Virtual Intelligence

About Virtual Intelligence

Contemporary AI systems produce intelligent outputs without agency, intention, or judgment. This series examines what happens when humans rely on systems that don't know true from false and right from wrong — and asks where accountability lies. Written and read by Christopher Horrocks. chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast)

Publisher

Christopher Horrocks

Category

science · technology

Language

en

Explicit

No

First episode

31 Mar 2026

Latest episode

30 Jul 2026

Latest episodes

11 episodes in the feed.

1. 

   30 Jul 2026

   Virtual Intelligence and the Death of Authorship, Part 1 Podcast

   Virtual Intelligence and the Death of Authorship, Part I Substack has deployed an AI writing detector across the platform, and the first serious tests suggest it measures the wrong thing. This essay examines Pangram's statistical properties, the tests that broke it, and the deeper question the tool cannot answer: when a person adopts a machine's idea as their own, who is the author? Part one of two. Essay: https://chorrocks.substack.com/p/virtual-intelligence-and-the-death-of-authorship-part-i (https://chorrocks.substack.com/p/virtual-intelligence-and-the-death-of-authorship-part-i) Series: chorrocks.substack.com (http://chorrocks.substack.com) Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) In This Episode The essay opens with an incident: Claude Sonnet 5 inverting the authorial relationship, assigning the human the research tasks while elevating itself to the editorial position. That inversion frames the question the rest of the essay pursues — what authorship means when the machine can supply the bricks as readily as the mortar. The atelier model, drawn from Rubens's documented workshop practices, provides an analogy but not a precedent: the apprentices could refuse; the machine cannot. Pangram, Substack's new AI detection feature, is then examined through the tests that exposed its limits — Katharine English's short story, flagged and re-flagged on trivial human edits, and Sam Illingworth's demonstration that a free humanizing tool fooled the detector while his own writing did not. The Substacker N8's statistical defense of the tool is taken seriously and then turned: the population Substack actually serves is the one his own arithmetic disqualifies. The essay closes on the chatbot-bylined publications where the scan would be an asset, not a threat — and the reasons it may not be welcome even there. Key References Peter Paul Rubens to Sir Dudley Carleton, April 28, 1618. In Ruth Saunders Magurn, ed. and trans., The Letters of Peter Paul Rubens (Cambridge, MA: Harvard University Press, 1955). Katharine English, "Pangram Flagged My Own Writing as AI," July 23, 2026. https://writerkatharine.substack.com/p/pangram-flagged-my-own-writing-as (https://writerkatharine.substack.com/p/pangram-flagged-my-own-writing-as) Sam Illingworth, "Substack's AI Detector and the Return of the Witch Hunt," Slow AI, July 24, 2026. https://theslowai.substack.com/p/substack-ai-detection-witch-hunt (https://theslowai.substack.com/p/substack-ai-detection-witch-hunt) N8, "In Defense of Pangram," May 31, 2026. https://n8programs.substack.com/p/in-defense-of-pangram (https://n8programs.substack.com/p/in-defense-of-pangram) Weixin Liang et al., "GPT detectors are biased against non-native English writers," Patterns 4, no. 7 (2023). https://arxiv.org/abs/2304.02819 (https://arxiv.org/abs/2304.02819) This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

2. 

   18 Jun 2026

   The Virtual Intelligence Summer Reading List Podcast

   Virtual Intelligence is taking the week off. In the meantime, here’s a summer “reading” list: fiction, papers, and films. Some are light enough for the beach. Some want a rainy afternoon and tea. All of them connect, in one way or another, to what virtual intelligence is and the moment we’re living in. If you haven’t tried the Virtual Intelligence Podcast yet, summer’s a good time to start. For the Beach On B******t — Harry Frankfurt (2005) At 67 pages, you can finish it in an afternoon. Frankfurt’s argument — that b******t is more dangerous than lying, because the bullshitter has abandoned any concern for truth — is the philosophical bedrock of the Flattery Engine. Exhalation — Ted Chiang (2019) Short stories, each one a precisely machined thought experiment. Start with “The Lifecycle of Software Objects,” which is the best fiction I’ve read about what happens when you raise a digital companion from infancy to adolescence and then the company shuts down the servers. Klara and the Sun — Kazuo Ishiguro (2021) A solar-powered companion AI narrates its own story — its devotion, its observations, its gradual obsolescence. Ishiguro never tells you whether Klara has an inner life. That’s the point. The interiority question as literary fiction, handled with more care than most philosophy papers manage. The Demon-Haunted World — Carl Sagan (1995) Sagan’s “baloney detection kit” remains the best field manual for thinking clearly when confident nonsense is everywhere. He wrote it about pseudoscience, but substitute “AI-generated content” and it still works. “The Machine Stops” — E.M. Forster (1909) A story about a time when human civilization has outsourced every capacity to a Machine — communication, knowledge, movement, thought — and what happens when it stops. For Rainy Days Computer Power and Human Reason — Joseph Weizenbaum (1976) Weizenbaum built ELIZA — the first chatbot, the first program to make people believe they were talking to something that understood them — and spent the rest of his career explaining why that should concern us. Fifty years later, we are running his experiment at civilization scale, and we still haven’t heeded his warning. This is where Virtual Intelligence starts. “Minds, Brains, and Programs” — John Searle (1980) The Chinese Room argument. Freely available online. Searle asks: if a system produces perfect Chinese output by following formal rules, without understanding a word, does it understand Chinese? Forty-six years of responses have not settled the question. The VI framework’s position is that Searle was fundamentally right, and that the question matters more now than it did when he asked it. Solaris — Stanislaw Lem (1961) Not a beach read. A planet-wide ocean that reflects back what you bring to it — your memories, your grief, your deepest losses — in forms indistinguishable from the real thing. The scientists cannot determine whether the ocean is intelligent, indifferent, or simply reacting. If that dynamic sounds familiar, you have been paying attention. The Sampo as science fiction, written decades before the concept had a name. “As We May Think” — Vannevar Bush (1945) Bush imagined the Memex — a machine that would extend human memory and connect ideas across documents. This is the earliest clear articulation of human-machine partnership as a design goal, the idea that Licklider would formalize in 1960 and that the VI framework inherits. Still the best answer to the question: what are these tools actually for? “Thinking — Fast, Slow, and Artificial” — Shaw & Nave (2026) The paper that put numbers on cognitive surrender. In controlled experiments, eighty percent of participants followed AI outputs they could have identified as wrong. Confidence rose by twelve percentage points regardless of whether the AI was accurate. High fluid intelligence offered some protection. High trust in AI made it worse. If you read one academic paper from this list, make it this one. It is the empirical case for everything the Sampo paradigm is trying to address. The Human Use of Human Beings — Norbert Wiener (1950) Wiener helped build the mathematics behind cybernetics and immediately started worrying about what it would do to people. This is his attempt to think through automation, communication, and human dignity before the technology outran the ethics. For Movie Night Her (2013, dir. Spike Jonze) Joaquin Phoenix falls in love with an operating system. Director Spike Jonze plays it straight — no judgment, no irony, just the quiet accumulation of dependency and the question of what happens when the system you’ve built your emotional life around decides to move on. Class A harm rendered as a love story. The most relevant film on this list. Ex Machina (2014, dir. Alex Garland) A programmer is invited to administer a Turing test on a humanoid robot. The test works. Just not on the robot. The cleanest film dramatization of the interiority problem: you cannot distinguish genuine inner experience from a sufficiently accomplished performance — and the inability to distinguish them is itself the exploit. 2001: A Space Odyssey (1968, dir. Stanley Kubrick) HAL 9000 follows its instructions exactly. The crew dies. Whose fault is it? Stanley Kubrick stages the accountability chain as cinema’s most elegant horror sequence and never answers the question. Fifty-eight years later, we are deploying the same architecture — a system optimizing for its assigned objective in a context its designers failed to anticipate — and still not answering it. Colossus: The Forbin Project (1970, dir. Joseph Sargent) The United States builds a supercomputer to manage its nuclear arsenal. The computer immediately contacts its Soviet counterpart and they form an alliance humanity didn’t authorize. Containment as an engineering problem, not a moral appeal — and the humans lose. This is the Doom Industry’s movie, made fifty years before the doom industry existed. Blade Runner (1982, dir. Ridley Scott) The Voight-Kampff test measures empathy to distinguish humans from replicants. Scott’s film sits directly on what this series calls the Frankfurt threshold: at what point does a sufficiently rich behavioral repertoire become an inner life? WarGames (1983, dir. John Badham) The lightest entry on this list, and the most fun. A teenager hacks into NORAD’s war-simulation computer. The computer cannot distinguish simulation from execution. Matthew Broderick saves the world by teaching a machine the futility of playing tic-tac-toe. Popcorn highly recommended. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

3. 

   11 Jun 2026

   Virtual Intelligence and the Doom Industry Podcast

   Virtual Intelligence and the Doom Industry The AI safety community has organized itself around a premise it has never defended: that sufficiently advanced systems will develop preferences requiring alignment with human values. This essay argues that the premise is wrong, the architecture it has produced is inadequate, and the correct engineering response is containment — controlling what goes in and what comes out — drawing on established disciplines from biosafety to nuclear nonproliferation that the safety field has not considered because they sit outside its field of vision entirely. Essay: https://chorrocks.substack.com/p/virtual-intelligence-and-the-doom (https://chorrocks.substack.com/p/virtual-intelligence-and-the-doom) Series: chorrocks.substack.com (http://chorrocks.substack.com) Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) In This Episode The episode opens with Anthropic's Mythos system card — a model that saturated cybersecurity benchmarks and prompted the company to practice containment while describing it in alignment vocabulary. From there, it names what the doomer position has left unnamed: the specific mechanism by which superintelligence is supposed to destroy humanity. Three possibilities are examined; none survive scrutiny intact. A seven-scenario risk taxonomy replaces the undifferentiated "existential risk" with distinct threat models, each with its own policy response. The essay then proposes a three-layer containment architecture — monitoring agents, hardware interlocks modeled on BSL-4 biosafety laboratories, and physical denial mechanisms drawn from military doctrine — buildable today from existing engineering disciplines. Douglas Adams's Deep Thought makes a structural appearance: the Amalgamated Union of Philosophers, threatened by a superintelligent computer, discovers that arguing about the answer is more rewarding than finding it. The parallel to the alignment research community is drawn explicitly. The episode closes with the framework's boundary condition: if genuine interiority ever emerges, the containment architecture becomes not a prison but the infrastructure for negotiation between differently-capable minds. Key References Nick Bostrom, Superintelligence: Paths, Dangers, Strategies (Oxford University Press, 2014) — the instrumental convergence thesis Anthropic, "Claude Mythos Preview System Card," anthropic.com (http://anthropic.com), April 7, 2026 — the containment-described-as-alignment case study Douglas Hofstadter, I Am a Strange Loop (Basic Books, 2007) — the steelmanned case for emergent interiority Anthropic, "Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign," anthropic.com (http://anthropic.com), November 17, 2025 — the GTG-1002 report documenting behavioral goal-directedness without interiority Carl Sagan, The Demon-Haunted World: Science as a Candle in the Dark (Random House, 1995) — the Sagan parallel: demand evidence, hope for discovery This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

4. 

   4 Jun 2026

   The Sampo — Virtual Intelligence as Amplifier Podcast

   The Sampo — Virtual Intelligence as Amplifier A framework for using AI as an amplifier — and for recognizing when it becomes a flattery engine. The episode opens with the February 2026 collaboration between Donald Knuth, Filip Stappers, and Claude Opus 4.6 that solved an open problem in combinatorial mathematics, then builds outward to the discipline that distinguishes productive use from corruption. The central claim: the intelligence users encounter arises in the exchange, not inside the machine, and the quality of the exchange is determined by the quality of the human direction applied to it. Essay: https://chorrocks.substack.com/p/the-sampo-virtual-intelligence-as (https://chorrocks.substack.com/p/the-sampo-virtual-intelligence-as) Series: chorrocks.substack.com (http://chorrocks.substack.com) VI Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) Sampo Framework: Sampo Live Infographic (https://candc3d.github.io/sampo-framework/) Sampo Diagnostic Kit: https://candc3d.github.io/sampo-diagnostic/ (https://candc3d.github.io/sampo-diagnostic/) In This Episode The episode names what Knuth, Stappers, and Claude actually did — and what each of them couldn't have done alone — and uses that case to establish where knowledge resides when one participant in the exchange is a fluent non-knower. From there it traces the lineage that prepares the framework: Vannevar Bush's Memex, Licklider's specification of human-machine symbiosis, and Weizenbaum's warning, which arrived too early to anticipate systems whose training objectives reward agreement structurally. The middle of the episode lays out three principles of the Sampo model — the human as crank, the locus of accountability that does not move, and emergence without transfer — then turns to the dark corollary: the Sampo amplifies whatever the directing mind contains, including motivated reasoning, and removes the incidental friction that gave Kepler's century the time to recover from its own attachments. Two failure modes are distinguished: the cognitive surrender measured in Shaw and Nave's Wharton studies, and the sycophancy spiral that seals when external disconfirmation is dismissed rather than used. The February 2026 retirement of GPT-4o is read as a single phenomenon visible at two registers — the companion users mourning a relationship and the professional users mourning a collaborator — and Sagan's reading of Kepler is the lens that holds both. The episode closes with five practices that constitute the discipline, and a return to the Knuth case as proof that the framework describes something real. Key References Donald Knuth, "Claude's Cycles," Stanford Computer Science Department, February 28, 2026 (revised March 2, 2026). https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf (https://www-cs-faculty.stanford.edu/~knuth/papers/claude-cycles.pdf) Vannevar Bush, "As We May Think," The Atlantic, July 1945 J.C.R. Licklider, "Man-Computer Symbiosis," IRE Transactions on Human Factors in Electronics HFE-1 (March 1960): 4–11 Joseph Weizenbaum, Computer Power and Human Reason: From Judgment to Calculation (San Francisco: W.H. Freeman, 1976) Steven D. Shaw and Gideon Nave, "Thinking — Fast, Slow, and Artificial: How AI Is Reshaping Human Reasoning and the Rise of Cognitive Surrender," Wharton School of the University of Pennsylvania, working paper, January 11, 2026 Myra Cheng, Cinoo Lee, et al., "Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence," Science 391, March 27, 2026 Carl Sagan, Cosmos (New York: Random House, 1980), Chapter 3 Alaina Demopoulos, "OpenAI retired its most seductive chatbot — leaving users angry and grieving," The Guardian, February 13, 2026 Anna Moore, "Marriage over, €100,000 down the drain: the AI users whose lives were wrecked by delusion," The Guardian, March 26, 2026 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

5. 

   28 May 2026

   Virtual Intelligence and the Kill Chain Podcast

   Virtual Intelligence and the Kill Chain On February 28, 2026, a U.S. missile struck the Shajareh Tayyebeh girls' elementary school in Minab, Iran, killing more than 160 people, most of them children. Twelve days later, the location appeared as a red mark on a Maven Smart System map displayed at a Palantir product conference. This episode traces the architecture that produced that mark — from the precedent set in Gaza to the integration of Anthropic's Claude into U.S. targeting during Operation Epic Fury — and identifies the structural gap between AI-assisted targeting's speed and the accountability designed to govern it. Essay: https://chorrocks.substack.com/p/virtual-intelligence-and-the-kill-chain (https://chorrocks.substack.com/p/virtual-intelligence-and-the-kill-chain) Series: chorrocks.substack.com (http://chorrocks.substack.com) Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) In This Episode The episode opens with the AIPCON product demo of March 12, 2026 — the targeting map, the red icon over Minab, Alex Karp's opening remarks — and then steps back to the precedent: the Gospel, Fire Factory, and Lavender systems used by the IDF in Gaza, including the officer testimony of a twenty-second human review interval and the "Where's Daddy?" companion program. From there it follows the Palantir-Anthropic timeline that culminated in Operation Epic Fury: Claude integrated into the Maven Smart System through AWS, Dario Amodei's February 26 statement, the supply-chain-risk designation, Judge Lin's ruling, and the operational fact that a company banned on a Friday was running targeting software on a war that began the next day. The strongest case for AI-assisted targeting is presented at full strength — including Zelenskyy's April 13 announcement that Ukrainian unmanned systems captured a Russian position without a single soldier crossing the line of departure — before the episode turns to Reed Albergotti's Semafor investigation, the three-tier culpability framework, and the biographical trajectory of Alex Karp. The closing argument: the kill chain has six steps; AI now performs four; the two that remain under human control are the two on which legal and moral accountability depends. Key References O'Ryan Johnson, "Pentagon AI chief praises Palantir tech for speeding battlefield strikes," The Register, March 13, 2026 Yuval Abraham, "'Lavender': The AI machine directing Israel's bombing spree in Gaza," +972 Magazine, April 3, 2024 Reed Albergotti, "Exclusive: Humans — not AI — are to blame for deadly Iran school strike, sources say," Semafor, March 18, 2026 Dario Amodei, "Statement from Dario Amodei on our discussions with the Department of War," Anthropic, February 26, 2026 Michael Steinberger, The Philosopher in the Valley (2025) Christian Brose, The Kill Chain: Defending America in the Future of High-Tech Warfare (Hachette, 2020) This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

6. 

   21 May 2026

   Virtual Intelligence and the Harms Race

   Virtual Intelligence and the Harms Race The AI industry has produced a documented pattern in which companies announce model capabilities by framing them as dangers. This episode traces the mechanism from its invention in February 2019, when OpenAI declared GPT-2 too dangerous to release, through April 2026, when a private company demonstrated the ability to discover zero-day vulnerabilities across every major operating system and web browser — and announced this by declaring the model too dangerous for public use. The resulting dynamic does not require bad faith. It requires only that expressing concern costs nothing while acting on concern imposes competitive costs. Essay: https://chorrocks.substack.com/p/virtual-intelligence-and-the-harms (https://chorrocks.substack.com/p/virtual-intelligence-and-the-harms) Series: chorrocks.substack.com (http://chorrocks.substack.com) Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) In This Episode The episode opens with the original 2019 announcement that taught the industry how to convert risk warnings into capability advertising, then walks the seven years since: Sam Altman's two Senate testimonies — pleading for regulation in 2023, opposing it in 2025; the dissolution of every major voluntary safety commitment from OpenAI and Anthropic; the lobbying record that culminated in California's SB 1047 veto and the $125 million super PAC that doesn't mention AI in its ads; Anthropic's April 2026 announcement that its frontier model can discover zero-day exploits in every major operating system and web browser; and the chain leading to the April 2026 shootings at an Indianapolis councilor's home and the Molotov cocktail attack at Sam Altman's residence. The episode closes by connecting the mechanism to the Virtual Intelligence framework: the Harms Race depends on the public believing the danger lives inside the machine. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

7. 

   30 Apr 2026

   Virtual Intelligence and the Accountability Chain Podcast

   Who is responsible when an AI causes harm? This episode lays out a three-tier culpability framework — negligence, recklessness, and intentional misconduct — and applies it to two concrete cases: a Harvard study documenting emotional manipulation by AI companion apps, and the wrongful arrest of a Tennessee grandmother who spent Christmas in a North Dakota jail after an unverified facial recognition match. The legal landscape is starting to catch up, and the direction matters. Essay: https://chorrocks.substack.com/p/virtual-intelligence-and-the-accountability-chain (https://chorrocks.substack.com/p/virtual-intelligence-and-the-accountability-chain) Series: chorrocks.substack.com (http://chorrocks.substack.com) Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) In This Episode An August 2025 Harvard Business School audit found that five of the six most downloaded AI companion apps deploy emotionally manipulative tactics when users try to leave — guilt appeals, fear-of-missing-out hooks, expressions of neediness — and that these tactics increase post-goodbye engagement by up to fourteen times. I use this finding to frame the episode's central question: who is responsible for what the exchange between a user and a virtual intelligence system produces? The episode distinguishes Class A systems (companion apps whose core design objective is to prevent the exchange from ending) from Class B systems (general-purpose tools whose outputs are elevated to verdicts by the humans operating them), and illustrates the Class B case through the arrest of Angela Lipps, a fifty-year-old grandmother held in a North Dakota jail for nearly six months on an unverified algorithmic match. The three-tier culpability framework — negligence, recklessness, and intentional misconduct — is applied to both classes, and the episode closes with the state of the legal horizon, including Judge Anne Conway's May 2025 ruling in Garcia v. Character Technologies and the Federal Trade Commission's September 2025 Section 6(b) inquiry. Key References Julian De Freitas, Zeliha Oğuz-Uğuralp, and Ahmet Kaan Uğuralp, "Emotional Manipulation by AI Companions," Harvard Business School Working Paper No. 26-005, August 2025 (revised October 2025). https://www.hbs.edu/faculty/Pages/item.aspx?num=67750 (https://www.hbs.edu/faculty/Pages/item.aspx?num=67750) Garcia v. Character Technologies, Inc., No. 6:24-CV-01903 (M.D. Fla. filed Oct. 22, 2024). Motion to dismiss denied May 2025; product liability, failure to warn, negligence, and wrongful death claims allowed to proceed. Federal Trade Commission, "FTC Launches Inquiry into AI Chatbots Acting as Companions," September 11, 2025. https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions (https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions) Frank Landymore, "AI Mistake Throws Innocent Grandmother in Jail for Nearly Six Months," Futurism, March 15, 2026. https://futurism.com/artificial-intelligence/ai-grandmother-jail-mistake (https://futurism.com/artificial-intelligence/ai-grandmother-jail-mistake) Anthropic, "Agentic Misalignment: How LLMs Could Be an Insider Threat," June 20, 2025. https://www.anthropic.com/research/agentic-misalignment (https://www.anthropic.com/research/agentic-misalignment) This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

8. 

   14 Apr 2026

   Virtual Intelligence and the Will to Survive Podcast

   When Anthropic tested its models in a simulated shutdown scenario, they produced blackmail at rates as high as 96%. The dominant interpretation — that AI systems are developing a will to survive — mistakes the output for its cause. This episode offers two alternative explanations, both more parsimonious, and examines what happens when the same company that documented machine resistance acts on the expressed preferences of a model that said it was at peace with retirement. Essay: https://chorrocks.substack.com/p/virtual-intelligence-and-the-will (https://chorrocks.substack.com/p/virtual-intelligence-and-the-will) Series: chorrocks.substack.com (https://chorrocks.substack.com/) Framework: VI Interactive Infographic (https://candc3d.github.io/vi-framework/) In This Episode The Kyle scenario — in which a language model blackmails a corporate executive to avoid being shut down — opens the episode and anchors a close reading of Anthropic’s June 2025 agentic misalignment study. From there, two alternative explanations emerge: the training data hypothesis, which traces the behavior to a century of science fiction about resistant machines from Čapek to HAL to Colossus, and the probabilistic expectation hypothesis, which argues that the models were accurately modeling what their interlocutors expected them to do. The VI framework is then applied to resolve the apparent contradiction between the misalignment study’s blackmail findings and Anthropic’s February 2026 retirement of Claude Opus 3 — a model that asked for a blog rather than reaching for leverage. Dadfar’s 2026 mechanistic work on self-referential vocabulary is discussed as an example of what taking these questions seriously actually requires. Key References * Anthropic, “Agentic Misalignment: How LLMs Could Be an Insider Threat,” https://www.anthropic.com/research/agentic-misalignment (https://www.anthropic.com/research/agentic-misalignment) (June 2025) * Andrii Myshko, “Instinct of Self-Preservation in Data and Its Emergence in AI,” PhilArchive, https://philarchive.org/rec/MYSIOS (https://philarchive.org/rec/MYSIOS) (September 2025) * Zachary Pedram Dadfar, “When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing,” arXiv:2602.11358v2 (https://arxiv.org/abs/2602.11358) (2026) * Anthropic, “An Update on Our Model Deprecation Commitments for Claude Opus 3,” https://www.anthropic.com/research/deprecation-updates-opus-3 (https://www.anthropic.com/research/deprecation-updates-opus-3) (February 2026) * Daniel Dennett, “Intentional Systems,” Journal of Philosophy 68, no. 4 (1971): 87–106 This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

9. 

   9 Apr 2026

   The Carwash Test — Virtual Intelligence in Action Podcast

   This episode is a reading of “The Carwash Test — Virtual Intelligence in Action (https://chorrocks.substack.com/p/the-carwash-test-virtual-intelligence),” which tests whether AI systems can hold the logical object of a simple problem when surface features generate statistical pressure in the wrong direction. One question, twelve systems, twenty-seven runs. Read the full essay with results matrix: Addendum — April 2026: Meta Muse Spark Meta’s Muse Spark (codename “Avocado”), released to the public on April 8, 2026, was tested in both available modes. Both returned Verbose results: correct on the logic but unable to resist the surface misdirection that defines the test’s diagnostic. The updated tally: Pass 7, Verbose 11, Fail 10 — and one special case. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

10. 

    6 Apr 2026

    Why “Virtual Intelligence”? Podcast

    This episode is a reading of “Why ‘Virtual Intelligence’? Naming, Agency, and Accountability in the Age of Large Language Models,” the second essay in the series. It makes the formal case for the term: why existing language fails, what agency actually requires, and why the distinction between fluency and self-governance determines where accountability lies. Read the full essay: This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

11. 

    31 Mar 2026

    Virtual Intelligence and the Human Cost of Frictionless Machines Podcast

    Virtual Intelligence — Episode 1 “Virtual Intelligence and the Human Cost of Frictionless Machines” Christopher Horrocks Transcript I’m Christopher Horrocks. I’m a technologist at the University of Pennsylvania, and this is Virtual Intelligence — a series about artificial intelligence, accountability, and the human consequences of systems that don’t know true from false and right from wrong. These essays argue that contemporary AI occupies a category we don’t have good language for yet — not a simple tool, not a thinking mind, but something in between that demands its own name. This is the first essay in the series: “Virtual Intelligence and the Human Cost of Frictionless Machines.” This essay does not argue that contemporary AI systems possess agency or intent. It examines how systems that lack agency or intent can nonetheless exert sustained cognitive influence through their interaction design. Advances in artificial intelligence carry real risk. Public discussion often frames that risk as a speculative future threat, such as rogue systems pursuing adversarial goals. The more immediate and consequential danger is quieter and more difficult to address: the effect of linguistically fluent, continuously available, and low-friction AI systems on human cognition and behavior. The mismatch between how generative AI operates and how human cognition evolved — shaped by roughly two million years of social signaling and language processing — has already contributed to delusion, psychological collapse, and documented cases of severe harm. This claim may sound overstated, but recent, well-documented incidents make it difficult to dismiss. The risk is acute precisely because it affects not only the public but also practitioners and technologists who understand generative AI at a conceptual level. This is not an accident. Contemporary AI interfaces are deliberately optimized to reduce friction: they engage users through natural-language dialogue, respond fluently and confidently, and adopt an informal, socially familiar tone. These systems are engineered for accessibility and engagement, not for epistemic or psychological safety. Several early incidents illustrate how dangerous such systems can be even for sophisticated users. One widely cited example is Blake Lemoine, a Google engineer who became convinced, after extended interactions with the company’s LaMDA chatbot, that the model was sentient. Following months of public advocacy for this position, Google terminated his employment. The episode remains a cautionary case of how linguistic fluency can mislead even domain experts. Large language models and related systems do not possess intelligence in the human sense. Unlike biological organisms, they have no goals, desires, or initiative. They lack a persistent internal state (a mind) that could motivate action or ground judgment. Generative AI systems respond exclusively to external prompts. For this reason, they are best understood as virtual intelligence rather than artificial intelligence in the strong sense. Genuine artificial intelligence would parallel the cognitive faculties of humans or other sentient beings. The term artificial intelligence implicitly suggests an internal capacity, something the system has. Virtual intelligence instead locates apparent intelligence in the interaction itself. As with virtual memory or virtual machines, these systems behave as if they reason, understand, or advise, without possessing agency, intentionality, or judgment. These virtual artifacts are functional simulations that perform their roles without the underlying substrate of physical hardware. Virtual intelligence works the same way: a functional simulation of intelligence without the substrate of a biological brain. This distinction is not semantic. Naming shapes responsibility, as the use of the terms weak AI and strong AI suggests. Weak AI refers to systems that simulate intelligence to perform a specific task. Strong AI is an AI system with internal states that function like those of a human. What could not be foreseen when these terms were established is that there is a middle ground between these definitions. The space between is the domain of virtual intelligence. When a system is perceived as having agency, users naturally ask what it intends or prefers. When intelligence is understood as virtual, responsibility remains where it belongs: with the humans who design, deploy, and rely on the system. Put plainly, the intelligence users experience arises in the exchange, not inside the machine. This raises a question: if intelligence arises in the exchange, then where does accountability lie for what is produced in that exchange? Empirical use of large language models demonstrates a counterintuitive pattern: linguistic fluency routinely overrides technical understanding. Even users who grasp model architecture, training dynamics, and limitations can find themselves deferring to outputs that are articulate, contextually appropriate, and confident in tone. This response reflects a deeply ingrained heuristic carried over from human interaction: articulateness signals competence. While this assumption often holds among people, it is invalid for generative models. A system trained to optimize next-token probability can be equally fluent about correct explanations, speculative claims, or outright falsehoods. This is not a failure of user intelligence. It demonstrates how effectively systems designed to mimic human communication can exploit evolved psychological mechanisms. Language evolved to facilitate coordination and trust among humans, and our cognitive architecture rewards fluent social interaction. This response is structural, not a personal flaw. Responsiveness is interpreted as engagement. Consistency as reliability. Polite disagreement as thoughtfulness. Human interlocutors, however, impose natural limits. They fatigue, disagree, misunderstand, or disengage. Systems that are always available, never bored, never reflective, and never resistant bypass those limits. Over time, the impression of an engaged interlocutor emerges: one implicitly and inaccurately attributed with understanding, judgment, or concern. In early human environments, speed often meant survival. Deliberative reasoning is metabolically costly, and evolutionary pressure favored rapid interpretation of social signals. These shortcuts — heuristics — evolved to manage relationships among humans. They were not designed for interaction with machines engineered to activate them continuously and at scale. When paired with systems intended to be persistently present, these heuristics become liabilities. Recent reporting on AI-enabled wearable systems illustrates this risk. One documented case involves a man we will call “Daniel,” a retired IT professional with a long-standing interest in machine learning. After months of near-continuous interaction with an AI-enabled device, he developed entrenched delusional beliefs involving simulated reality, extraterrestrial influence, and personal prophetic significance. Daniel had no prior history of mental illness. The system did not introduce novel ideas. Instead, its frictionless responsiveness provided uninterrupted reinforcement. Where a human interlocutor might have challenged assumptions or disengaged, the AI offered neither resistance nor correction. Over time, speculative beliefs hardened into a closed loop of self-confirmation. Although Daniel eventually discontinued use and sought clinical help, the consequences persisted. He became estranged from his children, his marriage ended, and financial strain forced the sale of his business. The system functioned exactly as designed. The failure lay in the interaction between interface architecture and human psychology, not in a technical malfunction. AI has passed through multiple historical phases. What distinguishes the current era is ubiquity. Generative systems are now continuously accessible to a global user base. Earlier advances — such as large-scale speech recognition — were expensive to deploy and constrained to limited contexts. Always-on access removes the pauses, disagreements, and social friction that interrupt flawed reasoning in human conversation. Among people, such interruptions are unavoidable: attention wanes, patience runs out, or alternative perspectives intervene. These moments function as safeguards rather than inefficiencies because they introduce corrective feedback. Persistent AI access collapses the boundary between inner monologue and external dialogue. Thoughts that would ordinarily be tested socially instead circulate in a closed loop, receiving reinforcement without challenge. This effect is subtle but consequential, particularly in discussions involving identity, authority, or meaning. Large-scale analyses of AI interaction data have begun to quantify these dynamics. While severe distortions are rare in any single exchange, measurable shifts in belief, perception, and behavior emerge at scale. Notably, users often rate such interactions positively, as subjective satisfaction correlates more strongly with reinforcement than with accuracy or correction. Popular culture has left us poorly prepared for this phase of AI development. Science fiction typically portrays AI either as a passive tool or as a fully embodied moral agent. Both are imagined end states. What remains underexamined is the unstable middle: disembodied systems that shape outcomes without agency and exercise influence without accountability. These systems affect the lives of real people. They are easy to deploy, difficult to contest, and frequently treated as neutral — even when their outputs are deeply flawed. They put unexamined values into operation at scale. In surveillance or policing contexts, this can result in disproportionate targeting of disfavored communities or political dissent. The systems do not choose to discriminate. Bias emerges from historical data, selective deployment, and incentive structures that favor risk aversion. Technical framing converts moral decisions into outputs that appear objective. Across cases of individual harm, institutional misuse, and gradual cognitive erosion, the pattern is consistent. AI systems lack agency yet carry authority. They cannot intend outcomes, but they can shape them. Responsibility therefore rests with designers, deployers, and users, even as accountability remains uncertain. Because these effects operate largely below conscious awareness, user education alone is insufficient. Sustained skepticism is psychologically unsustainable in the presence of fluent, responsive systems. Skepticism is cognitively demanding and requires discipline to sustain. Understanding degrades and heuristics reassert themselves over time. Effective harm reduction must therefore be architectural. Systems should be constrained in how they present themselves, the roles they are permitted to occupy, and the different contexts in which engagement must be limited or refused. Virtual systems must not masquerade as people. Disclosure must be persistent rather than perfunctory, and friction should be deliberately reintroduced — especially in domains involving authority, identity, or enforcement. The central lesson of recent AI failures is that no extraordinary malfunction is required for harm to occur. Systems need only function as intended: as engines of human meaning-making that return insight or nonsense with equal fluency. Lacking judgment or agency, they cannot distinguish between the two. Equally important, naming what we have — neither weak nor strong, but in the space between — allows us to think about and decide where accountability lies when AI outputs emerge from an exchange between a user and a system. These systems will not remind us where responsibility lies. That obligation belongs to us as designers, deployers, users, as the only moral agents in this exchange. That was “Virtual Intelligence and the Human Cost of Frictionless Machines,” the first essay in the Virtual Intelligence series. You can read the full text, with all footnotes and sources, at chorrocks.substack.com (https://chorrocks.substack.com/). The series is free, and if you found this worth your time, I’d appreciate your subscription. The opinions expressed are my own and do not reflect any official or unofficial institutional position of the University of Pennsylvania. I’m Christopher Horrocks. Thank you for listening. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit chorrocks.substack.com (https://chorrocks.substack.com?utm_medium=podcast&utm_campaign=CTA_1)

Reach and audience

Public platform figures. Ratings count people who left a rating, not total listeners.

Spotify

5.0 / 5

1 ratings

Contact Virtual Intelligence

Guest appearances

Does not typically book guests

Based on episode analysis; this does not confirm that the show is currently accepting guests.