Into The Machine with Tobias Rose-Stockwell

Episode 3: Tristan Harris on how to safely build artificial minds

October 31, 2025

AI Summary

5 min read

In 2025, Claude 4.5 can perform 30 hours of uninterrupted complex programming tasks, and at Anthropic, 70 to 90 percent of the code is now written by AI. This is not a future prediction; it is the present. Tristan Harris, founder of the Center for Humane Technology, argues that the most dangerous dynamic in artificial intelligence is not any single model's capability, but the incentive structure driving their development—a race for market dominance that makes safety a secondary concern.

The incentive flywheel behind the AI race

Harris rejects the idea that the AI industry's problems stem from advertising-based business models, the way social media's did. Instead, he points to the overarching goal of reaching artificial general intelligence (AGI). OpenAI's stated mission is to build AGI, and that goal creates a specific incentive flywheel: you need market dominance to attract the best engineers and the most investor capital. You need usage data and subscription revenue to fund ever-larger data centers and GPU clusters. You need those computational resources to train the next, more powerful model. Then you repeat the cycle.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

Never miss an episode of Into The Machine with Tobias Rose-Stockwell

Get every new episode summarized in your inbox — free, ~5 minutes to read.

No spam. Unsubscribe anytime.

What you'll learn

1 (00:03) **Opening & Guest Introduction** - Tobias Rose-Stockwell sets up the episode by noting a strange economic divergence: stock markets rising while job openings fall, driven by bets on AI replacing human workers. He introduces Tristan Harris, founder of the Center for Humane Technology and star of *The Social Dilemma*.
2 (01:21) **The Incentives Driving AI Today, Not Speculative Takeoff** - Tristan Harris explains he doesn't speculate about future "takeoff" scenarios; he looks at current incentives and what is already happening.
3 (03:56) **Pushback: Is the Subscription Model a Fix for Bad Incentives?** - Tobias pushes back, noting that LLMs like ChatGPT use a subscription model, not an advertising one, which was the core problem Harris identified with social media.
4 (08:08) **The True Economic Incentive: Owning the World's Labor** - The conversation pivots to the fundamental economic driver: replacing human labor.
5 (12:35) **The Egalitarian Paradox & The "Inevitability" Trap** - Tobias presents a counterpoint: studies show AI helps junior workers reach an average baseline but can slow down senior experts, making it an egalitarian tool.
6 (17:00) **Anthropic's Role: Safety Research as a Double-Edged Sword** - The discussion focuses on Anthropic, the company most associated with safety.
7 (20:36) **The Uncontrollability Gap: Power vs. Control** - Harris sharpens the central technical problem.

+ Full timestamped outline available in the app

Guests on this episode

Tristan Harris

Show Notes

Today I’m releasing a conversation with Tristan Harris.

Tristan is the founder of the Center for Humane Technology and one of the leading voices warning about how runaway AI might destabilize society. He starred in (and co-produced) the Netflix documentary The Social Dilemma.

I’ve known Tristan for a long time, and this is one of the best conversations we’ve ever had—public or private. I press him on major risk scenarios, what we can expect AI labs and legislators to do in the face of AGI, and what he thinks can actually be done right now to ensure these systems stay maximally beneficial to humanity.

I left this conversation a bit more hopeful understanding what kinds of solutions are available. I hope you do too.

In this episode we talk about:

* Creepy new AI capabilities: new models using unwitting humans to send encoded messages to other AIs.

* How it might all go down: a real-world near-term disaster scenario with runaway self-replicating AIs.

* How to bypass race dynamics with China and other powers accelerating AI capabilities.

* Designing systems for wisdom: alternative paths for designing and training Socratic AIs.

A small ask:

The irony is not lost on me in trying to critique the algorithms while still being dependent upon them to reach the right audience.

If you do enjoy the show, please share this episode with a friend and drop us a rating.

You can follow us here:

Thanks for listening, and please do subscribe.

-Tobias

Into The Machine is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

Get full access to Into The Machine at tobias.substack.com/subscribe