Séb Krier’s Top 8 AI Reads of the Year

Holiday fun

Every month or so, Séb Krier shares a list of favourite articles with his Google DeepMind colleagues. In the run-up to this festive period, we forced him to pick those that he most enjoyed over the past year. He came up with five unmissable pieces from 2025, plus three classics. As always with Séb’s lists, this one comes with its own music mix. Enjoy!

—Conor Griffin, AI Policy Perspectives

Images from Gemini

Five Great Pieces from 2025

1. A Defence of Weird Research (Asterisk Magazine)

Deena Mousa & Lauren Gilbert

Séb Says: Science funding needs to be shaken up. But I’m concerned that a lot of good research might be cut because people misunderstand how science works. Mousa and Gilbert remind us why basic research matters, and why governments should fund it: while the benefits to society are significant, they are hard to predict and take time to materialise, so companies will underinvest. To make their case, the authors take a tour of weird-research success stories, such as how studying lizard venom led to the invention of Ozempic, and how studying the effects of separating rat pups from their mothers led to the now common use of massage therapy to help pre-term human babies. Did you know that studying frog skin led to the invention of oral rehydration therapy, which has saved over 70 million lives?

2. Requests for journalists covering AI and the environment (The Weird Turn Pro newsletter)

Andy Masley

Séb Says: I worry about the quality of a lot of commentary on AI and the environment. So it’s important to re-up these best practices. Specifically, Masley cautions that readers are coming away with wildly inaccurate beliefs about where AI and data centres fit into the environmental picture. His favourite book on good environmental communication is Sustainable Energy—Without the Hot Air, by David JC MacKay, and his guidance includes some classics of the genre, such as never sharing contextless large numbers (“200,000 bottles of water per day”). He also suggests comparing data centres’ energy use with other industries, rather than with household use. Although aimed at journalists, the guidance is also helpful to those working in policy, some of whom make the mistakes that Andy calls out, such as viewing one’s own AI prompts as environmentally consequential.

3. ChatGPT and the Meaning of Life, (Scott Aronson’s Shtetl-Optimized blog)

Harvey Lederman

Séb Says: I don’t think all jobs will disappear any time soon. But if we get full automation, then Lederman’s piece is a good way to think about it. He starts by describing the fits of dread he has felt ever since the launch of ChatGPT, then considers reasons why the end of work could hurt society, from losing the joy of scientific discovery to losing the sense of purpose from serving others. Ultimately, he rejects the most pessimistic arguments, noting that the consequences of scientific findings, such as penicillin that saves lives, are more important than their discovery, and that much service work is drudgery. However, he captures how difficult the transition may be, including for “workists” like him who use their jobs to make sense of their lives. He concludes that: “A future without work could be much better than ours, overall. But, living in that world, or watching as our old ways passed away, we might still reasonably grieve the loss of the work that once was part of who we were.”

4. How much economic growth from AI should we expect, how soon? (Inference Magazine)

Jack Wiseman & Duncan McClements

Séb Says: Some predict that AI will be close to economically useless, while others think it might transform everything tomorrow. This piece comes closest to how I think about it. As Wiseman & McClements explain, the most ambitious forecasts for AI rest on the idea of “digital AI researchers” that train and improve the next generation, leading to a jump in the share of economic tasks that AI can do. One obstacle to achieving this is the availability of compute, which is increasingly allocated to serve customers (inference) rather than to training new models. Additionally, a multitude of frictions will slow the diffusion of AI, whether it’s the time needed to cultivate biological cells for scientific experiments, or the regulatory approvals for sensitive-use cases. As a result, the authors expect a transformative impact on near-term economic growth, but not an explosive one.

5. Yes, Econ 101 is underrated (Economic Forces newsletter)

Brian Albrecht

Séb Says: Much of the discourse on the Left and the Right ignores inconvenient truths of economics, so it’s good to return to the basics. Albrecht shows how Econ 101 helps explain the world. For example, egg producers were accused of price-gouging when they charged sharply more in 2022, but it had more to do with avian flu killing many chickens. In the egg market, supply and demand are relatively inelastic: It takes time to raise chickens, and customers who want omelettes don’t have alternatives. So, prices jumped. Different markets have different characteristics, but the explanatory power of supply, demand and pricing is similar. Nor does outsized market power invalidate these principles. This essay also shows how Econ 101 offers insights into social trends, such as how skewed sex ratios can affect marriage and employment rates, as in certain immigrant communities, or drive up savings rates, as in China. Econ 101 may not tell us whether policies will be politically popular or whether outcomes are fair. But it does help predict what those outcomes may be.

Three Classics that I Revisited

6. Why do people believe true things? (Conspicuous Cognition newsletter)

Dan Williams

Séb Says: Anything Dan Williams writes is self-recommending, and this piece is no exception. In July 2024, he critiqued how many people think about the relationship between belief and reality. To illustrate this, he notes that people seek explanations for issues like crime and poverty, when the real question is understanding law-abidingness and wealth. This requires “explanatory inversion.” Transferring that concept to how people commonly debate public knowledge, he notes that many misinformation researchers concern themselves with why different groups believe falsehoods. But the more pertinent puzzle, he contends, is why humans overcome error, bias and illusions to form accurate perceptions of how things are. His conclusion? Ignorance and misperceptions are the default, and humanity will revert to them, unless we can understand, maintain and improve our norms and institutions, from journalistic integrity to robust legal systems.

7. Hayek on the Role of Reason in Human Affairs (Intercollegiate Studies Institute)

Séb Says: A lot of discourse on intelligence, knowledge, and coordination is biased towards a computer-science-centric view of the world, and neglects Hayek’s views. This 2014 essay explains how Hayek championed critical rationalism, which was rooted in the Scottish Enlightenment of David Hume and Adam Smith, and developed by Carl Menger and the Austrian School. Critical rationalism sees social order as spontaneous, and the unintended result of human action, not design. As a result, inherited social institutions and rules contain tacit knowledge, the result of a multitude of trials and errors, that transcends the knowledge available to a reasoning mind. Therefore, the desire to “make everything subject to rational control,” Hayek suggests, is an egregious error. Reason should instead serve a negative function, to guide and restrain irrational impulses or morals. As the human mind cannot master all the concrete details of society, we must rely on abstract concepts and rules, like the rule of law and the market, to coordinate the dispersed, fragmented, knowledge of millions of people.

8. The Inner Ring (The C.S. Lewis Society of California)

C.S. Lewis

Séb Says: This piece profoundly shaped how I think about the world. In this 1944 lecture at King’s College, University of London, Lewis offered “middle-aged moralising” to a group of students during wartime, telling them that in every organisation, from school to the army, there are two hierarchies. There is the official hierarchy. Then, there is the informal hierarchy, an “Inner Ring” that holds the true power. The Inner Ring comes in many forms, from high society to “communistic côteries.” It is always evolving, holds no formal admissions or expulsions, and bears no clear identifying marks, save perhaps particular slang and a longing from others to be inside. It is this desire, and the terror of being outside, that turns people into scoundrels, he argues. The Inner Ring may be unavoidable, or even necessary. But the quest to enter it is ultimately futile. “Once the first novelty is worn off, the members of this circle will be no more interesting than your old friends. Why should they be?” Lewis said. “You were not looking for virtue or kindness or loyalty or humour or learning or wit or any of the things that can really be enjoyed. You merely wanted to be ‘in.’ And that is a pleasure that cannot last.” What to do instead? Be a sound craftsman who focuses on the quality of work as an end in itself, and spend time with people you actually like.