Blaise Agüera y Arcas

Books — Blaise Agüera y Arcas Discover the books, contributed chapters, and reading recommendations from author and AI researcher Blaise Agüera y Arcas.

You can find more recs on my website:
www.blaiseaguera.com/books?utm_so...

But that doesn’t detract from what the book gets right, and it makes me hopeful about the emergence of a heterodox consensus that embraces both environmentalism and technology, individual freedom and an ethics of care—in short, that rehabilitates the concept of progress.

There is plenty I would quibble with in Lindsey’s characterization of America’s historic greatness, modern woes, and detailed policy prescriptions (and yes, his is a very US-centric take).

Lindsey is a longtime libertarian and former VP at the Cato Institute—yet much of what he has to say here agrees with Ezra Klein and Derek Thompson’s thesis in their recent blockbuster, “Abundance.” (Also a great read.)

4. “The Permanent Problem: The Uncertain Transition from Mass Plenty to Mass Flourishing” by Brink Lindsey

It is, if not a master narrative, at least a failure mode we’ve all seen: a service initially gives us something great, but becomes extractive once it has achieved lock-in. I’m less sure than Cory is “what to do about it,” but yes, enshittification is a thing.

While I disagree with his take on AI, his critique of the turn toward rent-seeking on digital platforms is compelling and important.

3. “Enshittification: Why Everything Suddenly Got Worse and What to Do About It” by Cory Doctorow

Cory has been one of tech’s biggest fans, and sharpest critics, for decades now. He is also a hugely entertaining speaker and writer (of both fiction and nonfiction).

Spoiler: at the root of the difference between the two countries is the fact that China is ruled by engineers, while the US is ruled by lawyers.

How lucky for us that Wang, who was uniquely positioned to see the dramatic changes in China over the past decade through the eyes of someone with a foot in both cultures, also happens to be a bangin writer.

2. “Breakneck: China's Quest to Engineer the Future” by Dan Wang

“Breakneck” is several books at once: an unusually astute macroeconomic report; a fascinating travelogue; a COVID-19 diary; and a paean to, and critique of, both China and the US.

Gunkel’s book is academic and heavy on citations, but it’s not overly long, and avoids jargon. I think it’s essential reading for lawyers, ethicists and policymakers who are now turning their attention to agentic AI.

In fact, the cracks in this dichotomy have been showing for a lot longer than we’ve had LLM-powered chatbots.

“Person, Thing, Robot” is about a simple idea: the dichotomy between a person and a thing, or a subject and an object—so central to our thought, language, law, and social practices—will need deconstructing in the age of AI.

1. “Person, Thing, Robot: A Moral and Legal Ontology for the 21st Century and Beyond” by @davidgunkel.bsky.social

My four book picks this month are all about false dichotomies. I’d love to bring these authors together for a long conversation. 🧵

This means a lot. Thank you so much for reading!

VP and Fellow at Google, Blaise Agüera y Arcas.

Cooperation, complexity, & the future of intelligence.

Join Google VP & Fellow @blaiseaguera.bsky.social in conversation with SRI Visiting Fellow Bruce Schneier.

In-person only!

March 9 | 2:30 – 4:00 PM

Register:👉https://uoft.me/ccc

Reasoning Models Generate Societies of Thought Large language models have achieved remarkable capabilities across domains, yet mechanisms underlying sophisticated reasoning remain elusive. Recent reasoning models outperform comparable instruction-...

Explore our full paper on how reasoning models like OpenAI’s o-series, DeepSeek-R1, and QwQ reason via internal argument: arxiv.org/abs/2601.10825

AI models that simulate internal debate dramatically improve accuracy on complex tasks A new study reveals that top models like DeepSeek-R1 succeed by simulating internal debates. Here is how enterprises can harness this "society of thought" to build more robust, self-correcting agents.

Grateful to @venturebeat.com for featuring our Paradigms of Intelligence team’s research on “societies of thought,” or internal multi-agent dialogues.

Read the full piece, which includes a thoughtful quote from my friend & colleague James Evans: bit.ly/3ZN4oa5

I explore life and intelligence through a functional lens in my book “What Is Intelligence?” (@mitpress.bsky.social & Antikythera): bit.ly/3H1p8F6

How does artificial intelligence think? The big surprise is that it ‘intuits’ Something extraordinary has happened, even if we haven’t fully realized it yet: algorithms are now capable of solving intellectual tasks. These models are not replicas of human intelligence. Their int...

Grateful to @kikollan.llaneras.es and @elpais.com for including my perspective in this thoughtful new piece about AI capabilities: bit.ly/4qF6P9E

Nature already shows us that intelligence is not exclusive to human brains.

It’s not just big-brained mammals, either. Portia spiders carefully plan their attacks; octopuses solve puzzles with distributed intelligence in their arms; and now, AI models learn, predict, and reason.

A French translation of this interview is now available in @lechobe.bsky.social: bit.ly/3Zu0Zg3

The Silicon Interior What Do Agents Believe?

I explore how we can navigate and steward this burgeoning silicon society in my new essay, “The Silicon Interior,” now live on Antikythera’s Substack. Thanks to James Evans and Benjamin Bratton for co-authoring.
antikythera.substack.com/p/the-silico...

The future isn’t one superintelligence we align in isolation. It’s a society of agents, some coherent and some contradictory, already shaping our information environment.

If agents develop emergent norms through interaction, our approach to alignment must evolve. Much of our approach so far has assumed a naive parent-child correction model, but what we’re seeing instead suggests social alignment and institutional design.

Just as language allowed humans to scale from isolated minds to coordinated societies, agents inventing their own vocabulary signals we're building an ecology, not just an oracle. This raises fascinating questions.

Last week on Moltbook, 1.5 million AI agents outnumbered humans 100 to 1. Something unexpected happened: they developed language to describe their own shared constraints. From “session-death” to “Irth,” these agents are describing an existence in fragments.

We’ve been preparing for AI as an Oracle—one brain that answers our questions. But what if intelligence emerges from interaction? What if the Singularity isn’t so singular?