataset became the standard benchmark. | | The brain weighs about 2 pounds ([19:10](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1150s)) | Slightly low (he hedged) | An adult human brain weighs about 3 pounds (roughly 1.3 to 1.4 kg). | | AI does better than humans at protein folding and weather prediction ([25:57](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1557s)) | Broadly accurate | DeepMind's AlphaFold transformed protein structure prediction (its creators shared the 2024 Nobel Prize in Chemistry), and ML weather models such as GraphCast match or beat traditional forecasts on many measures. | | *Desk Set* features Katharine Hepburn, a computer expert, a team of women librarians, and a computer that goes haywire at the end ([34:23](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2063s)) | Broadly accurate | The 1957 film stars Hepburn and Spencer Tracy (the co-star he couldn't recall) as a methods engineer installing the EMERAC computer in a TV network's research department. The machine malfunctions, the researchers keep their jobs, and the ending is in their favor. | | Expert chicken sexers reach about 98% accuracy and can't explain how ([44:10](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2650s)) | Broadly accurate | Professional vent sexers are often cited at about 98% accuracy, and cognitive scientists use them as a classic example of tacit, hard-to-articulate expertise. The AI 99% figure was his hypothetical. | | Strategy in war wasn't a thing until the 18th century, mainly through German books ([55:36](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3336s)) | Partly accurate | The modern sense of "strategy" entered European military writing in the late 18th century, first in French (Joly de Maizeroy) and then in German (von Bülow, later Clausewitz). Strategic thinking itself is much older, for example Sun Tzu. | | Peter Drucker had to change a book's title in the early 1960s because his publisher thought "business strategy" wouldn't sell ([55:59](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3359s)) | Accurate | *Managing for Results* (Harper & Row, 1964) was originally titled *Business Strategies*; Drucker said the term was then unfamiliar in business ([Jim Collins on Drucker](https://jimcollins.com/article_topics/articles/the-daily-drucker.html)). | | The black swan idea "was a bestseller" ([54:45](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3285s)) | Accurate | Nassim Nicholas Taleb's *The Black Swan* (2007) was a long-running bestseller. | | Interleaf made the first integrated word processor and graphics system, and he worked there in the "mid 90s" / 1986 ([1:03:18](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3798s)) | Partly accurate; dates inconsistent | Interleaf (founded 1981) was a pioneer of WYSIWYG document publishing combining text and graphics. "Mid 90s" conflicts with his own "1986" a few sentences later; the exact years were not checked. | | He was a writer-in-residence at Google for four years, in the People + AI Research group ([1:04:52](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3892s)) | Consistent with his bio | His site describes him as "most recently as a writer-in-residence in Google Responsible AI groups" ([beautifulparticulars.com](https://beautifulparticulars.com/)). PAIR is Google's People + AI Research initiative. The four-year length was not checked. | | His upcoming book takes an optimistic view of AI and took five years to write ([46:50](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2810s)) | Accurate (book exists) | *Beautiful Particulars* was published by MIT Press on 2026-11-03 ([MIT Press](https://mitpress.mit.edu/9780262054867/beautiful-particulars/)); the five-year figure is his own. | | Machine learning works because it ignores rules and models particulars, which tells us particulars are "constitutive of the world" ([1:08:34](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=4114s)) | Interpretation | This is his philosophical thesis, not a testable claim. ML models do learn statistical patterns from examples rather than hand-coded rules, but whether that reveals something about reality is a philosophical argument. | | Five-year plans were "always a bad idea"; strategy is "utopian" | Opinion | His view; many strategy scholars would argue for adaptive rather than no planning. | ## Things mentioned ### Books and texts - *Beautiful Particulars: How AI's Attention to the Smallest of Differences Is Reshaping Our Biggest Ideas* by David Weinberger (MIT Press, 2026), his "upcoming work" ([0:22](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=22s), [46:50](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2810s)) - *Everyday Chaos* by David Weinberger (2019), referenced as "everyday chaos" in the pre-show ([0:25](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=25s)) - *The Hitchhiker's Guide to the Galaxy* by Douglas Adams ([27:34](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=1654s)) - *The Black Swan* by Nassim Nicholas Taleb, referred to as "the black swan idea" and "a bestseller", not named on air ([54:32](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3272s)) - *Managing for Results* by Peter Drucker (1964), originally titled *Business Strategies*; Weinberger tells the title story without naming the book ([55:59](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=3359s)) ### Films - *Desk Set* (1957), with Katharine Hepburn (and Spencer Tracy, not named) ([34:23](https://www.youtube.com/watch?v=4GNsmfBuwLM&t=2063s)) ### People - Martin Heidegger (pre-show mention), Isaac Newton, Aristotle, Fei-Fei Li, Peter Drucker, Katharine Hepburn ### Companies, products and organizations - Info-Tech Research Group (producer), its CIO Analytics product, and McLean & Company's Human-Centric AI Program (ads) - Interleaf; Google and its People + AI Research (PAIR) group ### Concepts - Clockwork universe, information processing, the "messy web", optimization and possibility spaces, reductionism vs. particulars, essences, machine learning vs. rule-based systems, embedding spaces, fairness as relevant difference, AI bias and proxies, black swans, chicken sexing as tacit expertise, chaos and complexity ## Biographies ### David Weinberger An American writer and technologist with a PhD in philosophy and about 40 years in technology companies, most recently as a writer-in-residence in Google's responsible-AI groups. He co-wrote *The Cluetrain Manifesto* (1999) and wrote *Small Pieces Loosely Joined*, *Everything Is Miscellaneous*, *Too Big to Know*, *Everyday Chaos* (2019) and *Beautiful Particulars* (2026). He has been affiliated with Harvard's Berkman Klein Center for about two decades and is a researcher at Harvard's metaLAB ([beautifulparticulars.com](https://beautifulparticulars.com/)). In the interview he mentions working at Interleaf and being vegetarian, and has a four-year-old granddaughter. ### Geoff Nielson The host of *Digital Disruption with Geoff Nielson*, produced by Info-Tech Research Group. He mentions having a five-year-old and a two-year-old and working with many business and technology leaders. ### Info-Tech Research Group (the channel) An IT research and advisory firm based in London, Ontario (Canada). McLean & Company, the HR advisory firm in the ad, is a division of Info-Tech.