I really, really do not like this Algernon argument!
Most trivially: Evolution optimises for inclusive genetic fitness, not intelligence: if some simple solution/intervention made us more intelligent but less reproductively fit (let's say some chemical that's both a nootropic and a contraceptive) that solution would be on the table for us but not for evolution.
Tradeoffs: We don't have the same tradeoffs as evolution. If some solution made us more intelligent but (say) burnt several hundred extra calories per day (brain overclocking..?) evolution would reject that solution but we would gladly accept it. Likewise some solution (some weird blood-brain barrier chemistry maybe?) that made us fractionally more vulnerable to pathogens that are much less scary today than they would have been 10,000 years ago, or any of dozens of other things that were deal-breakers (as it were!) for evolution.
Gradient ascent: Evolution doesn't traverse the entire solution-space: it only traverses those areas of the solution-space that are accessible via simple gradient ascent. If there's some better design that's very close to our current design within solution-space but there's a (sufficiently deep) fitness trough in between the two designs, evolution won't find the better design. We can, though, since we're not limited to searching only by gradient ascent!
Other domains: We improve upon evolution through simple solutions all the time. If you had a sore foot but you had to catch an antelope today or your tribe would go hungry, paracetamol and ibuprofen (both chemically pretty simple) would have come in frightfully handy. If you had to eat immediately after doing something slightly unhygienic (such as killing an antelope with your bare hands..) isopropyl alcohol sanitiser (chemically simple; synthesised entirely from biologically-produced chemicals) would be useful. If simple solutions that evolution didn't find can be shown to exist in other domains, I think the burden of proof is on the Algernon-proponent to show why intelligence is somehow different to anything else that increases inclusive genetic fitness.
The Flynn Effect: Intelligence increased *hugely* over the 19th and 20th Centuries (https://en.m.wikipedia.org/wiki/Flynn_effect) and most of the explanations for this (better nutrition, education, parenting methods, hygiene, etc.) seem to be far less complicated/advanced/impressive than the stuff evolution makes (eyes, hummingbirds, etc.) We seem to have increased intelligence steadily over many generations almost by accident, whilst aiming for other things - the Algernon argument would have us believe this is Impossible!
My favorite theory is that there basically was no "Bronze Age Collapse" as it's presented. There was no single horde of Sea-peoples that destroyed the Mycenae, Hittites, and Assyrians and almost the Egyptians in that order. It was a period of decline over a half-century with far less destruction than is pretended by the people who claim the year 1177 was somehow a pivotal change.
There were of course population migrations from the Mediterranean, and there were some unknown peoples invading the Hittites, Assyria and Egypt, but it was spread over a far longer period of time, and the power of the Sea Peoples themselves is far overstated. More impactful was the internal decay of Assyria and the Hittites which caused their empires to dissolve.
I've seen convinced evidence that some of the sea peoples were from Sardinia. I think the Egyptian word used was Shardish or something, and I've seen claimed the hats depicted in the Egyptian battle scene match archaeological evidence from Sardinia. One thing that's definitely true is that some of the Sea Peoples were from the Mycenaean civilization (eventually becoming the Hellenes or Greeks).
The Peleset were definitely Mycenaean (archaeological and genetic evidence is as conclusive as it gets without writing. Similar religious shrines, near-identical house construction not native to middle east, etc.) and the Peleset were explicitly mentioned in the Egyptian war stela you show above. The Peleset are well known from the Bible as the Philistines that battled with the Jews, who give their name to modern Palestine.
Now who has claims to what in Palestine is a contentious issue. But I think that if we identify the Peleset as a Mycenaean people, and the Mycenaean's became the Classical Greeks, and the Classical Greek civilization was subsumed by the Roman Republic, and we look for the still-extant nation that has the oldest tradition of claiming the spirit of the Roman Republic, we are left with the United States (the founding father's were heavily inspired by Polybius' theories on mixed government). Therefore I think it's legitimate Trump wants to build a casino in Gaza.
My understanding is that this is a wholly respectable theory, not considered fringe at all. (Opinions might vary on the last paragraph) Basically, some historians see the Sea Peoples as a cause of the Bronze Age collapse, and others see them as a symptom, where internal decay made it difficult for the empires to fend off attacks that were always a risk. What I don't quite understand about the latter theories is why different civilizations would suffer from internal decay around the same time. Surely there is some common cause. (Maybe climate or something?)
I’ve always been partial to the idea that coincidences in history just happen, and the explanations we come up with are often just stories that vaguely gesture at an explanation, but in reality weren’t causal.
The expected lifetime of an Empire isn’t exactly that long, and if both the Assyrians and Hittites just happened to be on their last legs, the collapse of one could harm trade enough to collapse the other within a few decades. Maybe there was some climate change involved, maybe the migration of some nomads due to even more nomads farther beyond, or maybe one charismatic or abnormally tall guy in Sardinia got the idea to loot the entire known world, with people piling on as he went. All could be the cause, or it could be that the rulers of both were particularly inept, and then faced a crisis that couldn’t be dealt with.
I basically buy it. I think that people in our corner of the internet especially tend to overrate "trends and forces" and underrate the importance of single individuals and random chance. I was taught in school that WWI was inevitable. But nowadays I tend to think that if Franz Ferdinand's driver hadn't made that wrong turn, the world would probably be very different today.
Thanks for the concept-handle of "land chunks vs. water chunks" — that's helpful!
I know that "slides" is the ordinary nomenclature that geologists use to talk about plates (as you described them: "solid chunks that sort of slide around"), but I've come to think that's an unhelpful image, connoting frictionless motion when what we're talking about is two titanic rocks scraping against one another (the sound of whose scraping are literal earthquakes).
Missing from the popularized images we use to talk about tectonics is a clear sense of what's driving them. "Mantle upwelling" is true, but abstract: I think a better answer is one particular upwelling, which we now see as the biggest volcano in the world — the floor of the Atlantic Ocean.
It began erupting in the Triassic, and split Pangaea apart. It's continuing to erupt now — the Atlantic grows by about an inch a year. Its vent is the mid-Atlantic ridge; Iceland is its peak. It's shoving the shards of Pangaea over the Pacific Ocean — which itself used to be "Panthalassa", the world-sea.
If I understand my geography rightly (and someone please check this), it's growing in length, snaking around Africa into the Indian Ocean, and splitting both north through the Levant (hence the Red Sea and Dead Sea) and also south through Africa (hence the East African Rift, which turned forests into grasslands and brought our ancestors down from the trees).
...now that I think of it, it's remarkable that this thing doesn't have its own Wikipedia page.
You clearly know more about this than me. I felt like "slides" is correct if you look at the process at the right time-scale, but maybe that's wrong. Is the objection that "slides" that it implies a process governed by dynamic friction?
Regarding mantle upwelling, is it correct to think of this as driven by heat buildup under Pangea?
Oh, you're entirely right: "slides" accurately states the nature of the motion between the plates (one over the other, in one direction), and what I dislike about it is the connotations of frictionlessness. (As a kid, I kept wondering why they wouldn't stop moving, and figured the answer must be some abstruse physics thing. Nope! Just the world's biggest volcano.)
As to the cause of Pangaea's split, my understanding is that there are a few theories in play at the moment, but my favorite is the one you gave, of supercontinents acting as hens incubating the eggs of their own destruction.
Not explicitly seeking. But a course of action that leads to greater unhappiness is sub-optimal. (I'm thinking of the Lisa Simpson graph of the inverse correlation between intelligence and happiness.)
Regarding your point about there not being a mental exercise to get you better at cognitive skills in general, I think you might be making a bit of a category error, going from the specific "running more makes you better at running" to the hyper general "thinking exercise makes you better at all thinking". It seems relevant since running doesn't get you better at e.g. power lifting, which doesn't get you better at throwing at a target, which doesn't get you better at swimming, etc. and even running marathons doesn't help get you better at sprinting past a certain low level. Some exercise types get you better at other things incidentally, like better aerobic capacity helps running and swimming, but others are kind of irrelevant while some are actively negative (being able to bench 700lbs probably makes you a worse marathon runner or tennis player). I am also reminded of the difference between lifting weights strength and "functional" strength.
Cognitive skills are probably fairly similar, in that getting better at some (e.g. reading or abstract math) probably help a bit of everything, but different types are specialized enough that there isn't a great deal of overlap past a certain point, and possibly some trade offs once you get to a certain frontier. I expect that much like exercise, once you get above a certain threshold of "not the mental equivalent of a couch potato" we tend to reliably get better only at what we specifically use and work on, with any cross functional benefits being close to zero.
Also, don't stop writing. I need something to read to make me late for work and this is my preferred style of content. :)
I may be wrong about this, but my claim was that if you run, that will in fact make you better at a "decently broad" range of activities (tennis, hiking, standing up, being alive instead of dying of heart disease). It does seem intuitive that cognitive skills should work the same way, but to the best of my knowledge there is no "brain exercise" you can do that's actually established to increase performance in a "decently broad" range of cognitive tasks.
Again, I stress that this may be wrong! The picture you paint in your second paragraph seems very plausible to me. I guess it's all a matter of degree. My reading is that "cognitive exercise" does have somewhat more narrow benefits than "physical exercise" but I don't really have any proof of this other than the total failure of the "brain training" trend some years ago.
(I might look into this more—it's an interesting subject.)
I think your claim is right, at lowish levels of physical fitness. For the average human in the USA running probably improves your performance across all sports that require running, and maybe even a little bit in those that don't (e.g. power lifting). However, that benefit flattens out pretty quickly for almost all activities; basketball players are not running 30 miles a week to train, for instance, and fencers probably don't run much at all. The value of running for most Americans is that most Americans are quite low on the physical fitness curve, running closer to 0 miles a week than say 10.
However, I think the average American is a little further along the mental activity curve. Most Americans read a bit, if only for their jobs and internet slop, we engage in a lot of speaking and negotiating with other people, and probably just watching TV and movies is better than nothing. The mental equivalent of being a couch potato is probably harder than is obvious, basically having an extremely menial job (or no job) and hardly interacting with people or things. Those people certainly exist, and we would probably find that if we got them to read a certain amount per week, or do a certain amount of mathematical problem solving they would show gains in mental activities across the board.
Another way of thinking of it might be going to a tribe of traditional pursuit hunters and seeing if running a bit more would make them better at whatever sports they like to play. Chances are they are all maxed out on running's general benefit, so we would see little gain if any. By contrast the humans in normal western nations can't get much general value from more cognitive exercise because most have largely maxed out the general value already.
I feel like I learned a good amount about shoes, but nothing about sea peoples. Can you further explain the significance of the sea people carving to me please?
The Sea Peoples are one of history's biggest mysteries. (Where did they come from? Why did they suddenly appear and destroy everything?) So I just think it's cool that you can (sort of) see a picture of them in action.
I'm not remotely knowledgeable about this, but the impression I got from a few sources eg https://www.goodreads.com/book/show/17736859-the-story-of-the-human-body was that most of our early adapations were for high-efficiency long distance walking (possibly because the rainforests receded and food source were much further apart) and then the running was much more recent and might have kept improving with a few more million years eg our ACLs should keep getting thicker, some of our pointless foot/ankle bones might fuse together.
The algernon idea always made more sense to me for changes which were easier mutations eg just raising the levels of some neuro-transmitter. But it seems obviously weak for anything more complex or for more recent adaptations. We just haven't been evolving into our niche for that long.
> By reviewing studies of illiterate subjects, we propose specific hypotheses on how the functions of core brain systems are partially reoriented or ‘recycled’ when learning to read. Literacy acquisition improves early visual processing and reorganizes the ventral occipito-temporal pathway: responses to written characters are increased in the left occipito-temporal sulcus, whereas responses to faces shift towards the right hemisphere. Literacy also modifies phonological coding and strengthens the functional and anatomical link between phonemic and graphemic representations. Literacy acquisition therefore provides a remarkable example of how the brain reorganizes to accommodate a novel cultural skill.
One of the researchers, Blaise Dubois, mentions a study where cutting the carbon plate into pieces didn't reduce the performance of the shoe https://youtube.com/watch?v=lRtEr1NMnew&t=5058
(Sorry the link is in French)
Unfortunately I can't find the study
It seems like most of the benefit of modern running shoes is the efficiency of the foam sole, which is only interesting because of the thickness
Sole thickness does seem quite helpful (when did you last see a top-level marathon runner with minimalist shoes?)
Now, I don't think having a hugely thick sole is a Pareto improvement. Not only does the large lever arm probably increase the risk of ankle or knee injury (reminding me of the horse metaphor in this post https://www.psychiatrymargins.com/p/schizophrenia-is-the-price-we-pay), it's probably not the best for quick direction changes (which would be why tennis shoes have thinner soles?).
I like minimalist shoes not because they are optimal in any situation (they're probably slower on flat ground, you have to be more precise on some parkour jumps and some technical trails), but because they work okay for a wide range of activities
But I bet that for most activity they are a nuisance
I'm not sure those are knockdown arguments for the Algernon principle. For example, I'm pretty sure we're close to having powered exoskeletons that improve performance across the board. Is an external power source cheating? If so, what about a genetic edit that increases performance but also makes you eat twice as much? Is that cheating?
I feel like the reference class is a hard thing to pin down.
I wouldn't call it cheating, but I do think that Algernon-type reasoning would suggest that you shouldn't be too surprised if something that adds extra energy (or food) to the equation leads to big improvements. The argument I'd make for that is simply that evolution has strongly optimized us to reduce energy expenditures. So even if you really buy into Algernon, it only says that we should be somewhere on the "energy x {fast, smart} pareto curve", not that we should be maximally fast or smart.
Yeah, I guess a similar argument applies to that, too. If the shoes made you better at all types of running at the same time that would be a much stronger refutation of Algernon-type reasoning than if they make you better at one type of running and worse at others. It seems that these shoes are more like the latter, making you better at long-distance but worse at sprinting. So I think it's reasonable to argue that Algernon passes the "shoes test" even better than I implied in the post.
Interesting claim: "But there doesn’t seem to be any cognitive task that you can practice and make yourself better at other cognitive tasks.". I have felt at least anecdotally that critical thinking problems, puzzles, debate, etc. give experience that generalizes relatively well to other cognitive tasks. It seems to be a point of disagreement between other commenters like @Blattyman and @DashWieland, curious to know what led to your initial position on this and what might change your mind?
Truthfully, it's a weakly held view, and I would probably shift my position in response to a strong gust of wind. When I wrote that, I was thinking of a proposition like, "There is no known brain-training exercise that will reliably increase your score on an IQ test." I *think* that's true? But I also agree with you that, anecdotally, people who spend a lot of time doing cognitive tasks seem to get better at doing cognitive tasks. I'm not entirely sure how to resolve the tension between those two issues,
The other relevant thing re: shoes is that you can take them on and off. Which is not really an option for evolution, at least for anything structural. So if there are downsides to having the plates equipped at all times, then you're right back to the Algernon argument.
What's the cognitive equivalent of that? Maybe stimulants? I'm pretty sure the performance improvements are zero-sum over time, but still useful in that you can choose when is the most convenient time to get a smartness boost.
PS: your blog is a pure delight. I'm finding a lot of the value of using LLMs is in thinking of interesting questions to ask (how to square supershoes with the Algernon effect?). So there is unlikely to be a world where I could get Dynomight-quality research and whimsy on demand, which is sad for me, but good news for you.
I agree with that argument re: shoes. After a brief review, it seems to me that the carbon-fiber plates are in fact known to be harmful for sprinting. That makes the Algernon argument survive the "shoes test" even better than it already does.
Regarding stimulants, I'd actually bet that they aren't zero-sum over time. That is, I'd bet that if you take (say) 100 mg of caffeine every day for the rest of your life, 20 years from now your adenosine receptors are still somewhat less active than if you never took any caffeine at all. (Though you could still be neutral or even net-negative if you consider sleep and so on.) I have essentially zero evidence for this belief, but for some reason it's a strong intuition.
Man that would be cool if you're right. I wonder how we'd go about trying to figure that out? Vaguely remember a gwern post on the subject, or maybe it was the algernon one itself.
It's quite hard to test because you need a long cycle even to get one datapoint, plus how do you measure the adenosine receptors? (And do it all blinded and placebo-controlled.) I think this makes it borderline-impossible to do as a self-experiment.
Trying to think of disadvantages of supershoes: they're really tall, which probably makes them a bit riskier for stability, especially on uneven terrain. Also, I think it's actually the special high-tech foam that's doing as much work as the plate, and that wears out really quickly. I went to a shoe shop and the clerk showed me a single-use marathon shoe! So not sure if evolution could come up with a biological tissue with the same properties that doesn't have to be disposed of after n kilometres.
That meta analysis compared to traditional running shoes, which may be worse than barefoot running (my knowledge admittedly stops at having really enjoyed Born to Run so I’m not sure what the state of the literature is currently but I wonder how the tarahumara would feel about the new shoes).
Also worth noting re the Algernon argument that our feet are not just adapted to marathons, they’re optimized for many things at once (endurance running, jumping, sprinting, turning, going on uneven surfaces as opposed to paved road, etc). Therefore even if new running shoes are in fact superior to barefoot that’s not too surprising
I think it's fun and interesting that long-distance triathletes are now much faster than they ever have been, not primarily because of increased aerodynamic testing/design of bikes/clothing, better carbon running shoes, power meters + better data tracking + better training principles, but simply because triathletes realized you could eat 3x the amount of sugar per hour than was previously thought beneficial.
I'm not sure if there is a good report on this, but my sense is that the increase in carb consumption started around 2022 and became really cemented in the last two years. I believe this was led by "lowercase science" rather than "Science™"—just people experimenting. (Perhaps it was after Kristian won gold in Tokyo and the Norwegians moved over to long course that they were the first ones to figure it out). The old belief was that one could consume 60-90g of carbs per hour, but now everyone is consuming at-least 120g, with 180-200 being quite common. The biggest impact is in bike power expenditure. In the last ten years, watts expended during an Ironman bike ride have increased by something like 30% for champion contenders. Note, this is just watts, not times, which had started to reduce before this as aero gains were happening before the increase in carb consumption. It's also worth noting that run times saw a bigger improvement after the increase in carb consumption rather than immediately after the introduction of carbon shoes.
You can see how race times have evolved over time by clicking through links like this:
Very interesting, thanks. One thought that occurs to me is that large improvements from from consuming dramatically more energy seems more compatible with Algernon than large improvements from using different shoes, since we evolved under strong pressure to minimize energy expenditures. It's also interesting that the benefits would be larger in biking than running, i.e. in the less optimized activity.
In a web increasingly full of AI slop, I am seeking more content like yours, not less! Keep doing what you do best, and I will keep reading.
I also want to comment on this claim:
"there doesn’t seem to be any cognitive task that you can practice and make yourself better at other cognitive tasks."
While this is true for fluid intelligence (e.g., abstract reasoning, spatial ability), it is not strictly true for crystallised intelligence (e.g., general knowledge, verbal knowledge). Research shows that crystallised intelligence typically increases into one's 50s or even 60s before it begins to decline. This pattern occurs because we continually acquire new knowledge and can reliably retrieve it from memory. In theory, someone who devotes much more time to learning should develop higher abilities that fall under crystallized intelligence than someone who devotes much less time.
I think of the idea of crystallized intelligence as a summation across all your cognitive skills, so if you learn more skills or some skills more deeply, you are considered to have higher crystallized intelligence. But I don't think that concept addresses Dynomight's point about whether you can have spillover from one cognitive task to others. Having said that, my prior is that this happens all the time. How can you not have spillover from common activities like talking, reading, writing, and analyzing problems to other things? People say that socializing is protective of cognitive ability as you get older, and I bet that a big part of the reason is that you have to carry on a conversation when you're socializing, or at least be ready to respond to people with more than a grunt.
Crystallised intelligence is primarily accumulated knowledge and language comprehension. I agree there’s a lot of near transfer where practicing one thing strengthens related abilities. If you read a lot, your vocabulary, general knowledge, and language comprehension should all improve, and sustained learning may improve some brain functioning. But that’s different from far transfer, where training one task boosts dissimilar abilities.
The running analogy fits. Training makes you better at running (and probably a bit better at walking or jumping), and improves general fitness. But running doesn’t make you a faster reader. Similarly, reading won’t raise fluid intelligence like abstract reasoning or make you a faster runner.
My point is, we see a similar pattern across physical and cognitive domains: practice spills over to related tasks but far-transfer effects are rare.
My Spatial abilities used to be fairly mediocre (maybe above average since I was good at math which I mostly visualize geometrically) until I started a job that involved a lot of CAD and mechanism design.
Now I feel like i’m a much stronger shape rotator than I used to be, I’m not sure if this is from combination of more crystallized intelligence or an increase in fluidized though, or maybe just aging to the peak fluidized intelligence age around that time.
"Or, perhaps thinking is just different from running. If you start running, you will get better at it, in a way that spills over into lots of other physical abilities. But there doesn’t seem to be any cognitive task that you can practice and make yourself better at other cognitive tasks."
This is a very fun prompt. In chatting about it with some folks around me we think there are actually cognitive tasks where practice generalizes. E.g., reading and synthesizing complex texts might make you better at ingesting and condensing information in general, which is useful outside of specifically reading complex texts. However, someone in the group noted that this only feels true for them if they're actively doing the meta-cognitive task of thinking about how their new strategies for ingesting and condensing information can be generalized. That without the conscious effort to generalize, the cognitive skill does remain isolated to, in our example, reading Kant, or whatever.
So maybe the difference is the physical body automates the generalization of the skill acquisition, while the mind requires more intervention to get generalizable skill?
Good point! As I was writing that, somewhere in the back of my head was also the idea that staying mentally active seems important later in life. That seems like decent evidence that "cognitive exercise" can at least do *something*.
I feel like this is actually the kind of blog post that survives because it's about a selection of things I didn't intend to think about today but find pleasing, which is probably going to be easier for a human to put together than a LLM for some time yet, especially if we keep training them to be bland sycophants because that's what most people want for most purposes.
In fact, I think I do believe that the "curation" aspect of this kind of post will continue to be useful for at least a while. But I'm less sure of the "explainer" aspect.
For example, I wonder how many people saw the section on super-continents and thought "looks interesting" and then instead of reading the rest of the section, went over to their favorite LLM and typed in "explain supercontinent cycles."
(To be clear, there are quite valid reasons someone might do that. It's very hard to beat a personalized/interactive explanation with a single static block of text!)
Use of LLMs may be less generalised than you imagine. I don't use one and nobody in my social circle uses one. Only at work do I find a boss whose go to is an LLM and my research consistently turns up better results than his ChatGpt, so I'm not likely to be finding any use for one soon.
I guess you could structure it sort of like a links post, but instead of linking to articles, you just give LLM prompts? I find that a bit depressing, but it does seem plausibly useful.
Why depressing? Conversations with my AI that I've enjoyed:
1. Who do you think are the objects of this statement by me: "In their commitment, in their fidelity, in their vision, in their execution, they're our superiors"?
2. What would make me 50% happier?
3. What conversation that we've had do you think I found the most interesting? How could you tell?
4. In the context of optimization, evolution, and AI stochastic gradient descent, I'd like to explore the tension between a Constraint and an Objective.
TBF I would never use a current LLM for that kind of thing because without having surrounding background knowledge I expect it would be plausible and wrong and pollute my mental store of facts at a much higher rate than a blogger that I understand the context of and thus have a better idea on how much trust to exercise in different areas.
I really, really do not like this Algernon argument!
Most trivially: Evolution optimises for inclusive genetic fitness, not intelligence: if some simple solution/intervention made us more intelligent but less reproductively fit (let's say some chemical that's both a nootropic and a contraceptive) that solution would be on the table for us but not for evolution.
Tradeoffs: We don't have the same tradeoffs as evolution. If some solution made us more intelligent but (say) burnt several hundred extra calories per day (brain overclocking..?) evolution would reject that solution but we would gladly accept it. Likewise some solution (some weird blood-brain barrier chemistry maybe?) that made us fractionally more vulnerable to pathogens that are much less scary today than they would have been 10,000 years ago, or any of dozens of other things that were deal-breakers (as it were!) for evolution.
Gradient ascent: Evolution doesn't traverse the entire solution-space: it only traverses those areas of the solution-space that are accessible via simple gradient ascent. If there's some better design that's very close to our current design within solution-space but there's a (sufficiently deep) fitness trough in between the two designs, evolution won't find the better design. We can, though, since we're not limited to searching only by gradient ascent!
Other domains: We improve upon evolution through simple solutions all the time. If you had a sore foot but you had to catch an antelope today or your tribe would go hungry, paracetamol and ibuprofen (both chemically pretty simple) would have come in frightfully handy. If you had to eat immediately after doing something slightly unhygienic (such as killing an antelope with your bare hands..) isopropyl alcohol sanitiser (chemically simple; synthesised entirely from biologically-produced chemicals) would be useful. If simple solutions that evolution didn't find can be shown to exist in other domains, I think the burden of proof is on the Algernon-proponent to show why intelligence is somehow different to anything else that increases inclusive genetic fitness.
The Flynn Effect: Intelligence increased *hugely* over the 19th and 20th Centuries (https://en.m.wikipedia.org/wiki/Flynn_effect) and most of the explanations for this (better nutrition, education, parenting methods, hygiene, etc.) seem to be far less complicated/advanced/impressive than the stuff evolution makes (eyes, hummingbirds, etc.) We seem to have increased intelligence steadily over many generations almost by accident, whilst aiming for other things - the Algernon argument would have us believe this is Impossible!
My favorite theory is that there basically was no "Bronze Age Collapse" as it's presented. There was no single horde of Sea-peoples that destroyed the Mycenae, Hittites, and Assyrians and almost the Egyptians in that order. It was a period of decline over a half-century with far less destruction than is pretended by the people who claim the year 1177 was somehow a pivotal change.
There were of course population migrations from the Mediterranean, and there were some unknown peoples invading the Hittites, Assyria and Egypt, but it was spread over a far longer period of time, and the power of the Sea Peoples themselves is far overstated. More impactful was the internal decay of Assyria and the Hittites which caused their empires to dissolve.
I've seen convinced evidence that some of the sea peoples were from Sardinia. I think the Egyptian word used was Shardish or something, and I've seen claimed the hats depicted in the Egyptian battle scene match archaeological evidence from Sardinia. One thing that's definitely true is that some of the Sea Peoples were from the Mycenaean civilization (eventually becoming the Hellenes or Greeks).
Good video on this: https://www.youtube.com/watch?v=s-J8VGFG1Bg
The Peleset were definitely Mycenaean (archaeological and genetic evidence is as conclusive as it gets without writing. Similar religious shrines, near-identical house construction not native to middle east, etc.) and the Peleset were explicitly mentioned in the Egyptian war stela you show above. The Peleset are well known from the Bible as the Philistines that battled with the Jews, who give their name to modern Palestine.
Now who has claims to what in Palestine is a contentious issue. But I think that if we identify the Peleset as a Mycenaean people, and the Mycenaean's became the Classical Greeks, and the Classical Greek civilization was subsumed by the Roman Republic, and we look for the still-extant nation that has the oldest tradition of claiming the spirit of the Roman Republic, we are left with the United States (the founding father's were heavily inspired by Polybius' theories on mixed government). Therefore I think it's legitimate Trump wants to build a casino in Gaza.
My understanding is that this is a wholly respectable theory, not considered fringe at all. (Opinions might vary on the last paragraph) Basically, some historians see the Sea Peoples as a cause of the Bronze Age collapse, and others see them as a symptom, where internal decay made it difficult for the empires to fend off attacks that were always a risk. What I don't quite understand about the latter theories is why different civilizations would suffer from internal decay around the same time. Surely there is some common cause. (Maybe climate or something?)
I’ve always been partial to the idea that coincidences in history just happen, and the explanations we come up with are often just stories that vaguely gesture at an explanation, but in reality weren’t causal.
The expected lifetime of an Empire isn’t exactly that long, and if both the Assyrians and Hittites just happened to be on their last legs, the collapse of one could harm trade enough to collapse the other within a few decades. Maybe there was some climate change involved, maybe the migration of some nomads due to even more nomads farther beyond, or maybe one charismatic or abnormally tall guy in Sardinia got the idea to loot the entire known world, with people piling on as he went. All could be the cause, or it could be that the rulers of both were particularly inept, and then faced a crisis that couldn’t be dealt with.
I basically buy it. I think that people in our corner of the internet especially tend to overrate "trends and forces" and underrate the importance of single individuals and random chance. I was taught in school that WWI was inevitable. But nowadays I tend to think that if Franz Ferdinand's driver hadn't made that wrong turn, the world would probably be very different today.
Thanks for the concept-handle of "land chunks vs. water chunks" — that's helpful!
I know that "slides" is the ordinary nomenclature that geologists use to talk about plates (as you described them: "solid chunks that sort of slide around"), but I've come to think that's an unhelpful image, connoting frictionless motion when what we're talking about is two titanic rocks scraping against one another (the sound of whose scraping are literal earthquakes).
Missing from the popularized images we use to talk about tectonics is a clear sense of what's driving them. "Mantle upwelling" is true, but abstract: I think a better answer is one particular upwelling, which we now see as the biggest volcano in the world — the floor of the Atlantic Ocean.
It began erupting in the Triassic, and split Pangaea apart. It's continuing to erupt now — the Atlantic grows by about an inch a year. Its vent is the mid-Atlantic ridge; Iceland is its peak. It's shoving the shards of Pangaea over the Pacific Ocean — which itself used to be "Panthalassa", the world-sea.
If I understand my geography rightly (and someone please check this), it's growing in length, snaking around Africa into the Indian Ocean, and splitting both north through the Levant (hence the Red Sea and Dead Sea) and also south through Africa (hence the East African Rift, which turned forests into grasslands and brought our ancestors down from the trees).
...now that I think of it, it's remarkable that this thing doesn't have its own Wikipedia page.
You clearly know more about this than me. I felt like "slides" is correct if you look at the process at the right time-scale, but maybe that's wrong. Is the objection that "slides" that it implies a process governed by dynamic friction?
Regarding mantle upwelling, is it correct to think of this as driven by heat buildup under Pangea?
Oh, you're entirely right: "slides" accurately states the nature of the motion between the plates (one over the other, in one direction), and what I dislike about it is the connotations of frictionlessness. (As a kid, I kept wondering why they wouldn't stop moving, and figured the answer must be some abstruse physics thing. Nope! Just the world's biggest volcano.)
As to the cause of Pangaea's split, my understanding is that there are a few theories in play at the moment, but my favorite is the one you gave, of supercontinents acting as hens incubating the eggs of their own destruction.
Would you be happier if you were 2.7% smarter?
Strongly reject the implied premise that the best way to live life is to explicitly seek happiness!
Not explicitly seeking. But a course of action that leads to greater unhappiness is sub-optimal. (I'm thinking of the Lisa Simpson graph of the inverse correlation between intelligence and happiness.)
Regarding your point about there not being a mental exercise to get you better at cognitive skills in general, I think you might be making a bit of a category error, going from the specific "running more makes you better at running" to the hyper general "thinking exercise makes you better at all thinking". It seems relevant since running doesn't get you better at e.g. power lifting, which doesn't get you better at throwing at a target, which doesn't get you better at swimming, etc. and even running marathons doesn't help get you better at sprinting past a certain low level. Some exercise types get you better at other things incidentally, like better aerobic capacity helps running and swimming, but others are kind of irrelevant while some are actively negative (being able to bench 700lbs probably makes you a worse marathon runner or tennis player). I am also reminded of the difference between lifting weights strength and "functional" strength.
Cognitive skills are probably fairly similar, in that getting better at some (e.g. reading or abstract math) probably help a bit of everything, but different types are specialized enough that there isn't a great deal of overlap past a certain point, and possibly some trade offs once you get to a certain frontier. I expect that much like exercise, once you get above a certain threshold of "not the mental equivalent of a couch potato" we tend to reliably get better only at what we specifically use and work on, with any cross functional benefits being close to zero.
Also, don't stop writing. I need something to read to make me late for work and this is my preferred style of content. :)
I may be wrong about this, but my claim was that if you run, that will in fact make you better at a "decently broad" range of activities (tennis, hiking, standing up, being alive instead of dying of heart disease). It does seem intuitive that cognitive skills should work the same way, but to the best of my knowledge there is no "brain exercise" you can do that's actually established to increase performance in a "decently broad" range of cognitive tasks.
Again, I stress that this may be wrong! The picture you paint in your second paragraph seems very plausible to me. I guess it's all a matter of degree. My reading is that "cognitive exercise" does have somewhat more narrow benefits than "physical exercise" but I don't really have any proof of this other than the total failure of the "brain training" trend some years ago.
(I might look into this more—it's an interesting subject.)
I think your claim is right, at lowish levels of physical fitness. For the average human in the USA running probably improves your performance across all sports that require running, and maybe even a little bit in those that don't (e.g. power lifting). However, that benefit flattens out pretty quickly for almost all activities; basketball players are not running 30 miles a week to train, for instance, and fencers probably don't run much at all. The value of running for most Americans is that most Americans are quite low on the physical fitness curve, running closer to 0 miles a week than say 10.
However, I think the average American is a little further along the mental activity curve. Most Americans read a bit, if only for their jobs and internet slop, we engage in a lot of speaking and negotiating with other people, and probably just watching TV and movies is better than nothing. The mental equivalent of being a couch potato is probably harder than is obvious, basically having an extremely menial job (or no job) and hardly interacting with people or things. Those people certainly exist, and we would probably find that if we got them to read a certain amount per week, or do a certain amount of mathematical problem solving they would show gains in mental activities across the board.
Another way of thinking of it might be going to a tribe of traditional pursuit hunters and seeing if running a bit more would make them better at whatever sports they like to play. Chances are they are all maxed out on running's general benefit, so we would see little gain if any. By contrast the humans in normal western nations can't get much general value from more cognitive exercise because most have largely maxed out the general value already.
I feel like I learned a good amount about shoes, but nothing about sea peoples. Can you further explain the significance of the sea people carving to me please?
The Sea Peoples are one of history's biggest mysteries. (Where did they come from? Why did they suddenly appear and destroy everything?) So I just think it's cool that you can (sort of) see a picture of them in action.
I'm not remotely knowledgeable about this, but the impression I got from a few sources eg https://www.goodreads.com/book/show/17736859-the-story-of-the-human-body was that most of our early adapations were for high-efficiency long distance walking (possibly because the rainforests receded and food source were much further apart) and then the running was much more recent and might have kept improving with a few more million years eg our ACLs should keep getting thicker, some of our pointless foot/ankle bones might fuse together.
The algernon idea always made more sense to me for changes which were easier mutations eg just raising the levels of some neuro-transmitter. But it seems obviously weak for anything more complex or for more recent adaptations. We just haven't been evolving into our niche for that long.
Also, a cool example of cognitive exercise is https://www.acesin.letras.ufrj.br/uploads/2/7/1/1/27113691/dehaene2015.pdf
> By reviewing studies of illiterate subjects, we propose specific hypotheses on how the functions of core brain systems are partially reoriented or ‘recycled’ when learning to read. Literacy acquisition improves early visual processing and reorganizes the ventral occipito-temporal pathway: responses to written characters are increased in the left occipito-temporal sulcus, whereas responses to faces shift towards the right hemisphere. Literacy also modifies phonological coding and strengthens the functional and anatomical link between phonemic and graphemic representations. Literacy acquisition therefore provides a remarkable example of how the brain reorganizes to accommodate a novel cultural skill.
Apparently the data are not that conclusive on the carbone plate shoes! https://researchcommons.waikato.ac.nz/entities/publication/383f0b95-d763-48dd-b157-6597ad1bdb2c
One of the researchers, Blaise Dubois, mentions a study where cutting the carbon plate into pieces didn't reduce the performance of the shoe https://youtube.com/watch?v=lRtEr1NMnew&t=5058
(Sorry the link is in French)
Unfortunately I can't find the study
It seems like most of the benefit of modern running shoes is the efficiency of the foam sole, which is only interesting because of the thickness
Sole thickness does seem quite helpful (when did you last see a top-level marathon runner with minimalist shoes?)
And that's probably why it's regulated (page 15 of this document: https://worldathletics.org/download/download?filename=b723c6b6-7d1f-40ad-8b27-1d3f956c6c99.pdf&urlslug=C2.1A+%E2%80%93+Athletic+Shoe+Regulations+(effective+from+01+January+2022)), we don't want to watch people running on quasi-stilts.
Now, I don't think having a hugely thick sole is a Pareto improvement. Not only does the large lever arm probably increase the risk of ankle or knee injury (reminding me of the horse metaphor in this post https://www.psychiatrymargins.com/p/schizophrenia-is-the-price-we-pay), it's probably not the best for quick direction changes (which would be why tennis shoes have thinner soles?).
I like minimalist shoes not because they are optimal in any situation (they're probably slower on flat ground, you have to be more precise on some parkour jumps and some technical trails), but because they work okay for a wide range of activities
Which makes sense evolutionarily
I don't know much about non-powered exoskeletons, apparently they can be helpful for running or walking (https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-021-00893-5)
But I bet that for most activity they are a nuisance
I'm not sure those are knockdown arguments for the Algernon principle. For example, I'm pretty sure we're close to having powered exoskeletons that improve performance across the board. Is an external power source cheating? If so, what about a genetic edit that increases performance but also makes you eat twice as much? Is that cheating?
I feel like the reference class is a hard thing to pin down.
I wouldn't call it cheating, but I do think that Algernon-type reasoning would suggest that you shouldn't be too surprised if something that adds extra energy (or food) to the equation leads to big improvements. The argument I'd make for that is simply that evolution has strongly optimized us to reduce energy expenditures. So even if you really buy into Algernon, it only says that we should be somewhere on the "energy x {fast, smart} pareto curve", not that we should be maximally fast or smart.
Yes I saw the other comment later, good point
But we still need to account for diversity of action right? A train is probably much more efficient than an animal
Yeah, I guess a similar argument applies to that, too. If the shoes made you better at all types of running at the same time that would be a much stronger refutation of Algernon-type reasoning than if they make you better at one type of running and worse at others. It seems that these shoes are more like the latter, making you better at long-distance but worse at sprinting. So I think it's reasonable to argue that Algernon passes the "shoes test" even better than I implied in the post.
Interesting claim: "But there doesn’t seem to be any cognitive task that you can practice and make yourself better at other cognitive tasks.". I have felt at least anecdotally that critical thinking problems, puzzles, debate, etc. give experience that generalizes relatively well to other cognitive tasks. It seems to be a point of disagreement between other commenters like @Blattyman and @DashWieland, curious to know what led to your initial position on this and what might change your mind?
Truthfully, it's a weakly held view, and I would probably shift my position in response to a strong gust of wind. When I wrote that, I was thinking of a proposition like, "There is no known brain-training exercise that will reliably increase your score on an IQ test." I *think* that's true? But I also agree with you that, anecdotally, people who spend a lot of time doing cognitive tasks seem to get better at doing cognitive tasks. I'm not entirely sure how to resolve the tension between those two issues,
Plenty to read on here https://gwern.net/dnb-faq#iq-tests
The other relevant thing re: shoes is that you can take them on and off. Which is not really an option for evolution, at least for anything structural. So if there are downsides to having the plates equipped at all times, then you're right back to the Algernon argument.
What's the cognitive equivalent of that? Maybe stimulants? I'm pretty sure the performance improvements are zero-sum over time, but still useful in that you can choose when is the most convenient time to get a smartness boost.
PS: your blog is a pure delight. I'm finding a lot of the value of using LLMs is in thinking of interesting questions to ask (how to square supershoes with the Algernon effect?). So there is unlikely to be a world where I could get Dynomight-quality research and whimsy on demand, which is sad for me, but good news for you.
I agree with that argument re: shoes. After a brief review, it seems to me that the carbon-fiber plates are in fact known to be harmful for sprinting. That makes the Algernon argument survive the "shoes test" even better than it already does.
Regarding stimulants, I'd actually bet that they aren't zero-sum over time. That is, I'd bet that if you take (say) 100 mg of caffeine every day for the rest of your life, 20 years from now your adenosine receptors are still somewhat less active than if you never took any caffeine at all. (Though you could still be neutral or even net-negative if you consider sleep and so on.) I have essentially zero evidence for this belief, but for some reason it's a strong intuition.
Regarding the blog, thank you!
Man that would be cool if you're right. I wonder how we'd go about trying to figure that out? Vaguely remember a gwern post on the subject, or maybe it was the algernon one itself.
It's quite hard to test because you need a long cycle even to get one datapoint, plus how do you measure the adenosine receptors? (And do it all blinded and placebo-controlled.) I think this makes it borderline-impossible to do as a self-experiment.
Trying to think of disadvantages of supershoes: they're really tall, which probably makes them a bit riskier for stability, especially on uneven terrain. Also, I think it's actually the special high-tech foam that's doing as much work as the plate, and that wears out really quickly. I went to a shoe shop and the clerk showed me a single-use marathon shoe! So not sure if evolution could come up with a biological tissue with the same properties that doesn't have to be disposed of after n kilometres.
Pangea Ultima sounds like a great concept for a movie and/or video game.
That meta analysis compared to traditional running shoes, which may be worse than barefoot running (my knowledge admittedly stops at having really enjoyed Born to Run so I’m not sure what the state of the literature is currently but I wonder how the tarahumara would feel about the new shoes).
Also worth noting re the Algernon argument that our feet are not just adapted to marathons, they’re optimized for many things at once (endurance running, jumping, sprinting, turning, going on uneven surfaces as opposed to paved road, etc). Therefore even if new running shoes are in fact superior to barefoot that’s not too surprising
Good point! To the best of my knowledge, those carbon-fiber plates are actively harmful for sprinting.
NON-SEQUITUR COMMENT:
I think it's fun and interesting that long-distance triathletes are now much faster than they ever have been, not primarily because of increased aerodynamic testing/design of bikes/clothing, better carbon running shoes, power meters + better data tracking + better training principles, but simply because triathletes realized you could eat 3x the amount of sugar per hour than was previously thought beneficial.
I'd love to read more about this. Who figured this out? How? When? How much did it speed people up?
(I looked a bit but couldn't find anything obvious.)
I'm not sure if there is a good report on this, but my sense is that the increase in carb consumption started around 2022 and became really cemented in the last two years. I believe this was led by "lowercase science" rather than "Science™"—just people experimenting. (Perhaps it was after Kristian won gold in Tokyo and the Norwegians moved over to long course that they were the first ones to figure it out). The old belief was that one could consume 60-90g of carbs per hour, but now everyone is consuming at-least 120g, with 180-200 being quite common. The biggest impact is in bike power expenditure. In the last ten years, watts expended during an Ironman bike ride have increased by something like 30% for champion contenders. Note, this is just watts, not times, which had started to reduce before this as aero gains were happening before the increase in carb consumption. It's also worth noting that run times saw a bigger improvement after the increase in carb consumption rather than immediately after the introduction of carbon shoes.
You can see how race times have evolved over time by clicking through links like this:
https://stats.protriathletes.org/race/challenge-roth
https://stats.protriathletes.org/race/im-hawaii
Very interesting, thanks. One thought that occurs to me is that large improvements from from consuming dramatically more energy seems more compatible with Algernon than large improvements from using different shoes, since we evolved under strong pressure to minimize energy expenditures. It's also interesting that the benefits would be larger in biking than running, i.e. in the less optimized activity.
In a web increasingly full of AI slop, I am seeking more content like yours, not less! Keep doing what you do best, and I will keep reading.
I also want to comment on this claim:
"there doesn’t seem to be any cognitive task that you can practice and make yourself better at other cognitive tasks."
While this is true for fluid intelligence (e.g., abstract reasoning, spatial ability), it is not strictly true for crystallised intelligence (e.g., general knowledge, verbal knowledge). Research shows that crystallised intelligence typically increases into one's 50s or even 60s before it begins to decline. This pattern occurs because we continually acquire new knowledge and can reliably retrieve it from memory. In theory, someone who devotes much more time to learning should develop higher abilities that fall under crystallized intelligence than someone who devotes much less time.
As someone with a dreadful memory, this description of crystalized intelligence does not make me optimistic about my prospects while aging!
I think of the idea of crystallized intelligence as a summation across all your cognitive skills, so if you learn more skills or some skills more deeply, you are considered to have higher crystallized intelligence. But I don't think that concept addresses Dynomight's point about whether you can have spillover from one cognitive task to others. Having said that, my prior is that this happens all the time. How can you not have spillover from common activities like talking, reading, writing, and analyzing problems to other things? People say that socializing is protective of cognitive ability as you get older, and I bet that a big part of the reason is that you have to carry on a conversation when you're socializing, or at least be ready to respond to people with more than a grunt.
Crystallised intelligence is primarily accumulated knowledge and language comprehension. I agree there’s a lot of near transfer where practicing one thing strengthens related abilities. If you read a lot, your vocabulary, general knowledge, and language comprehension should all improve, and sustained learning may improve some brain functioning. But that’s different from far transfer, where training one task boosts dissimilar abilities.
The running analogy fits. Training makes you better at running (and probably a bit better at walking or jumping), and improves general fitness. But running doesn’t make you a faster reader. Similarly, reading won’t raise fluid intelligence like abstract reasoning or make you a faster runner.
My point is, we see a similar pattern across physical and cognitive domains: practice spills over to related tasks but far-transfer effects are rare.
Also want to add as a data point:
My Spatial abilities used to be fairly mediocre (maybe above average since I was good at math which I mostly visualize geometrically) until I started a job that involved a lot of CAD and mechanism design.
Now I feel like i’m a much stronger shape rotator than I used to be, I’m not sure if this is from combination of more crystallized intelligence or an increase in fluidized though, or maybe just aging to the peak fluidized intelligence age around that time.
"Or, perhaps thinking is just different from running. If you start running, you will get better at it, in a way that spills over into lots of other physical abilities. But there doesn’t seem to be any cognitive task that you can practice and make yourself better at other cognitive tasks."
This is a very fun prompt. In chatting about it with some folks around me we think there are actually cognitive tasks where practice generalizes. E.g., reading and synthesizing complex texts might make you better at ingesting and condensing information in general, which is useful outside of specifically reading complex texts. However, someone in the group noted that this only feels true for them if they're actively doing the meta-cognitive task of thinking about how their new strategies for ingesting and condensing information can be generalized. That without the conscious effort to generalize, the cognitive skill does remain isolated to, in our example, reading Kant, or whatever.
So maybe the difference is the physical body automates the generalization of the skill acquisition, while the mind requires more intervention to get generalizable skill?
Thanks for the conversation starter :)
Good point! As I was writing that, somewhere in the back of my head was also the idea that staying mentally active seems important later in life. That seems like decent evidence that "cognitive exercise" can at least do *something*.
I feel like this is actually the kind of blog post that survives because it's about a selection of things I didn't intend to think about today but find pleasing, which is probably going to be easier for a human to put together than a LLM for some time yet, especially if we keep training them to be bland sycophants because that's what most people want for most purposes.
I want to believe!
In fact, I think I do believe that the "curation" aspect of this kind of post will continue to be useful for at least a while. But I'm less sure of the "explainer" aspect.
For example, I wonder how many people saw the section on super-continents and thought "looks interesting" and then instead of reading the rest of the section, went over to their favorite LLM and typed in "explain supercontinent cycles."
(To be clear, there are quite valid reasons someone might do that. It's very hard to beat a personalized/interactive explanation with a single static block of text!)
Use of LLMs may be less generalised than you imagine. I don't use one and nobody in my social circle uses one. Only at work do I find a boss whose go to is an LLM and my research consistently turns up better results than his ChatGpt, so I'm not likely to be finding any use for one soon.
What does a "curation" look like without the explanations, in a world of going to one's favourite LLM?
Are you planning any posts with collections of great things with which to prompt your favourite LLM?
... I think I'd like that.
I guess you could structure it sort of like a links post, but instead of linking to articles, you just give LLM prompts? I find that a bit depressing, but it does seem plausibly useful.
Why depressing? Conversations with my AI that I've enjoyed:
1. Who do you think are the objects of this statement by me: "In their commitment, in their fidelity, in their vision, in their execution, they're our superiors"?
2. What would make me 50% happier?
3. What conversation that we've had do you think I found the most interesting? How could you tell?
4. In the context of optimization, evolution, and AI stochastic gradient descent, I'd like to explore the tension between a Constraint and an Objective.
Oh, sure. It's just the idea of sending out a list of such prompts as a post that strikes me as depressing. (Not entirely sure why...)
Because it's absent any humanity and you blog therefore you are.
TBF I would never use a current LLM for that kind of thing because without having surrounding background knowledge I expect it would be plausible and wrong and pollute my mental store of facts at a much higher rate than a blogger that I understand the context of and thus have a better idea on how much trust to exercise in different areas.