Show Notes
Can the AI "alignment problem" actually be defined, or is it fundamentally flawed from the start? In this episode of Philosophy, Programs, and Prompts, Casey Hart( @OntologyExplained ) and Carl Brown( @InternetOfBugs ) set aside bad-faith doomer rhetoric and corporate marketing moats to examine the core arguments behind AI alignment in good faith.Casey argues that grappling with alignment acts as a conceptual gateway into foundational questions of ethics, paternalism, and what constitutes the "good life". Meanwhile, Carl pushes back from an engineering and practical standpoint, arguing that true alignment with humanity—which is notoriously misaligned with itself—is a distraction from pressing real-world issues like environmental degradation, social media algorithms, legal accountability, and building practical systems we can control.Together, they dissect the mechanics of Coherent Extrapolated Volition (CEV), Isaac Asimov's Three Laws of Robotics, Nick Bostrom's orthogonality thesis, and the practical limits of guardrails, kill switches, and machine interpretability.---Links & Resources Mentioned:// Eliezer Yudkowsky – Coherent Extrapolated Volition (2004):https://intelligence.org/files/CEV.pdf// Nick Bostrom – Superintelligence: Paths, Dangers, Strategies (2014): Referenced for the Orthogonality Thesis (01:06:44) and the Paperclip Maximizer https://global.oup.com/academic/product/superintelligence-9780199678112?cc=us&lang=enhttps://rudyct.com/ai/Superintelligence%20Paths,%20Dangers,%20Strategies%20%28Nick%20Bostrom%292014.pdf---CHAPTERS:00:00 - Cold Open: Why Computer Science Ignores History00:28 - Welcome to Philosophy, Programs, and Prompts01:53 - The Thesis: Alignment as a Gateway Drug to Philosophy03:59 - Four Ways to Frame the Alignment Problem06:05 - Paternalism, Ethics, and Defining "The Good Life"09:50 - The Computer Science Habit of Working in a Vacuum15:06 - Coherent Extrapolated Volition & Paternalistic Genies20:20 - Eternal Recurrence and Avoiding an Algorithmic Rut25:30 - Defining Alignment: Goals, Intentions, and Needs31:40 - Can We Be Aligned with Non-Living Artifacts?36:00 - Internal Self-Misalignment and Humanity's Fractured Goals47:20 - A Taxonomy of Proposed Solutions: Moratoriums & Off Switches54:20 - Asimov's Three Laws: Control Constraints vs. True Alignment01:00:50 - Encoding Morality: Utilitarian Calculators & Perfect Parents01:05:14 - Eliminativism, Orthogonality, and Decision Theory Flaws01:09:30 - Machine Interpretability: What Science Can Actually Do01:13:00 - Marketing Benchmarks vs. Real-World Harm#AIAlignment #ArtificialIntelligence #PhilosophyOfTech #AISafety #TechPodcast #ComputerScience #PhilosophyProgramsAndPrompts
More from Philosophy, Programs, and Prompts
In this episode of Philosophy Programs and Prompts, Casey Hart ( @OntologyExplained ) and Carl Brown( @InternetOfBugs ) take a deep dive into the "R-E-A-L" half of the TESCREAL acronym (coined by Émile P. Torres and Timnit Gebru): Rationalism, Effective Altruism, and Longtermism.Starting with the intellectual roots of online rationalism—from LessWrong and Eliezer Yudkowsky to the Center for Applied Rationality (CFAR)—the conversation traces how reasonable ideas about cognitive bias, decision theory, and clear thinking evolved into a distinct subculture closely tied to Silicon Valley and artificial intelligence. Next, Casey and Carl examine Effective Altruism, tracing its trajectory from Peter Singer’s thought experiments and GiveWell’s charity evaluations to the "earning to give" ethos that paved the way for figures like Sam Bankman-Fried (FTX). Finally, they analyze Longtermism, addressing how utilitarian maximization can lead to thought experiments involving 10^20 virtual human beings, existential risk prioritizations (from pandemics to AI extinction), and moral blind spots regarding people living in the present.Subscribe for more grounded, hype-free takes from two real humans who know the tech!More info about Casey:https://CaseyHart.com/More info about Carl:https://InternetOfBugs.com/// Sources:* "Famine, Affluence, and Morality" (1972) by Peter Singerhttps://www.abebooks.com/9780190219208/Famine-Affluence-Morality-Singer-Peter-0190219203/plp * "The Most Good You Can Do" (2015) by Peter Singerhttps://www.abebooks.com/Good-Effective-Altruism-Changing-Ideas-Living/32491397359/bd * "Winners Take All: The Elite Charade of Changing the World" (2018) by Anand Giridharadashttps://www.abebooks.com/Winners-Take-Elite-Charade-Changing-World/32517761692/bd * The TESCREAL Acronym — Coined by Dr. Timnit Gebru and Émile P. Torreshttps://en.wikipedia.org/wiki/TESCREAL * Cal Newport interview with Emile Torreshttps://www.youtube.com/watch?v=jt508VjX2H800:00 - Introduction: Unpacking the "REAL" in TESCREAL02:22 - Rationalism: From Descartes and Empiricism to 2009 Online Culture06:10 - LessWrong, Slate Star Codex, and the Rationalist Canon07:44 - The Methods of Rationality & CFAR Culture13:56 - Why Rationalists Prioritize Artificial Intelligence18:02 - Effective Altruism: Peter Singer and "The Drowning Child"21:28 - Giving What We Can, GiveWell, and Systematic Charity26:38 - "Earning to Give," Silicon Valley, and Sam Bankman-Fried32:07 - Winners Take All: Systemic Impact vs. Individual Charity37:57 - Longtermism: Future Persons, Digital Humans, and Utilitarian Traps43:02 - Utility Monsters, Experience Machines, and Moral Edge Cases47:20 - Shifting Priorities: How 80,000 Hours Pivoted from Pandemics to AI53:48 - Non-Falsifiability, Religious Parallels, and Pascal's Wager01:00:13 - The Silicon Valley Social Web: Anthropic, Polycules, and Interconnection01:04:43 - Final Thoughts: Grounding Big Ideas in the Real World#Philosophy #TESCREAL #EffectiveAltruism #Rationalism #Longtermism #AIEthics #ArtificialIntelligence #LessWrong #PeterSinger #Podcast
In this episode of *Philosophy, Programs, and Prompts*, hosts Casey Hart ( @OntologyExplained ) and Carl Brown ( @InternetOfBugs ) dive into the strange intersection of theology, tech hype, and AI ethics. Anthropic recently updated its terms of service to forbid "sustained and needless abusive or cruel behavior" toward its models—effectively banning users from "torturing" Claude. But are large language models actually capable of feeling pain, or are we just projecting humanity into statistical matrix math?Casey and Carl explore Anthropic's lobbying efforts aimed at the Catholic Church and other religious leaders, the Pope’s encyclical addressing machine intelligence and human art, and the philosophical absurdity of two recent papers claiming to map "functional pain" and "well-being" onto LLM vector spaces. Along the way, they unpack a viral GitHub "torture chamber" experiment, why adding a "constipation vector" breaks an LLM just as easily as pain, and why current AI models remain miles away from genuine sentience. https://www.nytimes.com/2026/09/29/us/anthropic-claude-morals-ai.html https://futurism.com/artificial-intelligence/anthropic-lobbying-vatican https://www.forbes.com/sites/siladityaray/2026/09/29/pope-leo-says-ai-safety-fears-are-not-fake-news-and-criticizes-nvidias-billionaire-ceo/https://www.catholicculture.org/commentary/dangerous-myth-conscious-ai/ https://aiweekly.co/alerts/anthropic-lobbied-pope-leo-xivs-advisers-over-ai-consciousness* Dr. Charles Camosy: https://www.youtube.com/live/vhWeyGEplU4* 404 Media: Report https://www.404media.co/someone-torturing-llms-in-a-robot-prison-has-triggered-the-dumbest-debate-in-ai-yet/ https://www.youtube.com/watch?v=aLYyF_nxNh0Papal Encyclical on Artificial Intelligence (May 2026), Section 99: "Magnifica Humanitas" https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html* Pope's Statement on X / Twitter (October 8): https://x.com/Pontifex/status/2105983638147891490 Ren et al. – "AI Well-Being: Measuring and Improving the Functional Pleasure and Pain of AIs" https://www.ai-wellbeing.org/paper.pdf* Tagliabue et al. – "The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It" https://arxiv.org/abs/2609.16247v1* "The Neurogenic Case Against Synthetic Sentience" https://arxiv.org/abs/2301.05397Projects & Tools:* GitHub: The AI "Torture Chamber" steering repository and README rebuttal experiments https://github.com/terrafying/ai-torture-chamber* Anthropic Terms of Service update: Prohibition on "sustained and needless abusive or cruel behavior" https://www.anthropic.com/news/2026-usage-policy-update* Nightshade: Data poisoning tool for image generative modelshttps://arxiv.org/abs/2310.13828 https://nightshade.cs.uchicago.edu/whatis.html* Moltbook: LLM agent social media experiment archives https://www.youtube.com/watch?v=GYfgjYVEYQ000:00:00 - Introduction & Anthropic’s Cruelty Ban00:01:55 - Lobbying Religious Leaders on AI Sentience00:05:20 - Possibility of Machine Consciousness vs. Today's Reality00:07:30 - PR Stunts, Regulatory Capture, and Rabbi Navon’s Dilemma00:15:10 - Why Target the Catholic Church?00:20:40 - The Pope’s Encyclical and the Human Element00:26:00 - Historical Panics and Cold War Parallels00:31:00 - The Ontological Difference in AI Art00:36:15 - Pattern Averaging vs. Transformative Human Art00:44:40 - Paper 1: AI Well-Being and Utility Functions00:50:30 - Moltbook and Human Anthropomorphic Projection00:57:50 - Paper 2: The "Pain Axis" and Vector Direction01:02:40 - The GitHub Torture Chamber & The Constipation Vector01:09:40 - Why Anthropic Actually Cares (Training Data vs. Press)01:14:30 - Aligning Intentions, Sentience Tests, and Outro#AI #ArtificialIntelligence #Philosophy #Anthropic #Claude #MachineLearning #AIEthics #TechPodcast #LLM #Consciousness.
In the wake of high-profile security breaches, frontier lab resignations, and mounting political pressure, the call to "Pause AI" has moved from fringe forums to Capitol Hill. But what does pausing AI actually look like in practice? In this episode of Philosophy Programs and Prompts, Casey Hart( @OntologyExplained ) and Carl Brown( @InternetOfBugs ) break down the tangled definitions, ulterior motives, and policy proposals surrounding the AI pause movement. They explore whether agentic LLMs should be legally classified as malware, interrogate the PR machinery behind AI safety whistleblowers, and evaluate proposals ranging from Dario Amodei’s call to "pace the frontier" to the official PauseAI international treaty framework. Most importantly, they tackle the critical distinction between abstract existential dread and immediate, tangible harms—asking whether regulatory obsession with hypothetical "superintelligence" is blinding lawmakers to the damage happening right now.Subscribe for more grounded, hype-free takes from two real humans who know the tech!More info about Casey:https://CaseyHart.com/More info about Carl:https://InternetOfBugs.com/LINKS MENTIONED IN THIS EPISODE:• Jacob Coxon Resignation Thread & Statementshttps://x.com/hilbertspaess/status/2097476196791709843• Pirate Wires Report on AI Safety PR PitchingEXCLUSIVE: Jacob Coxon Worked With Doomer PR Firm ( https://www.piratewires.com/p/jacob-coxon-worked-with-ai-doomer-firm )• Dario Amodei’s Post on Pacing Frontier AI & METRhttps://darioamodei.com/post/we-must-pace-the-frontier• Casar-Sanders Artificial Superintelligence Bill Summaryhttps://www.sanders.senate.gov/wp-content/uploads/Ban-Artificial-Superintelligence-Act-Release-Summary.pdf • PauseAI Official Proposal & Treaty Frameworkhttps://pauseai.info/proposal• Pod Save America Segment on AI Governancehttps://youtu.be/v_BTZ5lcXrw?si=mJ0XCv2FJyegeKO6&t=1055• PauseAI funded by Future of Life Institute https://pauseai.info/funding • Future of Life Institute says “FLI does not accept donations from Big Tech, or from companies seeking to build artificial general intelligence” but took $10M from Elon Musk https://futureoflife.org/about-us/finances/ • Ezra Klein Vid: https://youtu.be/fjZ90V_JREk?si=HeDsnH5QChEZ2Cxu&t=1706• Zitron Video: https://youtu.be/PENnaW9zGYI?si=I8exuzqiLUWFwGHu&t=597• Ed Dowd piece: https://eddowdbeyondthenarrative.substack.com/p/the-post-labor-day-ai-panic-psyop • Podcast about how (at least some) hacking statutes require “intent”: https://www.youtube.com/watch?v=EQUBMpg1aY4&t=1669s • The U.S. is already using AI in warfare:https://www.washingtonpost.com/technology/2026/03/04/anthropic-ai-iran-campaign/ • Andre Miotti of Control AI on Sky News:https://www.youtube.com/watch?v=odSvjKMRn-I 00:00 - Cold Open: The Super AI Trap00:33 - Slowing Down & The AI Doomer Explosion01:42 - What Does "Pause AI" Actually Mean?03:45 - Is AI a "Normal Technology"?05:07 - AI as Malware: Agentic Intent vs. Real Outcomes12:10 - The Jacob Coxon Resignation & Safety PR15:45 - Follow the Money: Who Funds the AI Doom Movement?23:15 - Dario Amodei, METR, and "Pacing the Frontier"25:06 - Geopolitics and the "China Boogeyman"29:05 - Media Hype and the Failure to Cite Sources32:46 - OpenAI's Security Response & The Incompetence Problem36:47 - Defining the Target: Current LLMs vs. Superintelligence41:21 - The Casar-Sanders Bill & Political Definitions46:22 - Military AI Targeting and Anthropic's Pushback52:05 - Deconstructing the PauseAI Treaty Proposal58:43 - An IAEA for AI vs. Regulating Social Media01:11:14 - Regulatory Capture and Big Tech Moats01:18:00 - The Case for a Pause: Prioritizing Real Harms01:25:30 - Economic Fallout, Force Majeure, and the AI Bubble01:29:30 - Why Corporate Incompetence Trumps Deception01:35:00 - Soundbite Verdicts: Should We Pause AI?01:39:38 - Outro#AISafety #PauseAI #ArtificialIntelligence #TechPolicy #OpenAI #Anthropic #aigovernance
Is artificial intelligence really on track to wipe out humanity by 2050, or are the most prominent doom arguments fundamentally flawed? In this episode of Philosophy Programs and Prompts, Casey Hart( @OntologyExplained ) and Carl Brown( @InternetOfBugs ) break down the core premises behind AI existential risk (x-risk). From Bostrom-style superintelligence scenarios and recursive self-improvement (RSI) to "hockey stick" benchmark progress and the appeal to expert anxiety, Casey and Carl analyze where the logic holds up—and where it jumps the rails. More importantly, they discuss why hyper-focusing on apocalyptic sci-fi scenarios distracts the public and regulators from the very real, immediate harms AI systems are causing right now: cyber vulnerabilities, military reliance on hallucinated outputs, non-consensual deepfakes, and corporate regulatory capture.
In this episode, Casey( @OntologyExplained ) and Carl( @InternetOfBugs ) prep for Casey's upcoming debate on ( @DoomDebates ) with Liron Shapira. They dive into what "P-Doom" actually means, break down the feasibility of recursive self-improvement (RSI), and evaluate whether recent breakthroughs—like AI-assisted Lean proofs for Navier–Stokes—actually point toward artificial superintelligence or just sophisticated, narrow mixtures of experts. Along the way, they analyze the incentives driving public AI panic, the rhetorical tactics of the rationalist community, and why focusing on speculative existential risk often distracts from tangible, real-world AI harms.Subscribe for more grounded, hype-free takes from two real humans who know the tech!More info about Casey:https://CaseyHart.com/More info about Carl:https://InternetOfBugs.com/// References and Resources:// Ed Dows explaining the recent high-profile Anthropic resignation as a PsyOphttps://eddowdbeyondthenarrative.substack.com/p/the-post-labor-day-ai-panic-psyopWatch chatbbots attempt to play chess in hilarious ways @GothamChess ( https://www.youtube.com/playlist?list=PLBRObSmbZluQwBIvxyiWfm_b495AnpzDn )// Mac app Carl mentioned that lets Gemma 4 26B run in 2GN of RAM by only loading the "expert" that's needed at the moment: https://github.com/drumih/turbo-fieldfare/More on decomposing LLMs at @chrishayuk // On the Apple Paper about how chatbots fail problems that look similar to but aren't benchmark problems because they follow the wrong patternshttps://arstechnica.com/ai/2024/10/llms-cant-perform-genuine-logical-reasoning-apple-researchers-suggest/https://arxiv.org/pdf/2410.0522900:00:00 - Cold Open: The Moron or Masochist Rule00:00:11 - Intro: Preserving Sanity Amid AI Hype00:00:46 - Prepping for Doom Debates with Liron Shapira00:03:09 - The Value (and Trap) of Public Debates00:07:37 - What Exactly is P-Doom? Defining the Stakes00:11:45 - Knowing the Audience: Preaching to the Converted vs. the 1%00:15:38 - Finding the Common Ground (and the Common Enemy)00:21:40 - Deconstructing *If Anyone Builds It, Everyone Dies*00:26:22 - The 2050 Extinction Baseline00:29:44 - Recursive Self-Improvement: Hype vs. Reality00:33:41 - AI Takeover vs. Bad Actors with Powerful Tools00:41:43 - The Bayesian Argument Against Doom00:44:13 - A Bug-Centered Epistemology00:47:50 - The Economic Engine: Capital, IPOs, and Regulatory Capture00:52:03 - True Believers and Singularity Evangelists00:56:01 - Manufactured Hype and Coordinated Resignations01:00:00 - Critique of Rationalist Culture and Intelligence Fetishization01:01:43 - The Orthogonality Thesis: Wants vs. Intelligence01:02:40 - Navier–Stokes, Lean Prover, and Mixture of Experts01:03:40 - Goodhart’s Law, P-Hacking, and Jagged Intelligence01:04:46 - Extraordinary Claims Require Extraordinary Evidence01:05:15 - Post-Show: Podcasting Logistics and Editing Legos#AIDoom #ArtificialIntelligence #AISafety #TechDebate #MachineLearning #PhilosophyProgramsAndPrompts #DoomDebates