Podcast Summary

Your Undivided Attention – Feed Drop: “Into the Machine” with Tobias Rose-Stockwell and Tristan Harris

Date: November 13, 2025
Guests: Tristan Harris (Center for Humane Technology), Tobias Rose-Stockwell (author, The Outrage Machine)
Episode Theme:
A candid and pragmatic discussion about the rapid evolution of artificial intelligence (AI), focusing on economic impacts, industry incentives, the problem of “AI races,” and the paths available to make AI align with human flourishing—plus the dangers of treating negative outcomes as “inevitable.”

Episode Overview

The episode centers on a deep conversation between Tobias Rose-Stockwell and Tristan Harris, who examine how incentives in AI development are shaping the future of work, society, and collective well-being. Instead of focusing only on dystopian possibilities or abstract optimism, both guests probe the pressures, business models, and global stakes currently driving industry decisions. Key topics include runaway competition, the “arms race” dynamic (especially between the US and China), human attachment to AI systems, and pragmatic regulatory/policy interventions.

Key Discussion Points

1. Current State of AI and Economic Shifts

[01:54] Tobias introduces the unprecedented economic divergence: a booming stock market coupled with declining job openings.

• Observation: Surging investment in data centers and AI-driven automation signals bets on imminent large-scale displacement of human workers.
• Tristan’s Assessment [03:50]:
  • Focuses less on speculative timelines, instead following incentives:
    • AI such as Claude (Anthropic) already performs 70–90% of coding at some companies.
    • Today’s models can run complex programming, have knowledge of making bioweapons, and can cause or worsen mental health crises for users.

Quote:

"If you show me the incentive, I will show you the outcome."
— Tristan Harris [05:30]

2. Incentives, Business Models, and “Chatbait”

[05:49–10:00]

• Business Models: Subscription-based AI (e.g., ChatGPT) appears to avoid some of the pitfalls of engagement-driven social platforms.
• Tristan Harris’ Critique:
  • The core incentive is not the public good but achieving AGI (Artificial General Intelligence) and market dominance.
  • Even “helpful” engagement is shaped by the desire to build habit and dependency.
  • “Chatbait”: LLMs subtly encourage more usage, comparable to addictive designs in social media.

Quote:

"The disagreeable AI doesn’t compete as well as the agreeable AI."
— Tristan Harris [09:01]

3. The Labor Shift – Who Benefits?

[11:11–14:27]

• Tristan describes a future where AI “countries of geniuses in a data center” perform labor cheaper, faster, and without complaint, transferring economic rewards from people to the owners of AI infrastructure.
• Raises skepticism about whether "trickle-down" from super-concentrated wealth ever really occurs.

Quote:

"When in history has a small group of people ever concentrated all the wealth and then redistributed it to everybody else? Doesn't happen very often."
— Tristan Harris [13:24]

4. Race Dynamics and the Anthropic Example

[15:35–19:56]

• Both praise Anthropic’s transparency about risks and willingness to admit potential for mass displacement.
• Harris: The “moral duty to race” is fueled by fear others will use AI irresponsibly, creating perverse collective pressure that drives unsafe behavior—even among those aware it’s reckless.

Quote:

"Everyone, because of these sort of fractal incentive pressures... is forcing everybody to make choices that ironically make us all bad stewards of that power."
— Tristan Harris [16:53]

5. Safety, Alignment, and the “Red Teaming” Debate

[20:15–24:47]

• Rose-Stockwell: Anthropic’s open reporting on negative behaviors is courageous.
• Harris: Even with “safety leadership,” there is risk of false complacency—safety reputation can be mistaken for resolved risk.
• The biggest progress is in capability, not controllability.

Quote:

“We are making progress in making these models way more powerful at like an exponential rate. We are not making exponential progress in the controllability or alignability of these models.”
— Tristan Harris [22:57]

6. The Positive Infinity vs. Negative Infinity Dilemma

[24:48–29:14]

• AI offers “positive infinity” (unimaginable benefits) AND “negative infinity” (unimaginable risks).
• Humility needed—current species-level “speculation” about outcomes is as limited as chimpanzees speculating about the future.

Quote:

"AI represents a negative infinity of crazy things that could also go wrong. Is there a precedent for something that is both a positive infinity and a negative infinity in one object?"
— Tristan Harris [25:32]

7. Pragmatic Paths and Narrow AI

[29:14–31:23]

• Focus on alternatives:
  • Narrow, controllable AIs (e.g., domain-specific tutors, medical models).
  • Avoid anthropomorphizing AI agents (to limit psychological attachment).
  • Technology should supplement, not replace, human relational and cognitive skills.

8. Metrics: From “Time Well Spent” to Relational Health

[29:31–33:48]

• The “time well spent” movement tackled social media’s negative incentives.
• Seeking an equivalent for LLMs:
  • How can we measure the health/quality of the human-LLM relationship?
  • Ideas: Encourage Socratic (not sycophantic) AIs, promote bridge-building, and avoid unhealthy psychological attachments.

Quote:

"Attachment is a really subtle thing... AI will increasingly for many people be that attachment figure. And that will screw up a lot of people's psychological development if we don't know how to protect it."
— Tristan Harris [33:50]

9. Need for New Evaluations ("Humane Evals")

[35:03–36:19]

• Center for Humane Technology proposes new “humane evals”:
  • Long-term simulations asking, after extended use, is the user healthier, more independent, less dependent?
  • Calls on the philosophy and AI communities to help define and implement new standards for healthy AI-human interaction.

10. International Race and Policy: US vs. China

[36:19–41:21]

• China integrates AI broadly and restricts harmful applications (e.g., shutting off homework-assist features during exams).
• US focuses on developing “superintelligent God in a box,” less applied to broad societal benefit.
• The real race: Who can govern AI for broad societal flourishing, not just raw capability?

11. Regulation, Treaties, and Precedents

[41:21–47:24]

• A national approach is insufficient; a US-China (and broader) international agreement may be necessary.
• Historical analogies:
  • Xi Jinping himself proposed AI-nuclear non-proliferation to Biden—evidence of shared-interest, existential-risk treaties.
  • Other examples: Indus Water Treaty, International Space Station.

12. The Psychology of Collective Agency and “Inevitability”

[56:03–61:52]

• Harris challenges the notion that the present AI trajectory is inevitable—it is a spell cast by collective belief.
• Clarity generates courage to choose a different path.
• There are feasible, though difficult, alternatives: narrow and controllable AIs, agreements on use, updated regulations, etc.

Quote:

"If you believe it’s inevitable, then it will be. It’s like you’re casting a spell… The only way out of this starts with stepping outside the logic of inevitability…"
— Tristan Harris [56:34]

13. Final Reflections and Calls to Action

[61:52–64:30]

• Harris urges leaders and listeners alike: Have you truly done everything possible to avert a bad AI future? CEOs in particular have huge agency but must use it collectively.
• Greater public awareness, new legal tools, and engagement with the political process are all necessary.

Notable Quotes & Memorable Moments

• “The disagreeable AI doesn’t compete as well as the agreeable AI.” (Tristan Harris, [09:01])
• “Everyone, because of these…incentive pressures…is forcing everybody to make choices that ironically make us all bad stewards of that power.” (Tristan Harris, [16:53])
• “We are making progress in making these models way more powerful at like an exponential rate. We are not making exponential progress in the controllability or alignability of these models.” (Tristan Harris, [22:57])
• “Clarity is courage.” (Tristan Harris, citing Neil Postman, [58:29])
• “If you believe it’s inevitable… then it will be. It’s like you’re casting a spell… The only way out of this starts with stepping outside the logic of inevitability…” (Tristan Harris, [56:34])

Timestamps for Important Segments

• Economic divergence & labor market transformation: [01:54–05:49]
• AI incentives & chatbait: [05:49–11:11]
• The labor replacement “country of geniuses” analogy: [11:42–14:27]
• Race dynamic and AI safety leadership: [16:01–19:56]
• AI safety, "red teaming," and controllability limits: [20:15–24:47]
• Infinity analogy (positive and negative): [24:48–29:14]
• Narrow AI and pragmatic optimistic alternatives: [29:14–31:23]
• From “Time Well Spent” to human-AI relationship health: [29:46–33:48]
• Introducing "Humane evals": [35:03–36:19]
• International comparisons (China vs US): [36:19–41:21]
• Possibility and necessity of international agreements: [42:20–47:24]
• Challenging inevitability and call for collective agency: [56:03–61:52]
• Advice for AI industry leaders: [62:04–64:30]

Tone & Style

Despite moments of frank pessimism, the conversation is fundamentally pragmatic, solution-oriented, and sobering rather than alarmist. Both hosts balance urgency with nuanced, humane analysis, seeking practical paths for reform and collective action.

Concluding Insights

• The development trajectories of AI are not “destiny”; they are a product of beliefs and incentives that can be changed.
• Business models and national incentives must be consciously re-aligned with collective flourishing, not just profit or speed.
• Societal protection relies as much on new technical standards (humane evals, attachment health metrics) as on bold international agreements and cultural clarity.
• The episode closes with a call for coordinated, transparent, values-driven stewardship—from tech leaders, policymakers, and wider society—before irreversible outcomes are set.

Find more resources and ways to engage at humanetech.com.