Future of Life Institute Podcast

Adam Gleave is co-founder and CEO of FAR.AI. In this cross-post from The Cognitive Revolution Podcast, he joins to discuss post-AGI scenarios and AI safety challenges. The conversation explores his three-tier framework for AI capabilities, gradual disempowerment concerns, defense-in-depth security, and research on training less deceptive models. Topics include timelines, interpretability limitations, scalable oversight techniques, and FAR.AI’s vertically integrated approach spanning technical research, policy advocacy, and field-building.LINKS:Adam Gleave - https://www.gleave.meFAR.AI - https://www.far.aiThe Cognitive Revolution Podcast - https://www.cognitiverevolution.aiPRODUCED BY:https://aipodcast.ingCHAPTERS:(00:00) A Positive Post-AGI Vision(10:07) Surviving Gradual Disempowerment(16:34) Defining Powerful AIs(27:02) Solving Continual Learning(35:49) The Just-in-Time Safety Problem(42:14) Can Defense-in-Depth Work?(49:18) Fixing Alignment Problems(58:03) Safer Training Formulas(01:02:24) The Role of Interpretability(01:09:25) FAR.AI's Vertically Integrated Approach(01:14:14) Hiring at FAR.AI(01:16:02) The Future of GovernanceSOCIAL LINKS:Website: https://podcast.futureoflife.orgTwitter (FLI): https://x.com/FLI_orgTwitter (Gus): https://x.com/gusdockerLinkedIn: https://www.linkedin.com/company/future-of-life-institute/YouTube: https://www.youtube.com/channel/UC-rCCy3FQ-GItDimSR9lhzw/Apple: https://geo.itunes.apple.com/us/podcast/id1170991978Spotify: https://open.spotify.com/show/2Op1WO3gwVwCrYHg4eoGyP

Beatrice works at the Foresight Institute running their Existential Hope program. She joins the podcast to discuss the AI pathways project, which explores two alternative scenarios to the default race toward AGI. We examine tool AI, which prioritizes human oversight and democratic control, and d/acc, which emphasizes decentralized, defensive development. The conversation covers trade-offs between safety and speed, how these pathways could be combined, and what different stakeholders can do to steer toward more positive AI futures.LINKS:AI Pathways - https://ai-pathways.existentialhope.comBeatrice Erkers - https://www.existentialhope.com/team/beatrice-erkersCHAPTERS:(00:00) Episode Preview(01:10) Introduction and Background(05:40) AI Pathways Project(11:10) Defining Tool AI(17:40) Tool AI Benefits(23:10) D/acc Pathway Explained(29:10) Decentralization Trade-offs(35:10) Combining Both Pathways(40:10) Uncertainties and Concerns(45:10) Future Evolution(01:01:21) Funding PilotsPRODUCED BY:https://aipodcast.ingSOCIAL LINKS:Website: https://podcast.futureoflife.orgTwitter (FLI): https://x.com/FLI_orgTwitter (Gus): https://x.com/gusdockerLinkedIn: https://www.linkedin.com/company/future-of-life-institute/YouTube: https://www.youtube.com/channel/UC-rCCy3FQ-GItDimSR9lhzw/Apple: https://geo.itunes.apple.com/us/podcast/id1170991978Spotify: https://open.spotify.com/show/2Op1WO3gwVwCrYHg4eoGyP

Nate Soares is president of the Machine Intelligence Research Institute. He joins the podcast to discuss his new book "If Anyone Builds It, Everyone Dies," co-authored with Eliezer Yudkowsky. We explore why current AI systems are "grown not crafted," making them unpredictable and difficult to control. The conversation covers threshold effects in intelligence, why computer security analogies suggest AI alignment is currently nearly impossible, and why we don't get retries with superintelligence. Soares argues for an international ban on AI research toward superintelligence.LINKS:If Anyone Builds It, Everyone Dies - https://ifanyonebuildsit.comMachine Intelligence Research Institute -  https://intelligence.orgNate Soares - https://intelligence.org/team/nate-soares/PRODUCED BY:https://aipodcast.ingCHAPTERS:(00:00) Episode Preview(01:05) Introduction and Book Discussion(03:34) Psychology of AI Alarmism(07:52) Intelligence Threshold Effects(11:38) Growing vs Crafting AI(18:23) Illusion of AI Control(26:45) Why Iteration Won't Work(34:35) The No Retries Problem(38:22) Computer Security Lessons(49:13) The Cursed Problem(59:32) Multiple Curses and Complications(01:09:44) AI's Infrastructure Advantage(01:16:26) Grading Humanity's Response(01:22:55) Time Needed for Solutions(01:32:07) International Ban NecessitySOCIAL LINKS:Website: https://podcast.futureoflife.orgTwitter (FLI): https://x.com/FLI_orgTwitter (Gus): https://x.com/gusdockerLinkedIn: https://www.linkedin.com/company/future-of-life-institute/YouTube: https://www.youtube.com/channel/UC-rCCy3FQ-GItDimSR9lhzw/Apple: https://geo.itunes.apple.com/us/podcast/id1170991978Spotify: https://open.spotify.com/show/2Op1WO3gwVwCrYHg4eoGyP

Luke Drago is the co-founder of Workshop Labs and co-author of the essay series "The Intelligence Curse". The essay series explores what happens if AI becomes the dominant factor of production thereby reducing incentives to invest in people. We explore pyramid replacement in firms, economic warning signs to monitor, automation barriers like tacit knowledge, privacy risks in AI training, and tensions between centralized AI safety and democratization. Luke discusses Workshop Labs' privacy-preserving approach and advises taking career risks during this technological transition.  "The Intelligence Curse" essay series by Luke Drago & Rudolf Laine: https://intelligence-curse.ai/ Luke's Substack: https://lukedrago.substack.com/ Workshop Labs: https://workshoplabs.ai/ CHAPTERS: (00:00) Episode Preview(00:55) Intelligence Curse Introduction(02:55) AI vs Historical Technology(07:22) Economic Metrics and Indicators(11:23) Pyramid Replacement Theory(17:28) Human Judgment and Taste(22:25) Data Privacy and Control(28:55) Dystopian Economic Scenario(35:04) Resource Curse Lessons(39:57) Culture vs Economic Forces(47:15) Open Source AI Debate(54:37) Corporate Mission Evolution(59:07) AI Alignment and Loyalty(01:05:56) Moonshots and Career Advice

Basil Halperin is an assistant professor of economics at the University of Virginia. He joins the podcast to discuss what economic indicators reveal about AI timelines. We explore why interest rates might rise if markets expect transformative AI, the gap between strong AI benchmarks and limited economic effects, and bottlenecks to AI-driven growth. We also cover market efficiency, automated AI research, and how financial markets may signal progress. Basil's essay on "Transformative AI, existential risk, and real interest rates": https://basilhalperin.com/papers/agi_emh.pdf Read more about Basil's work here: https://basilhalperin.com/CHAPTERS:(00:00) Episode Preview(00:49) Introduction and Background(05:19) Efficient Market Hypothesis Explained(10:34) Markets and Low Probability Events(16:09) Information Diffusion on Wall Street(24:34) Stock Prices vs Interest Rates(28:47) New Goods Counter-Argument(40:41) Why Focus on Interest Rates(45:00) AI Secrecy and Market Efficiency(50:52) Short Timeline Disagreements(55:13) Wealth Concentration Effects(01:01:55) Alternative Economic Indicators(01:12:47) Benchmarks vs Economic Impact(01:25:17) Open Research QuestionsSOCIAL LINKS:Website: https://future-of-life-institute-podcast.aipodcast.ingTwitter (FLI): https://x.com/FLI_orgTwitter (Gus): https://x.com/gusdockerLinkedIn: https://www.linkedin.com/company/future-of-life-institute/YouTube: https://www.youtube.com/channel/UC-rCCy3FQ-GItDimSR9lhzw/Apple Podcasts: https://geo.itunes.apple.com/us/podcast/id1170991978Spotify: https://open.spotify.com/show/2Op1WO3gwVwCrYHg4eoGyPPRODUCED BY: https://aipodcast.ing

Esben Kran joins the podcast to discuss why securing AGI requires more than traditional cybersecurity, exploring new attack surfaces, adaptive malware, and the societal shifts needed for resilient defenses. We cover protocols for safe agent communication, oversight without surveillance, and distributed safety models across companies and governments.   Learn more about Esben's work at: https://blog.kran.ai  00:00 – Intro and preview 01:13 – AGI security vs traditional cybersecurity 02:36 – Rebuilding societal infrastructure for embedded security 03:33 – Sentware: adaptive, self-improving malware 04:59 – New attack surfaces 05:38 – Social media as misaligned AI 06:46 – Personal vs societal defenses 09:13 – Why private companies underinvest in security 13:01 – Security as the foundation for any AI deployment 14:15 – Oversight without a surveillance state 17:19 – Protocols for safe agent communication 20:25 – The expensive internet hypothesis 23:30 – Distributed safety for companies and governments 28:20 – Cloudflare’s “agent labyrinth” example 31:08 – Positive vision for distributed security 33:49 – Human value when labor is automated 41:19 – Encoding law for machines: contracts and enforcement 44:36 – DarkBench: detecting manipulative LLM behavior 55:22 – The AGI endgame: default path vs designed future 57:37 – Powerful tool AI 01:09:55 – Fast takeoff risk 01:16:09 – Realistic optimism

Benjamin Todd joins the podcast to discuss how reasoning models changed AI, why agents may be next, where progress could stall, and what a self-improvement feedback loop in AI might mean for the economy and society. We explore concrete timelines (through 2030), compute and power bottlenecks, and the odds of an industrial explosion. We end by discussing how people can personally prepare for AGI: networks, skills, saving/investing, resilience, citizenship, and information hygiene.  Follow Benjamin's work at: https://benjamintodd.substack.com  Timestamps: 00:00 What are reasoning models?  04:04 Reinforcement learning supercharges reasoning 05:06 Reasoning models vs. agents 10:04 Economic impact of automated math/code 12:14 Compute as a bottleneck 15:20 Shift from giant pre-training to post-training/agents 17:02 Three feedback loops: algorithms, chips, robots 20:33 How fast could an algorithmic loop run? 22:03 Chip design and production acceleration 23:42 Industrial/robotics loop and growth dynamics 29:52 Society’s slow reaction; “warning shots” 33:03 Robotics: software and hardware bottlenecks 35:05 Scaling robot production 38:12 Robots at ~$0.20/hour?  43:13 Regulation and humans-in-the-loop 49:06 Personal prep: why it still matters 52:04 Build an information network 55:01 Save more money 58:58 Land, real estate, and scarcity in an AI world 01:02:15 Valuable skills: get close to AI, or far from it 01:06:49 Fame, relationships, citizenship 01:10:01 Redistribution, welfare, and politics under AI 01:12:04 Try to become more resilient  01:14:36 Information hygiene 01:22:16 Seven-year horizon and scaling limits by ~2030

On this episode, Calum Chace joins me to discuss the transformative impact of AI on employment, comparing the current wave of cognitive automation to historical technological revolutions. We talk about "universal generous income", fully-automated luxury capitalism, and redefining education with AI tutors. We end by examining verification of artificial agents and the ethics of attributing consciousness to machines.  Learn more about Calum's work here: https://calumchace.com  Timestamps:  00:00:00  Preview and intro 00:03:02  Past tech revolutions and AI-driven unemployment 00:05:43  Cognitive automation: from secretaries to every job 00:08:02  The “peak horse” analogy and avoiding human obsolescence 00:10:55  Infinite demand and lump of labor 00:18:30  Fully-automated luxury capitalism 00:23:31  Abundance economy and a potential employment cliff 00:29:37  Education reimagined with personalized AI tutors 00:36:22  Real-world uses of LLMs: memory, drafting, emotional insight 00:42:56  Meaning beyond jobs: aristocrats, retirees, and kids 00:49:51  Four futures of superintelligence 00:57:20  Conscious AI and empathy as a safety strategy 01:10:55  Verifying AI agents 01:25:20  Over-attributing vs under-attributing machine consciousness

On this episode, Tom Davidson joins me to discuss the emerging threat of AI-enabled coups, where advanced artificial intelligence could empower covert actors to seize power. We explore scenarios including secret loyalties within companies, rapid military automation, and how AI-driven democratic backsliding could differ significantly from historical precedents. Tom also outlines key mitigation strategies, risk indicators, and opportunities for individuals to help prevent these threats.  Learn more about Tom's work here: https://www.forethought.org  Timestamps:  00:00:00  Preview: why preventing AI-enabled coups matters 00:01:24  What do we mean by an “AI-enabled coup”? 00:01:59  Capabilities AIs would need (persuasion, strategy, productivity) 00:02:36  Cyber-offense and the road to robotized militaries 00:05:32  Step-by-step example of an AI-enabled military coup 00:08:35  How AI-enabled coups would differ from historical coups 00:09:24  Democratic backsliding (Venezuela, Hungary, U.S. parallels) 00:12:38  Singular loyalties, secret loyalties, exclusive access 00:14:01  Secret-loyalty scenario: CEO with hidden control 00:18:10  From sleeper agents to sophisticated covert AIs 00:22:22  Exclusive-access threat: one project races ahead 00:29:03  Could one country outgrow the rest of the world? 00:40:00  Could a single company dominate global GDP? 00:47:01  Autocracies vs democracies 00:54:43  Mitigations for singular and secret loyalties 01:06:25  Guardrails, monitoring, and controlled-use APIs 01:12:38  Using AI itself to preserve checks-and-balances 01:24:53  Risk indicators to watch for AI-enabled coups 01:33:05  Tom’s risk estimates for the next 5 and 30 years 01:46:50  How you can help – research, policy, and careers

Anders Sandberg joins me to discuss superintelligence and its profound implications for human psychology, markets, and governance. We talk about physical bottlenecks, tensions between the technosphere and the biosphere, and the long-term cultural and physical forces shaping civilization. We conclude with Sandberg explaining the difficulties of designing reliable AI systems amidst rapid change and coordination risks.  Learn more about Anders's work here: https://mimircenter.org/anders-sandberg  Timestamps:  00:00:00 Preview and intro 00:04:20 2030 superintelligence scenario 00:11:55 Status, post-scarcity, and reshaping human psychology 00:16:00 Physical limits: energy, datacenter, and waste-heat bottlenecks 00:23:48 Technosphere vs biosphere 00:28:42 Culture and physics as long-run drivers of civilization 00:40:38 How superintelligence could upend markets and governments 00:50:01 State inertia: why governments lag behind companies 00:59:06 Value lock-in, censorship, and model alignment 01:08:32 Emergent AI ecosystems and coordination-failure risks 01:19:34 Predictability vs reliability: designing safe systems 01:30:32 Crossing the reliability threshold 01:38:25 Personal reflections on accelerating change

On this episode, Daniel Kokotajlo joins me to discuss why artificial intelligence may surpass the transformative power of the Industrial Revolution, and just how much AI could accelerate AI research. We explore the implications of automated coding, the critical need for transparency in AI development, the prospect of AI-to-AI communication, and whether AI is an inherently risky technology. We end by discussing iterative forecasting and its role in anticipating AI's future trajectory.  You can learn more about Daniel's work at: https://ai-2027.com and https://ai-futures.org  Timestamps:  00:00:00 Preview and intro 00:00:50 Why AI will eclipse the Industrial Revolution  00:09:48 How much can AI speed up AI research?  00:16:13 Automated coding and diffusion 00:27:37 Transparency in AI development  00:34:52 Deploying AI internally  00:40:24 Communication between AIs  00:49:23 Is AI inherently risky? 00:59:54 Iterative forecasting

On this episode, Daniel Susskind joins me to discuss disagreements between AI researchers and economists, how we can best measure AI’s economic impact, how human values can influence economic outcomes, what meaningful work will remain for humans in the future, the role of commercial incentives in AI development, and the future of education.  You can learn more about Daniel's work here: https://www.danielsusskind.com  Timestamps:  00:00:00 Preview and intro  00:03:19 AI researchers versus economists  00:10:39 Measuring AI's economic effects  00:16:19 Can AI be steered in positive directions?  00:22:10 Human values and economic outcomes 00:28:21 What will remain for people to do?  00:44:58 Commercial incentives in AI 00:50:38 Will education move towards general skills? 00:58:46 Lessons for parents

Ed Newton-Rex joins me to discuss the issue of AI models trained on copyrighted data, and how we might develop fairer approaches that respect human creators. We talk about AI-generated music, Ed’s decision to resign from Stability AI, the industry’s attitude towards rights, authenticity in AI-generated art, and what the future holds for creators, society, and living standards in an increasingly AI-driven world.  Learn more about Ed's work here: https://ed.newtonrex.com  Timestamps:  00:00:00 Preview and intro  00:04:18 AI-generated music  00:12:15 Resigning from Stability AI  00:16:20 AI industry attitudes towards rights 00:26:22 Fairly Trained  00:37:16 Special kinds of training data  00:50:42 The longer-term future of AI  00:56:09 Will AI improve living standards?  01:03:10 AI versions of artists  01:13:28 Authenticity and art  01:18:45 Competitive pressures in AI 01:24:06 Priorities going forward

On this episode, Sarah Hastings-Woodhouse joins me to discuss what benchmarks actually measure, AI’s development trajectory in comparison to other technologies, tasks that AI systems can and cannot handle, capability profiles of present and future AIs, the notion of alignment by default, and the leading AI companies’ vague AGI plans. We also discuss the human psychology of AI, including the feelings of living in the "fast world" versus the "slow world", and navigating long-term projects given short timelines.  Timestamps:  00:00:00 Preview and intro00:00:46 What do benchmarks measure?  00:08:08 Will AI develop like other tech?  00:14:13 Which tasks can AIs do? 00:23:00 Capability profiles of AIs  00:34:04 Timelines and social effects 00:42:01 Alignment by default?  00:50:36 Can vague AGI plans be useful? 00:54:36 The fast world and the slow world 01:08:02 Long-term projects and short timelines

On this episode, Michael Nielsen joins me to discuss how humanity's growing understanding of nature poses dual-use challenges, whether existing institutions and governance frameworks can adapt to handle advanced AI safely, and how we might recognize signs of dangerous AI. We explore the distinction between AI as agents and tools, how power is latent in the world, implications of widespread powerful hardware, and finally touch upon the philosophical perspectives of deep atheism and optimistic cosmism.Timestamps:  00:00:00 Preview and intro 00:01:05 Understanding is dual-use  00:05:17 Can we handle AI like other tech?  00:12:08 Can institutions adapt to AI?  00:16:50 Recognizing signs of dangerous AI 00:22:45 Agents versus tools 00:25:43 Power is latent in the world 00:35:45 Widespread powerful hardware 00:42:09 Governance mechanisms for AI 00:53:55 Deep atheism and optimistic cosmism

On this episode, Ben Goertzel joins me to discuss what distinguishes the current AI boom from previous ones, important but overlooked AI research, simplicity versus complexity in the first AGI, the feasibility of alignment, benchmarks and economic impact, potential bottlenecks to superintelligence, and what humanity should do moving forward.   Timestamps:  00:00:00 Preview and intro  00:01:59 Thinking about AGI in the 1970s  00:07:28 What's different about this AI boom?  00:16:10 Former taboos about AGI 00:19:53 AI research worth revisiting  00:35:53 Will the first AGI be simple?  00:48:49 Is alignment achievable?  01:02:40 Benchmarks and economic impact  01:15:23 Bottlenecks to superintelligence 01:23:09 What should we do?

On this episode, Jeff Sebo joins me to discuss artificial consciousness, substrate-independence, possible tensions between AI risk and AI consciousness, the relationship between consciousness and cognitive complexity, and how intuitive versus intellectual approaches guide our understanding of these topics. We also discuss AI companions, AI rights, and how we might measure consciousness effectively.  You can follow Jeff’s work here: https://jeffsebo.net/  Timestamps:  00:00:00 Preview and intro 00:02:56 Imagining artificial consciousness  00:07:51 Substrate-independence? 00:11:26 Are we making progress?  00:18:03 Intuitions about explanations  00:24:43 AI risk and AI consciousness  00:40:01 Consciousness and cognitive complexity  00:51:20 Intuition versus intellect 00:58:48 AIs as companions  01:05:24 AI rights  01:13:00 Acting under time pressure 01:20:16 Measuring consciousness  01:32:11 How can you help?

On this episode, Zvi Mowshowitz joins me to discuss sycophantic AIs, bottlenecks limiting autonomous AI agents, and the true utility of benchmarks in measuring progress. We then turn to time horizons of AI agents, the impact of automating scientific research, and constraints on scaling inference compute. Zvi also addresses humanity’s uncertain AI-driven future, the unique features setting AI apart from other technologies, and AI’s growing influence in financial trading.  You can follow Zvi's excellent blog here: https://thezvi.substack.com  Timestamps:  00:00:00 Preview and introduction  00:02:01 Sycophantic AIs  00:07:28 Bottlenecks for AI agents  00:21:26 Are benchmarks useful?  00:32:39 AI agent time horizons  00:44:18 Impact of automating research 00:53:00 Limits to scaling inference compute  01:02:51 Will the future go well for humanity?  01:12:22 A good plan for safe AI  01:26:03 What makes AI different?  01:31:29 AI in trading

On this episode, Jeffrey Ding joins me to discuss diffusion of AI versus AI innovation, how US-China dynamics shape AI’s global trajectory, and whether there is an AI arms race between the two powers. We explore Chinese attitudes toward AI safety, the level of concentration of AI development, and lessons from historical technology diffusion. Jeffrey also shares insights from translating Chinese AI writings and the potential of automating translations to bridge knowledge gaps.  You can learn more about Jeffrey’s work at: https://jeffreyjding.github.io  Timestamps:  00:00:00 Preview and introduction  00:01:36 A US-China AI arms race?  00:10:58 Attitudes to AI safety in China  00:17:53 Diffusion of AI  00:25:13 Innovation without diffusion  00:34:29 AI development concentration  00:41:40 Learning from the history of technology  00:47:48 Translating Chinese AI writings  00:55:36 Automating translation of AI writings

On this episode, Allison Duettmann joins me to discuss centralized versus decentralized AI, how international governance could shape AI’s trajectory, how we might cooperate with future AIs, and the role of AI in improving human decision-making. We also explore which lessons from history apply to AI, the future of space law and property rights, whether technology is invented or discovered, and how AI will impact children. You can learn more about Allison's work at: https://foresight.org  Timestamps:  00:00:00 Preview 00:01:07 Centralized AI versus decentralized AI  00:13:02 Risks from decentralized AI  00:25:39 International AI governance  00:39:52 Cooperation with future AIs  00:53:51 AI for decision-making  01:05:58 Capital intensity of AI 01:09:11 Lessons from history  01:15:50 Future space law and property rights  01:27:28 Is technology invented or discovered?  01:32:34 Children in the age of AI