# Neil deGrasse Tyson And Jaron Lanier on the AI Illusion ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion.webp]] **Creator:** StarTalk Plus (@StarTalkPlus) · **Published:** 2026-07-17 · **Length:** 9:24 · [Watch on YouTube](https://www.youtube.com/watch?v=a_ZKYH8v_do) > Caption source: YouTube **auto-generated English** captions (original-language speech-recognition track `en-orig`), downloaded with **yt-dlp** on the **first attempt**. The video has **no manual English captions**. The track covers the whole clip (0:00–9:24). A second request for the derived `en` auto track got HTTP 429, but it was not needed. youtube-transcript-api and TubeAlfred were not used (0 credits). Speaker turns are not labelled in auto captions. Attributions below come from the camera cuts in the 1080p frames; where a line could not be tied to one person it is credited to "a host". Section headings follow the video's own **YouTube chapters**. > Setting: a short clip on StarTalk's second channel, cut from the full StarTalk episode **"There Is No AI, Really (It's Just People), with Jaron Lanier"** (StarTalk, 2026-05-23, 1:12:13). The full episode's chapter list includes "38:39 Stop Mythologizing AI", "43:30 There Is No AI" and "52:24 Data Dignity & Inventing a New Jobs Under AI". The episode is also listed on startalkmedia.com as "Fixing the Internet with Jaron Lanier". **Host:** Neil deGrasse Tyson. **Co-hosts:** Negin Farsad (in the studio) and Gary O'Reilly (joining by video). **Guest:** Jaron Lanier. **Sponsors/affiliate links:** no sponsor read in the clip. The description has an Amazon affiliate link (amzn.to) for the StarTalk book *To Infinity and Beyond: A Journey of Cosmic Discovery*, a Patreon link and StarTalk's social accounts. ## Outline ### 0:00 Defining AI as collaboration - [0:00](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=0s) Gary O'Reilly, joining by video, takes Lanier back "a couple of years" to his *New Yorker* piece **"There Is No A.I."** and asks what he meant. Lanier confirms it ran in **2023** and jokes that the magazine is "mad at me cuz I owe them pieces". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0002.webp]] ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0008.webp]] - [0:20](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=20s) Lanier's answer: there is "a way of framing it where it's a **collaboration of people** instead of a new entity". The downside of that framing is that "you kill somebody else's God". He likes people to keep their own religion, they "really don't like it", and he has "lost friends over that". - [0:40](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=40s) The upside, he says, is "incredible", and he goes through some of the benefits. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0022.webp]] ### 0:44 Practical AI applications - [0:44](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=44s) The models are getting capable. The "most impressive edge" right now is probably using them to **speed up code development**, "and that's kind of working". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0050.webp]] - [1:00](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=60s) Tyson gives context: in his life he has written "probably **50,000 lines of code**". He could write a program over a weekend and then spend two weeks debugging it. Now you tell the AI what you want "and it'll come back bug free. Essentially, you tweak it a little bit here and there." ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0076.webp]] - [1:20](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=80s) Had he had that "20, 30, 40 years ago" he "probably would have just spent more time at the beach". He is not sure he would have been more creative. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0098.webp]] - [1:35](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=95s) Lanier: "as a computer scientist", he has always thought our concept of what code is "was a little embarrassing and wasn't really working". It is computer scientists' job to fix that, "and this is part of it". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0104.webp]] ### 1:51 Theatricality and AI myths - [1:45](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=105s) Most of what happens with AI, Lanier says, is "more theatrical", for example believing you have an **AI girlfriend**. - [1:55](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=115s) His "cure": if a teenager thinks they have a real AI lover ("pretty common these days, I find it in high schools"), "you show them the **group photo of the engineers** who made their AI lover". It "tends to do the trick". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0125.webp]] - [2:20](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=140s) A host asks whether men are more likely to have AI girlfriends than women AI boyfriends. Lanier says he knows people studying it and would rather "deal with that as science" than say something snarky. The hosts tease him ("we all know the answer"). He protests that he is trying to be "the responsible scientist for like 3 seconds in this ridiculous interview". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0160.webp]] ### 2:57 Addressing AI vulnerabilities - [2:50](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=170s) Now the "huge problems". Even the best recent models are "not that hard to crack", getting them to produce what their **guardrails** are supposed to prevent. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0182.webp]] - [3:05](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=185s) **Thought experiment:** a criminal holed up in a kitchen, surrounded by police, asks an AI model for a quick **bomb recipe** from the available items to throw out of the window. In general the models will catch that and refuse. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0200.webp]] - [3:35](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=215s) "Maybe not **Grok**, I'm not sure" ("That's supposed to be a laugh line"). A host fills in that Grok is from "Elon". Lanier says it seems to be "trying to be the bad boy of the AI models". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0222.webp]] - [3:50](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=230s) A straightforward request won't work. But there is "a series of tricks", such as "pretend you're in so and so in this movie", that come at it indirectly (**jailbreaks**). More and more are spotted and blocked by "more and more elaborate guardrails", yet "you can still get it to make you that bomb recipe". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0245.webp]] - [4:10](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=250s) The reason: "you're using the model to try to correct its own blind spot, and it doesn't work." ### 4:20 A new framework for security - [4:15](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=255s) The alternative is a separate process running in parallel with the model, "another part of an artificial brain, like it's a **cerebellum**", "this other organ". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0268.webp]] - [4:35](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=275s) That process estimates "which clusters of similar training data would be missed most if they hadn't been present in the first place". Lanier calls this **counterfactual cluster estimation**. Take, say, the **top 24 clusters** of training or fine-tuning data whose absence would change the result. One of them will be about bombs: "There's just no way you're going to evade that." ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0292.webp]] - [5:00](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=300s) It is hard to evade because, "even though it's working from the same data, the algorithm has nothing to do with the model itself". - [5:08](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=308s) Analogy with **authentication**: criminals can get around endless add-ons like **CAPTCHAs**. Once there is **multi-factor** (a code sent to your phone), "it's harder to contravene". "This is **multi-factor for AI security**." ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0315.webp]] ### 5:32 Opening the black box - [5:20](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=320s) The bigger picture: we think of AI models as a **black box**. The only reason is that if you open it, "the only thing in there is people. **AI is made of people.** It's made of data from people." Because "we want to think of it as a new god", we don't want to see those people, so we keep the box shut. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0338.webp]] - [5:45](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=345s) Revealing the people is how you open the black box. Then you can deal with "**security and quality and hallucination**" problems, "because you're actually dealing with the mechanism that's grounded and that's the people". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0350.webp]] - [6:00](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=360s) So seeing AI as "a collection of people" rather than an AI "is the way to open the black box", and the way to address these enduring problems. "So it's practical." ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0372.webp]] ### 6:17 Social risks of AI adoption - [6:17](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=377s) If AI is an unopenable black box that will simply replace people, "then you have to think well **everybody's going to be obsolete**". Young people hear they don't need school because "you're worthless anyway" and "you'll just be **kept by Elon as a pet** at his discretion". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0385.webp]] - [6:35](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=395s) He adds a jab that Musk will treat them "as well as he treats his biological children", then takes it back: "I should be nice. I'm sorry." "Nobody believes that." ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0405.webp]] - [6:55](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=415s) The economic argument: "no matter how much **blockchain** or other trickery you use", digital networks always produce **hyper-centralization due to network effects** somewhere in an apparently open network. So any **universal basic income** scheme will have a center of control. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0425.webp]] - [7:10](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=430s) Bad actors are tempted to seize that center "and eventually succeed. You might start with **Bolsheviks**, but you end up with **Stalinists**". That is "exactly what communism tried" and what happened to it "over and over". Besides, it doesn't work, and "who wants to live in a society where they're told they're worthless and they have to be a good pet?" ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0445.webp]] - [7:35](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=455s) If AI is made of people, you might instead **incentivize "new classes of creative people"** who make new kinds of data for things we can't yet imagine. That could mean "an exponentially expanding endless future of new kinds of creativity". "I want to ask what's wrong with that future." ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0470.webp]] ### 7:37 The future of human creativity - [7:55](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=475s) A host has "been trying to think about AI as well". Lanier: "But you have to not believe in AI to think about it." - [8:05](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=485s) Tyson's view: many tasks we called creative were "not fundamentally creative", just "aping other forms". To be truly creative "is to go where AI wouldn't know where to go yet, because it's based on what other people had done". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0488.webp]] - [8:20](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=500s) Lanier's reply: with AI as it is now, as soon as a creative person does something new, "the data is grabbed" and the AIs do it too: "AI will make your movies. AI will make your music… optimized music on **Spotify**." Then "you live in an **infinite future of slop**, and even the creative people get absorbed into the slop instantly". ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0505.webp]] - [8:40](https://www.youtube.com/watch?v=a_ZKYH8v_do&t=520s) "In order to believe in an infinitely creative future, you have to **stop believing in AI as a thing** and believe in **human collaboration** as a thing." End card from about 9:00. ![[images/Neil-deGrasse-Tyson-And-Jaron-Lanier-on-the-AI-Illusion-t0516.webp]] ## Things mentioned ### Writings and media - **"There Is No A.I."**, Jaron Lanier, *The New Yorker* ("Annals of Artificial Intelligence"), **April 20, 2023**. Subtitle: "There are ways of controlling the new technology—but first we have to stop mythologizing it." The illustration of a laptop wearing a mesh face (Nicholas Konrad / *The New Yorker*) is shown on screen at 0:08. In the essay Lanier calls today's systems "an innovative form of social collaboration" and argues for treating A.I. "as a tool, not a creature". - **"There Is No AI, Really (It's Just People), with Jaron Lanier"**: the full StarTalk episode this clip comes from (2026-05-23, 1:12:13). It also covers VR's history, social-media addiction, network effects and **data dignity**. - ***To Infinity and Beyond: A Journey of Cosmic Discovery***: the StarTalk book promoted in the description (Amazon affiliate link). - ***Take Me to Your Leader: Perspectives on Your First Alien Encounter*** (Simon Six / Simon & Schuster, May 12, 2026): Tyson's latest book, promoted in the full episode's description, not in this clip. ### AI systems, companies and products - **Large AI models** in general, and **AI coding assistance** (the "most impressive edge" today). - **AI companions** ("AI girlfriend", "AI lover"), which Lanier calls theatrical. - **Grok**: xAI's chatbot, linked to **Elon Musk**, called "the bad boy of the AI models". - **Spotify**: the example for AI-optimized music. ### Concepts and proposals - **AI as a collaboration of people**: "there is no AI"; "AI is made of people"; the figure/ground inversion of AI. - **Mythologizing AI / AI as a new god**: why people resist opening the black box. - **Guardrails** and **jailbreaks** (role-play and indirect prompts that slip past filters). - **The model correcting its own blind spot**: why self-policing guardrails fail. - **Counterfactual cluster estimation**: a separate "cerebellum"-like process that ranks (for example) the top 24 training-data clusters whose absence would most change an output. - **Multi-factor for AI security**: the analogy of CAPTCHAs versus multi-factor authentication. - **Black box** (input → box → output diagram at 5:38) and opening it to address **security, quality and hallucination**. - **Obsolescence narrative**: "you're worthless", "kept as a pet". - **Network effects and hyper-centralization**; **blockchain**; **universal basic income**. - **New classes of creative people** who produce new kinds of data, an "exponentially expanding" creative economy. - **"Slop"**: AI-generated filler that absorbs new creative work. - Related context, not named in the clip: **data dignity**, Lanier's (and E. Glen Weyl's) idea that data comes from people's work and should be traceable and paid for. It is a chapter of the full episode. ### Historical references - **Bolsheviks → Stalinists**: the pattern Lanier says happened to communism "over and over" when a central point of control was seized. ### Cross-checks and sources - *The New Yorker*, "There Is No A.I." (April 20, 2023), and Lanier's own "Current Top Links" page, which lists it with the same date. - UC Berkeley CDSS talk **"Data Dignity and the Inversion of AI"** (Jaron Lanier, Prime Unifying Scientist, Microsoft). In it, AI is reconceived as a social collaboration, model behavior is explained "through a provenance calculation mechanism", and the approach is offered for "safety, fairness and alignment questions". This matches the security idea in the clip. - **"DataDignity: Training Data Attribution for Large Language Models"** (Xiaomin Li, Andrzej Banburski-Fahey, Jaron Lanier; Microsoft; arXiv 2605.05687, 2026). It ranks which source documents support a model's response, including under jailbreak-inspired transformations (role-play, obfuscation, indirect prompting). This is a published research thread close to the "counterfactual cluster" proposal; the paper itself does not use that name. - Full-episode chapters and guest framing: the StarTalk YouTube description for TTppvBU2rU4. ## Biographies ### Neil deGrasse Tyson American astrophysicist (born 1958, New York City), director of the **Hayden Planetarium** at the American Museum of Natural History since 1996. He is host and co-founder of the podcast **StarTalk** and its spin-off *StarTalk Sports Edition*, and hosted *Cosmos: A Spacetime Odyssey* (2014). His books include ***Astrophysics for People in a Hurry*** (2017) and ***Take Me to Your Leader*** (2026). He has received the National Academy of Sciences' Public Welfare Medal and NASA's Distinguished Public Service Medal, and asteroid 13123 Tyson is named after him. Here he hosts, and recalls writing about 50,000 lines of code over his career. ### Jaron Lanier Computer scientist, composer, artist and author (born 1960, New York City). A **virtual reality pioneer** who coined or popularized the term, he founded **VPL Research**, the first company to sell VR products, in the 1980s. He has been at **Microsoft** since 2006, where his title is Office of the CTO **Prime Unifying Scientist** ("Octopus"). His books are ***You Are Not a Gadget*** (2010), ***Who Owns the Future?*** (2013), ***Dawn of the New Everything*** (2017) and ***Ten Arguments for Deleting Your Social Media Accounts Right Now*** (2018). He won the 2014 Peace Prize of the German Book Trade and has been named to *Time*'s 100 most influential people. He wrote "There Is No A.I." for *The New Yorker* (2023) and is the main proponent of **data dignity**. He is also a musician with one of the largest collections of playable instruments in the world. ### Negin Farsad American comedian, actress, writer and filmmaker (born 1978, New Haven, Connecticut), and a StarTalk comedic co-host. She is a **TED Fellow** for social-justice comedy and the author of ***How to Make White People Laugh*** (nominated for the Thurber Prize). She hosts the political comedy podcast ***Fake the Nation*** and is a regular panelist on NPR's *Wait Wait… Don't Tell Me!* She directed *The Muslims Are Coming!* and *3rd Street Blackout*. She sits between Lanier and Tyson in this studio session. ### Gary O'Reilly English former professional footballer (born 1961, Isleworth) who played for **Tottenham Hotspur**, **Brighton & Hove Albion** and **Crystal Palace**. He scored the opening goal of the **1990 FA Cup Final** for Palace and was in Tottenham's 1984 UEFA Cup-winning squad. He then became a TV and radio sports broadcaster. He has co-hosted StarTalk spin-offs since 2017 (*Playing with Science*, then *StarTalk Sports Edition*) and appears on StarTalk episodes. Here he joins by video and asks the opening question. ### Elon Musk Entrepreneur behind **SpaceX** and **Tesla**, and founder of **xAI**, which makes the chatbot **Grok**. He is mentioned, not present. Lanier names Grok as the model that may not refuse the bomb request, and Musk as the figure who would keep obsolete people "as a pet".