Weekend Reading & Selected Links
Happy weekend, and hello from San Francisco (where I've been recording some interviews)! Here are some links to things I've been reading or watching that you might also enjoy:
- My new podcast episode, with Rob Boyd and Peter Richerson. At the bottom of this email, I've reprinted five excerpts from the conversation.
- Robin Hanson's comments on my Boyd & Richerson podcast.
- 'How Stewart Made Tucker', a 2022 article by Jon Askonas.
- 'Can AI Scaling Continue Through 2030?', a new Epoch AI report.
- Toby Ord revisits the x-risk landscape five years after publication of The Precipice.
- The Black Swan is the most influential book of the 21st century.
- 'AI existential risk probabilities are too unreliable to inform policy', a recent article by Arvind Narayanan and Sayash Kapoor.
- Which books, papers, and blogs are in the Bayesian canon?
- Recent Tablet profile of Palmer Luckey.
- Brian and Chris of Affix podcast discuss my review of Alan Kohler's Quarterly Essay.
- Nigel Warburton's five key philosophical texts in the Western Canon.
- Terry Tao on Black-Scholes, from 2008.
- A nice blog post on understanding statistical moments, by Gregory Gundersen.
- NEO, the new humanoid-like robot.
- In case you missed it, Patrick Collison's list of books in the tech canon.
Have a great weekend,
Joe
Excerpts from my podcast with Boyd & Richerson
1. How contingent was the evolution of cultural brains?
JOSEPH WALKER: Okay, so everything we've spoken about thus far, for me, raises the question of contingency. So it seems like – you can correct me if this is inaccurate — but for cultural brains to evolve took at least three, if you like, exogenous shocks.
One was apes becoming bipedal about 5 million years ago.
The second was the large amount of predation on the savannas in Africa. I read that there was about twice the level of predation there is today, both in terms of the number and the type of predators.
And then the third factor is the Pleistocene climate variation we've spoken about. Do you see all of those ingredients as just independent factors, and we won the lottery in a sense, in that we had these pre-adaptations, our lineage had become bipedal apes, were under the threat of lots of predation and the selection pressures that was creating for grouping socially, and then, just as that was happening, we entered this period of climate variation on a millennial and submillennial scale?
Or are those three factors somehow correlated? Maybe bipedalism was driven by the climate variation in some sense? Maybe apes had to change their foraging strategies because of the forests growing and shrinking or something like that?
So how contingent was the evolution of cultural brains?
ROBERT BOYD: I think it's really contingent. I'd add a bunch of other things.
WALKER: What would you add on your list?
BOYD: Well, I think this … So humans, australopithecines, were bipedal, and that placed constraints on brain size, because as brains get bigger, there's more obstetric complications. That led to … So in early Homo, there seems to have been a shift to a more predatory lifestyle. So now we've got apes out on the savannah, they're bipedal. Heads are little, so that's not a problem.
Then they start hunting. Females can't hunt with highly dependent offspring, and that leads to changes in social organisation and cooperative breeding, so-called. Cooperative breeding, that probably potentiates cultural transmission because there's more individuals together in social groups.
PETER RICHERSON: Also, it brings males more firmly into … Fathers have to teach their kids, their boys to hunt.
BOYD: Yeah. So there's a bunch of things, I think, all of which could have gone some other way. When I teach this in class, I say it's like little Eliza jumping from one ice floe to another. It's just a bunch of stuff happened and it ended up where we are. But it didn't have to be that way at all.
That'd be my story.
RICHERSON: Yeah. I would say basically the same thing.
But if you look at any evolving lineage looking backwards, it's one crazy thing after another. So, horses started as these little forest browsers, and as the climate got drier, more open, then you had a whole revolution, Miocene revolution, in mammals. You had the origin of the antelope and bovids and all these efficient grass-eating herbivores that exploited the more open vegetation, browsers that could … If the trees have their leaves way the hell above the ground, then there are no mammals that get up there very much. But if you've got low trees and shrubs and things, then you’ve got a bunch of browsing specialists that can get after the vegetation. And then there are top-down effects of the grazers and the browsers on the ecosystems.
So if you look back in history, it's just one damn thing after another. So historical contingency plays a big role.
2. On how climate variation has been driving brain size increase on Earth for millions of years
WALKER: So I want to play a counterfactual. Imagine in the next few years, we get some new data that stretches all the way back into the Pliocene that shows that climate variation has actually been continually increasing since the Pliocene, right up through the Pleistocene. From your perspective, what's the most important story that would enable us to tell? Would it enable us to tell some kind of grand story about how climate variation has been driving the evolution of brain size and intelligence on Earth?
RICHERSON: Well, I think that what we expect … in other words, doing what Rob and I didn't do in the ’85 book, turning the question around and arguing that brain size is a kind of a palaeoclimate indicator. In other words, brains are for coping with variable environments on different timescales. Culture being, as you say, in the sweet spot; there is a sweet spot for culture. So what? Well, to go even further back, brain size in mammals generally has been increasing for the last 65 million years.
WALKER: Since the dinosaurs.
RICHERSON: Since the dinosaurs, yeah. And birds probably too. But birds don't fossilise very well because they're so delicate. To fly, you have to have light bones, and light bones aren't very rugged. And so the bird record is poor, and the mammal record is quite poor. A guy named Harry Jerison at UCLA in 1973 in his book, I think, he used naturally occurring fossil endocasts to build a record of brain size evolution for the last 65, 70 million years. But it's a really crude record because fossil endocasts aren't very common. And nowadays somebody could go back and take all the skulls in all of these collections and put them in an x-ray machine and do computerised tomography and develop a much finer-resolution set of data. But I keep looking for this, and so far, nobody that I have run across has actually tried to do this.
BOYD: It's complicated because the other response to climate variation is to go small.
So apes – there were hundreds of species of apes 20 million years ago, and by 8 or 10 million years ago, most of them were gone, replaced by monkeys.
So apes came first, and monkeys replaced apes mainly, not completely.
So, if you can't predict, then just make a lot offspring; that’s the kind of invertebrate strategy. And that's going to be overlaid on top of …
RICHERSON: Well, yes. And Jerison’s data, how did he summarise it? Many mammalian lineages got bigger brains, but by no means all of them.
WALKER: So what's been driving that? If it's been happening for 65 million years?
RICHERSON: Well, what I think our models suggest, and models like them suggest is that the world has been getting more variable, on short timescales. Now, the trouble is that there are no short timescale records that I know of – certainly not to cover the whole last 65 million years. Now, I don't think you need to need that. You could just have samples. You could look at fossil tree rings or something like that and develop samples of high-frequency climate variation – old lake sediments, old marine sediments – and not try to have a complete record, just spot samples would give you some indication.
BOYD: You'd like to know the slope of the brain size versus time curve, right? Because I think we just know the average.
RICHERSON: Well, Jerison suggested that there was a long, slow … and then the transition to the Pleistocene was our story.
BOYD: Pleistocene, Miocene, Eocene variation being not that big: warm, wet planet. And then all of a sudden in the late Miocene things go to hell, basically, and it gets colder and drier and more variable. And so until 20 million years ago, there were tropical rainforests in Moscow
RICHERSON: Temperate rainforests in Antarctica.
BOYD: Yeah, exactly. So it was a much more … Now this is, of course about primates who are forest specialists. But then all of a sudden in the Miocene, things dry out. You can't just live in the forest anymore. There’s a lot more variation. And if it's true – I didn't know this about Jerison’s data – if there’s an acceleration …
RICHERSON: Well, again, the crudity of that data might cause some scepticism. But that's what he says.
BOYD: That's what I predict, okay, so I’ll stick my neck out here.
WALKER: I like it. That's so interesting.
RICHERSON: There is a great paper that you might look up by a geochemist at UC Santa Barbara and his co-authors. I think his first name is Zachos. A 2001 paper in Science, I think. If you have trouble finding it, just drop me an email and I'll give you the full reference. And I don't know how he derived it, but he and his co-authors derived it. But they have in the graph that they have a pattern of dots around the mean curve that they use to represent their impression of what the variation in climate has been. They don't specify timescale.
BOYD: I know this. Yeah, I know that paper.
RICHERSON: It's a great paper.
BOYD: I think the dots are just data points.
RICHERSON: I don't know what data they are.
BOYD: It's all from deep sea cores, yeah?
RICHERSON: Could be, yeah.
BOYD: Yeah, I think that's right. I think it's all from deep sea cores. And what's piece, the variant, the cloud of points around the mean gets much wider.
RICHERSON: As you get to … particularly during the Pleistocene, it just explodes. But I think this would be the orbital scale variation. If it's real data, it's orbital scale. Milankovitch. 100,000-year, 20,000-year, 41,000-year, 23,000 years the shortest. They aren't real cycles, they're more complicated than that. They're orbital perturbations that come from the gravitational effect of mainly Jupiter on the Earth. So that to a first approximation, the Earth and the other planets follow these orderly orbits, but under the massive influence of the gravitation of the Sun. But Jupiter has enough gravitational force on the other planets to perturb their orbits in a systematic but not exactly cyclical way. That's where you get the 100,000-year ellipticity and the 41,000-year tilt and the 23,000-year procession of the equinoxes by the gravitational effects. I suppose that Saturn and the other big planets also may have a measurable effect.
WALKER: I didn't know that we could connect the variation back to the orbital perturbations.
RICHERSON: But the orbital perturbations then are somehow filtered by the Earth ocean atmosphere system. So the 100,000-year, the 41,000-year and the 23,000-year quasi0cycles, they sometimes say: they haven't changed. As the Earth moved from the Pliocene to the Pleistocene, the early Pleistocene and then the later Pleistocene, those orbital scales didn't change. Something about the way the Earth responded to them is what generated the dominance of one cycle versus the other. Just exactly what that all amounts to is at least still a mystery to me. I don't know if it's a mystery to the palaeoclimatologists or not, but I think it is. And what generates the millennial and submillennial scale variation? And why has it been increasing over at least the last 1.5 million years?
One hypothesis I saw – speculative, I think – it's that the North American glacier is sort of the dominant ice mass during the high glacial episodes. Over the successive glacial cycles, it's gradually been planing the North American continent flatter, so it reduces the friction. So the reason that the amplitude and frequency have gone up is because the sliding of the glacier off the North American continent has gotten faster and faster.
3. Density and fertility
WALKER: One argument about the urban phenomenon is that there's some deep genetically evolved switch in our brains that once we're surrounded by density and crowds of people it causes us to downregulate our fertility in response. Does that seem likely to you?
RICHERSON: Well, one of the observations that people make of hunter-gatherer subsistence camps is they're tightly crowded together. They don't live miles away from each other.
BOYD: In a given camp, there are often 50 people.
WALKER: Living in each other's hair, so to speak.
RICHERSON: Yeah.
BOYD: Cheek by jowl. Denser than the village.
WALKER: Well, I guess that largely dispenses with that hypothesis.
BOYD: Yeah, it depends. They could be averaging. They're quite mobile … I don't know. I'm sceptical of those kind of arguments.
4. Have cities always reduced fertility?
BOYD: One of the predictions that I would love to see somebody test would be: I'd predict the same thing for urbanisation expansions in the past. So if we could go back to Tenochtitlan or Rome and get good measures of fertility, I'd predict the urban people would get sucked up into the same thing because they have these wider social networks. Prestige networks are …
WALKER: I think Rome was really good at just continuing to suck the rural population in.
BOYD: All those cities were, because they were death traps.
RICHERSON: Moving to the city was a little bit like a suicide.
WALKER: Still is, to an extent.
BOYD: Still is. I'd like to see somebody … There is demographic data for some of those places. Somebody needs to really work on that …
RICHERSON: Isn't there some decent data for early modern or mediaeval jews in Italy? That they underwent a precocious demographic transition …
BOYD: They would be exactly who you predict.
5. "AI will be cultural"
WALKER: Does your understanding of how human intelligence evolved give you any unique insights into how artificial intelligence might be created?
RICHERSON: Well, I attended a conference on AI about six weeks ago or so, and so I boned up just a little bit. And Alison Gopnik, do you know who she is? She's a developmental psychologist here at Berkeley, and she argues in a paper that artificial intelligence has nailed culture. That artificial intelligence, these large language models have made the accumulated wisdom of the world available in a very efficient way.
Now, it's also made the bullshit factor in culture equally exaggerated, right? So that's why the large language models have these flaws, because they're tapping into the craziness of culture as well as the sensible fraction of it. And she argues that … We also have learning models, individual-like learning models, models of innovation, but those aren't integrated with the large language models. And the kinds of success-based filters that we think are so important in making culture adaptive, it's not clear that they're built into the current thinking on AI. And other people make the point that AI has this problem that it's computationally extremely expensive.
If we say, with Gopnik, that AI has aced the cultural part of it. What about the cultural transmission kind of part of it? What about the innovative part of it? What about the quality filters part of it? Can we integrate those into AI without making it prohibitively expensive?
Very unschooled thoughts.
WALKER: Interesting, nonetheless. Do you have anything to add to that, Rob?
BOYD: AI will be cultural. It's got to be. But how it'll all work is a mystery to me.