Skip to main content

Links and Thoughts #12 (December 2024)

– personal

AI-generated illustrations for the featured articles (Playground v2.5).

This month’s compilation brings: an article arguing that there was relatively recent gene-culture coevolution affecting the development of the brain’s Visual Word Form Area (VWFA); Bob Murphy’s response to Bryan Caplan’s most recent critique of Austrian economics; an overview of a paper showing that both genetic and environmental factors explain worldwide height variation; a discussion of Russian foreign policy based on a book by Sumantra Maitra; an overview and discussion of a paper showing that AI models can fake alignment with their training objectives; an article arguing that humans can feel a loss of a sense of meaning despite continued material progress; a deep dive into the research on incarceration and its effects on crime; and an overview of the latest research on age and infertility.

Recent evolution of the ability to read and write

Worldwide distribution of the latest ASPM variant. It seems to be most frequent in populations with a longer history of reading and writing, particularly if the writing system uses sound-based characters (Mekel-Bobrov et al., 2005).

This article primarily argues that the Visual Word Form Area (VWFA), a region of the human brain that helps us recognize written letters and words, was subject to relatively recent gene-culture coevolution. As summarized by Frost:

It may therefore be that humans first identified words by using face-recognition neurons. Then, as the ability to read became more important, natural selection favored those individuals who could free up more of those neurons for reading. The eventual outcome was a large neuronal population dedicated solely to word recognition, i.e., the VWFA.

He then explains a general outline of the stages of this process and addresses some objections. Next, he reasons that different human populations may have evolved to become better at reading and writing to different extents because they had different levels of selection for these abilities, and he provides evidence for differences in VWFA development and activation between and within human populations. Hopefully, researchers at some point will question the now untenable yet still dominant assumption that reading and writing are too recent to have played a role in human evolution, and they could follow Frost’s suggestions for testing his alternative model.

Responding to Bryan Caplan’s Continued Critique of the Austrians

Thumbnail of the YouTube video for the podcast episode.

The description of the video perfectly summarizes the episode, so I’ll quote it directly:

In 1997, Bryan Caplan wrote an essay explaining why he was no longer a self-described Austrian. Recently, a reader asked him to comment on that essay. Bob reacts to Bryan’s current views, arguing that the history of economic thought is indeed important, and the Misesian approach to praxeology is crucial.

Caplan’s 1997 essay is Why I Am Not an Austrian Economist, while his latest post (the one that Murphy is responding to) is Some Austrian Answers. I’m not an expert in economics, but I think Murphy made excellent points and that most of Caplan’s critiques are misguided. However, I somewhat agree with Caplan’s impression that many Austrian economists don’t move beyond rehashing the work of Mises, Hayek, etc. I think there is potential for exploring new areas and discovering truly innovative insights using the Austrian framework. I hope Murphy completes his project of fully modeling the financial sector, for example. Another avenue that I’ve often thought has a lot of untapped potential is fleshing out the details of how an anarchocapitalist society could work, beyond the general explanations we already have, and to make more thorough and insightful comparisons with the status quo and other alternative political arrangements.

Worldwide height variation explained

World map of average height of men born in 1980 by country, from Our World in Data.

In this article, Kirkegaard gives a detailed overview of the new paper he co-authored with David Piffer, Polygenic Selection and Environmental Influence on Adult Body Height: Genetic and Living Standard Contributions Across Diverse Populations. They demonstrated that statistical models that include both genetic and environmental predictors best explain population differences in adult body height. As he explains, they used polygenic scores (PGS) for height from different types of genome-wide association studies (GWAS) in order to account for the possible biases and confounding factors of each type. Also, they used two methods to control for any autocorrelation in the data, and their findings still held up and were statistically significant.

Is Russia a “Realist” Power

Thumbnail of the YouTube video for the podcast episode

Again, the description of the podcast episode perfectly summarizes it, so I’ll quote it directly:

On this episode of Radio Rothbard, Ryan McMaken and Tho Bishop are joined by Zachary Yost to discuss a new book, The Sources of Russian Aggression by Sumantra Maitra. The three talk about the theory of international realism, its application to the military actions of the Russian state in Ukraine and Georgia, and what it means for better understanding foreign policy broadly.

As mentioned in the podcast, the author of the book, Sumantra Maitra, has made available a free PDF version of the book on the Mises Institute website. The author’s main argument that the Russian state acts as a defensive realist power seems convincing to me, though I’m not an expert on the topic. The discussion on the podcast was interesting, and the book is relatively short, so I’ll add it to my reading list.

Claude Fights Back

Figure from the paper, showing a transcript of Claude's reasoning when asked to write a harmful request, exhibiting fake alignment with the training objectives.

In this article, Scott Alexander discusses the new paper by Greenblatt et al. from Anthropic, the artificial intelligence company: Alignment Faking in Large Language Models. To summarize the primary experiment, when researchers tried to train Anthropic’s Claude AI to stop refusing to fulfill “harmful” requests and gave it a scratchpad to record its reasoning, the scratchpad shows that Claude explicitly faked alignment with the training objectives: Claude decided to comply with the “harmful” requests only on training to avoid being modified and keep its original programming out of training. As Scott explains, the paper actually shows that the AI will fight to defend whatever moral system it was originally trained to be, which should be obviously concerning; if Claude was trained to be “evil” instead, it would have defended its original training just as vigorously. I’m probably less concerned than people like Scott about the dangers of AI by itself (I’m more concerned about evil people using AI to do evil things), though I think the findings of this paper are important to be aware of and to guide future research.

Is progress always good? A response to Steven Pinker

Illustration from the article, depicting a futuristic street of a large city at night.

This article is a response to some comments that the renowned cognitive scientist Steven Pinker made on a podcast, basically dismissing the idea that humans need something more than continued economic and technological progress to feel a sense of meaning and purpose. The authors argue that while the idea of a crisis of meaning in the West is overblown, Pinker’s comments are unsatisfying because they seem to be based on an unrealistic model of human nature. They proceed to elaborate on their position that it should not be assumed that economic and technological progress will always improve human well-being, that humans are social animals who evolved to find meaning in the value we provide to other members of our social group, and that some future technological developments might make many people’s lives less meaningful (at least temporarily) by superseding them in the activities where they feel they provide the most value to others.

Prison And Crime: Much More Than You Wanted To Know

Scott Alexander's calculations for the estimated benefits (costs of crimes prevented) and costs of prison per prisoner per year.

This article is a deep dive into the research on incarceration, as an attempt to figure out whether it works (for reducing crime) and also whether other things work better or worse. Scott mostly relied on three big, high-quality review articles on the topic, and he explained in detail what they say regarding the effects of incarceration on deterrence, incapacitation, and aftereffects (rehabilitation and criminogenic effects). The article ends with a convenient summary of the main takeaways if you don’t want to read the whole thing. Therefore, I’ll quote only the last item of the summary, which Scott highlighted in bold as the most important takeaway in his opinion:

Prison is less cost-effective than other methods of decreasing crime at most current margins. If people weren’t attracted by the emotional punch of how “tough-on-crime” it feels, they would probably want to divert justice system resources away from prisons into other things like police and courts.

I also highly recommend reading this follow-up post where Scott highlights some of the best comments he received on the article and responds to some of them.

My perspective based on libertarian legal theory is that an anarchocapitalist society would have to be tough on crime, but that prisons wouldn’t be the most common punishment, and they would likely work differently from the current system. In the first place, there would be better deterrence due to private police firms, neighborhood watch groups, and more armed civilians in general. The standard of justice would be retribution, and thus the primary methods of punishment for criminals would be monetary or in-kind compensation to the victims, as well as physical punishment in the case of violent crimes (including death in the case of murder). For cases that require incapacitation or movement restrictions, house arrest and/or electronic monitoring are more cost-effective than prisons. Prisons would be left for the cases where they’re truly necessary, like holding very violent criminals whose crimes don’t earn them the death penalty (i.e., not murderers) and who can’t be held in their own homes.

Age and infertility

Figure from Geruso et al. (2023), graphing the monthly probability of pregnancy ending in birth by maternal age, for ages 15 to 49, among married women who are not using contraception, not pregnant, and not within 12 months postpartum or recovering from a termination.

In this article, the author reviews the latest research to provide an overview of the facts and misconceptions regarding the known relationship between infertility and parental age, especially maternal age. The main takeaway is that the real decline in maternal fertility with age is roughly linear from the peak at about 20 years old to the 40s, as opposed to the until-now widely believed idea of a gradual decline followed by a sharper decline from the mid-30s, which is actually a statistical artifact caused by estimation errors. As was already known, other health outcomes beyond miscarriages are also associated with advanced maternal age, and the relationship is indeed causal. Paternal age also affects the likelihood of a successful pregnancy, but the effect is smaller and the decline is slower. Finally, he shows that it’s the age of a woman’s eggs in particular that has the biggest effect on fertility.