>One of my distant cousins won't eat tomatoes. His parents say when he was very young, he bit into a cherry tomato and it exploded into goo in his mouth, and he was so upset he wouldn't eat tomatoes from then on. Now he’s in his 30s and still hates them. Is this fairly described as a “critical window” for food preferences?
Don't we all have a story like this? I ate sriracha as a kid and got food poisoning. I now have a strong negative reaction to the smell of sriracha.
>So at the beginning, you might start in a random place and want to see if 500 miles away is more mountainous. But once your hard work has brought you 1000 feet from Everest’s summit, you don’t want to take a 500 mile jump and end up in New Delhi and have to start all over again.
It seems pretty universal that as things scale upward in mass (and accumulated learning/knowledge can be viewed as mass), they become less "maneuverable". Speedboats vs ocean liners, etc.
Maybe this relates to your language post from yesterday, where a child with a vocabulary of 10,000 words can more easily learn a new language than an adult with a vocabulary of 20,000 words (and an increased number of deep rules that must be discarded, like "adjectives of quantity come before adjectives of color".)
> the AI starts at a very high learning rate (also called “temperature”)
For modern models we're talking about completely different things when we talk about temperature and learning rate. I think the confusion here is probably due to the idea of a learning schedule being similar to simulated annealing.
I think there may also be a different subsystems learn differently aspect here. I’ve personally had food poisoning from a particular fast food restaurant and then not eaten at that entire chain for years. Similarly my girlfriend won’t go anywhere near vicinity of any kind due to a terrible experience in high school with a concerning quantity of coconut rum. I think we have a system that jumps further on bad food experiences and is much much more willing to trap into priors in a negative direction for very understandable survival reasons. E.g. berries that look/smell like this caused me to vomit all of my food up once, therefore I will never eat them again in case they are poisonous.
I think the only way to get past food aversions is to try the food again (even a tiny nibble) a bunch of times, and if those experiences are good, you start liking the food and get over the aversion. My guess is that the tomato thing is less about the specifics of the original experience, or the fact that it was in childhood, and more about your cousin (presumably) deciding each and every day for the subsequent decades to never try cherry tomatoes again. By contrast, you've presumably had lots of positive beach experiences between the scary one and today. (Exposure therapy is obviously related.)
(A useful parenting tip that I got from the book "Bringing Up Bebe" is that it's inordinately important to try to get a kid to eat even just one bite of each kind of food you're serving. After 5 meals in 5 different weeks where they grudgingly eat a single bite of a certain food X, you might find that they have started liking X and now gobble it up.)
I always figured these were something like a manifestation of chaos, in the mathematical sense. When you’re growing everything that comes later depends on what came before. Signals or their absence matter more. There’s not a “go back and redo the last step” feature in most of biology. The best thing to that is our cognition and I think that only comes after it develops right.
How does this programming relate to life long diet, behavior, skills? And how much more effort is required to retrain? We invest on a thesis. We spend 1.7T in US on food, 1.9T on healthcare cost of poor nutrition. You use to go to grocery and were thin, now you get diabetes. We have been trained to like ultraprocessed food even though it's bad for us. We don't invest in behavior change. We do invest in things people want. They want to live longer but are stuck eating the things programmed in their brain in their youth. What is the innovation path of least resistance to better health and behavior change?
As a parent of an extremely premature child, I'm familiar with a few specific physical critical development windows. If you do not practice the swallowing reflex when you are an infant, you will struggle perhaps for the rest of your life (my son was intubated for months). Similarly, the development of your alveoli in your lungs seems to happen primarily in the first 5-7 years of life or so.
These examples don't relate to the idea of critical windows for developing preferences, but I thought I'd offer them as additional concrete examples.
For much of my life I was unable to eat any food located between two slices of bread (by definition, a sandwich), which my mind categorized as revolting and vomit-inducing, even though I like bread and food that comes on a single slice of bread (isn't that illogical?)
This was due to a bad experience in my childhood, not with food in itself, but because it was the food I had to eat during a very stressful period.
Childhood vs adulthood doesn't seem to be rocket science: children simply have less background to compare against. It is like wise old men vs. young naive ones, only the comparison is even worse.
Unlike AI - real intelligences have an ever ongoing comparison and update of input vs. experience. A simple example: if apples fall down - 100 past experiences of apples falling down won't be overwritten by one example of an apple falling up. A child, however, doesn't have 100 past experiences - maybe only has a couple so the fall-up outlier makes much more of an impression.
As for the ongoing fetish with "bad imprint" - the assumption being made is still that furries are just miswiring firing off random chance.
I say assumption because I do not think furries were a thing before television. Note that this is even though children have had stuffed animals for a long, long time; even animated animals in movies only constitute a relative handful of hours per year prior to the advent of television.
It is just as likely that it is the hundreds/thousands/tens of thousands of hours which children spend watching fake talking cartoon animals (basically people but that look like animals) on television that causes the miswiring. In this case, it is this prolonged conditioning in an entirely artificial experience as opposed to some neuron firing too strongly once during a child's viewing of a cute animal.
And if this is so, the problem isn't evolution - the problem is garbage being input into children at high rate:. GIGO. It is actually a testament to evolution that so few children are permanently miswired after such a regime of indoctrination that would make Guantanamo look like a week of spring break. If kids are watching 2000 hours of TV + internet videos a year from age 4 to 14 - that's 20000 hours of potentially bad info going in - and more importantly, is likely a greater source of input than any single other input including parents. Add in 1500 hours of schooling per year/12000 hours from 4 to 14 with a LibsofTikTok type teacher - holy crap.
It should only be a wonder why there aren't MORE furries.
Now consider how either a machine learning or neural network type AI would perform with this diet of input info.
People seriously think evolved intelligence is so easily reproduced? LOL
The two examples you gave could be highly salient/heavily imprinted because they are in areas highly important to reproductive fitness, ie reproduction itself and poison avoidance. Evolutionary pressure over generations is the training data for the genetic matrices that encode our neural matrices.
I think it's clear that formative stimuli like the exploding tomato happen and are not so uncommon. I don't share your puzzlement about why their effects are not more uniform.
* The brain is a large and complex system, and any given stimulus will hit different people differently.
* Processing of the stimulus in the years afterward can also greatly differ, based on subsequent formative stimuli, one's openness to new experience, etc.
It seems to like you're looking for a compact, unified, deterministic explanation of why X produces Y, whereas I don't expect such an explanation to exist in a large, somewhat chaotic system.
I suspect the tomato story is a slight misunderstanding.
My pet theory as to why so many people hate raw tomatoes (quite common in my experience) is something to do with them being vegetables (savoury tasting) masquerading as fruits.
You bite into broccoli or a cucumber and it tastes green and leafy, as you'd expect (so kids don't expect to like it or eat it raw)- you expect a cherry tomato to taste like a cherry or a grape (sweet), and instead it tastes like a raw tomato (savoury, awful). So your cousin wasn't imprinted with the unpleasant experience of a tomato exploding in his mouth, but with the fact that this unpleasant savoury-tasting juice came out, rather than sweet fruity goodness.
Counterexample to the tomato thing: I also thought I didn't like tomatoes since childhood, but that was a belief, not a taste. When someone convinced me to try one at age ~17, and it was good. Now I like tomatoes. Maybe the same would happen with your cousin if they tried one. Maybe the same would happen with me again if I tried a Cheeto (I won't though lol)
We'll learn a lot about possible critical windows in sexuality if and when specific cohorts of children hit puberty and exhibit widespread fetishes for face masks.
As someone that has tried tomatoes as an adult, your cousin made the right choice. "He still won't try tomatoes" is equivalent to "he still won't stick his hand in fire".
Do we know that all these childhood formative experiences are really the cause of anything, or if they are just the first time when something latent becomes salient?
Not "x made me gay" but "I realised i was Born This Way when I saw x".
This reminds me of training dogs. On the one hand they are incredibly trainable (for certain tasks). But they can carry one bad experience through their entire life.
So dogs having this incredible 'learning window' is the same trait that leads them to having 'cherry tomato-style' experiences inprint on them so strongly?
I remember it as if it were yesterday. As a young child I was walking to the shops with my Mum and she gave me a tomato to enjoy. I bit into it but a pip stuck painfully between my two front teeth and couldn't be dislodged without returning home. For a long time I wouldn't eat raw tomato, but now 70 years later I can eat cooked tomato or sliced tomato in salads. I still won't bite into a whole tomato.
Perhaps it is a matter of two influences colliding. The immediate strong event (e.g. the pip getting stuck) and a pre-existing strong aversion/desire (others messing with my teeth).
I didn’t like your ”just so” stories - kinks and fetisches come with incredible variations, so either you have to make up increasingly unlikely explanations, or you have to say that some fetisches can be rationally explained and others can’t, which seems strange. Whereas if you just believe it’s an unusual cross-wiring explained by childhood experiences, anything can fit that model.
Presumably the difference to tomatoes is that you’re _supposed_ to form sexual preferences at a young age and you’re mentally wired for it, but the mind being the messy thing it is it can easily get non-standard. This seems to include homosexuality, and I wouldn’t be the least bit surprised if it’s also how various trans things are formed.
Quoting ’Cryptonomicon’:
”I finally came clean to her about the stockings. It felt good. I'd been reading a lot about how the brain develops and had finally come to accept my stocking kink. It seems that when you are a certain age, somewhere between about two and five years, your mind just gels. The part of it that's responsible for sex becomes set into a pattern that you'll carry with you for the rest of your life. All of the gay people I've ever discussed it with have told me that they knew they were gay, or at least different, years before they even began thinking about sex, and all of them agree that gayness cannot be converted into straightness, or vice versa, no matter how hard you might try.
The part of your brain that handles sex frequently gets cross-wired into other, seemingly irrelevant areas at this age. This is when people pick up an orientation towards sexual dominance or submission, or when a lot of guys pick up highly specific kinks--say, rubber, feathers, or shoes. Some of them are unfortunate enough to get turned on by little kids, and those guys are essentially doomed from that point onwards--there is nothing to do except castrate them or lock them up. No therapy will unkink the brain once it has kinked.
So, all things considered, being turned on by black stockings wasn't such a bad sexual card to have been dealt. I laid this all out to Virginia during the trip home. I was surprised by how calmly she accepted it. I was too big of a jerk to realize that she was thinking about how it all applied to her.”
I doubt the exploding cherry tomato is the reason your cousin dislikes tomatoes. A chocolate truffle exploding in his mouth filling it with delicious goo is unlikely to have put him off chocolate.
It's common to find people who hate the taste of raw tomatoes (I'm one of them). I think it's like Coriander/Cilantro, where there's some compound that some people can taste and others can't.
I pride myself on being able to eat anything so spent many years developing a tolerance to raw tomato flavour, and so can now eat it without gagging but ugh, it's still awful. Even if I do sometimes voluntarily add it to burgers etc to maintain my tolerance. Many people have insisted that oh, I just haven't tried a GOOD tomato, try THIS organic heirloom homegrown flavoursome sweet cherry tomato... Nope they're all bad, sorry.
Re: the "bad childhood experiences cause lifelong specific food aversions", I'd put a personal anecdote on the table that I've never been able to make much sense of.
When I was a kid, I was a fairly picky eater (I guess most kids are, but I seem to have been worse than average - or so the adults told me). And I still vividly remember being totally (and irrationally) grossed out by a couple of specific foodstuffs, to the extent that I would throw up in my mouth if they were just brought near me.
Fast forward a few decades, and some of these are amongst my favourite foodstuffs.
The interesting thing is how the switch happened. In childhood, these foods were super yucky. And at some point during my teenage years, zap!, the aversion was suddenly gone. No laborious process of trying to re-condition myself. Just... seeing a plate of that food, and going "yum! let's have some of this!". If the aversion had not been as strong during childhood, this would not have been as surprising: but I really found these things revolting. And then all of a sudden... tasty!
Also interesting is that this did not happen with all food that my childhood brain considered suspect. A lot of cheese varieties are still off the map, particularly the more biologically active ones. But these are likely a special category, as we probably have some evolution-provided innate additional safeguard in our brains not to eat stuff that visibly has mold growing on it. A safeguard that I haven't been able to deactivate yet.
The term “temperature” usually wouldn’t be used for learning rates. Temperature, especially in the context of LLMs, is a parameter for softmax that determines how sharp you want the probability distribution of predictions to look (roughly, e.g., put 99.99% on the class that’s slightly better than any other or put approximately the same probability on everything almost regardless of the activations), controlling randomness of the output.
Having the learning rate (lr) higher at the beginning of the training and then decreasing it seems to speed up the training compared to having a small learning rate all along, but also, there are all sorts of lr schedulers, including those that decrease and increase the lr in a cycle (no one knows why this sometimes works better, and intuitions about mountains and hill-climbing don’t work well for spaces with millions of dimensions; gradient descent works from the derivative , not from looking around, and the you might be climbing weirdly combined counterintuitive mountains- https://www.lesswrong.com/posts/muLN8GRBdB8NLLX36/visible-loss-landscape-basins-don-t-correspond-to-distinct-algorithms).
People might call decreasing the learning rate “cooling”, but I think most wouldn’t use “temperature”.
There's a rich literature on learning of food preferences. The top level idea that you need to know is that omnivores are evolutionarily sensitized to the effects of food, for the very good reason that a varied diet implies regular encounters with foods that might kill you.
If you play the nasty game of using radiation to make a rat feel nauseous after ingesting some food, it will most likely avoid that food forever. Omnivores can't afford to give foods many second chances. Which is part of the story about tomatoes.
Since sexual experiences and tendencies don't usually kill you, its reasonable to think that evolution will be more tolerant of them. An awful lot of the psychoanalytic stories about childhood experiences with big effects are just stories. They might matter a lot when they become stories that we tell ourselves and come to believe. but that's another topic.
If at age n you experience that some experience E is really bad, the probability per Laplace's rule of succession that the next E will be really bad is 1/n per year. Because n is small, 1/n is large. So you never do it again, and therefore can't update.
I would assume that the reason why formative experiences can have such a drastic effect, like being against eating tomatoes, has almost nothing to do with age specifically and a lot more to do with having little else to judge the experience by.
It's entirely plausible that your thirty-something cousin could bite into a cheery tomato and get squicked out for the first time, but he would have three decades of experience with various species of tomato at that point, which is what lessens the impact of any new information. I would assume that, without experience with e.g. dragonfruit, he could look at the inside of one at any age *for the first time* and justifiably decide it's a horror too gross to deal with. We simply have less experience generally when we're younger, which is why these moments tend to happen in childhood.
You're the psychiatrist, I'm the bumpkin. You tell me if that's not accurate. All I have to go on is logic.
The high learning rate early in life likely does contribute to stronger imprinting/conditioning effects during childhood. Children's brains are still forming connections rapidly, so an intense experience can wire in certain associations more permanently.
Overcoming trapped priors later in life is difficult but not impossible. With enough contrary evidence and conscious effort, even adults can slowly weaken learnt associations. So while imprinting is stronger in childhood, neural plasticity continues through life.
i imagine childhood brain as less full of ruts and knotted pathways that adults have and swimming in BDNF—maybe the plasticity and new pathways that are triggered are done so in a way that is more potent bc there is a relative lack of others to compete with
Does valproic acid reopen critical windows in humans? I remember some excitement about this possibility about a decade ago. Was that just hype generated by a few underpowered studies?
"reacted in some profound way, like becoming a furry."
More likely ... let's back up to Freud, [paraphrased] "there are two kinds of deviation, deviation in aim, and deviation in objective." Is becoming a furry a deviation in objective: 'I tried to become a non-human ideation of an animal.' or is becoming a furry a failure in aim: 'I can't visualize a reasonably attainable human to human sexual model, I see more sexualization in animation than in humans.' I would think for a 'basement dwelling anime addict' the latter is the most likely case.
After writing this, I realized I have followed Freud into the trap of binary thinking, I do apologize for this.
> This still doesn’t explain the unpredictable nature of preference-changing events. My cousin had a bad experience with a tomato in childhood, but was that really his worst childhood experience?
I imagine it's probably something like "I didn't like tomatoes as a kid, so I didn't try them when I was older, so I didn't try them when I was older..." up to the present. Basically, he never had any reason to even gather more evidence about tomatoes being good or bad, so he never did, and never updated.
After your near-drowning, did you experience trepidation the first few times you went back to the beach?
I don't understand the logic here. How is a child who is updating stronger going to be more likely to end up in a trapped prior? If they're more willing to take 100-mile steps, why would we expect them to get trapped at a local minimum/maximum?
I think the implications of this theory on eating would be that children should be very adventurous with food, only to settle later in life on a narrow range of food preferences. That's opposite of reality.
And while there are some cases like that of a weird tomato aversion, that's an exception we shouldn't use to define the rule. My son had this issue with hot dogs, where he ate one while experiencing GI discomfort and decided that hot dogs make him sick. After a few months, he decided kielbasa and corn dogs are fine. I suspect he'll work his way toward the full hot dog experience again, so it definitely seems like a trapped prior. But if I accepted the hypothesis that children take larger step changes than adults, I wouldn't expect his priors to get trapped in the first place.
One other feedback loop to consider is Identity: as a young child I said I didn’t like mayonnaise and people found that surprising, so I became A Person Who Doesn’t Like Mayonnaise. This stuck around until after college when I realized that my preference wasn’t actually grounded in anything and I could switch my mayo preference and no one would care or give me a hard time for it.
For somatic, “feel like I need to vomit” reflexes this might be less of a factor, but for things like “I am American and so speak English” or “I am not a math person” or “I am a Democrat” these identity-based loops from childhood can be impactful and very hard to get out of, and they don’t happen as often to adults.
Naively, wouldn't it make sense for childhood priors to be stronger, just because they arrive without a lot of other signal to balance them out?
Although I wonder if there's something weird about food preference that makes it particularly rigid in children compared to other things. Anecdotally, I was an incredibly, obsessively picky child, and at least compared to how I was then I'm not at all picky now, and I can pinpoint, almost exactly, the age where that shifted for me (around 24) but I could not tell you why at all.
There's a surprising number of people here who dislike tomatoes, including me. Is this worth polling on? Maybe a question on the next ACX survey?
Personally, I think it's the remaining nightshade-like taste on fresh tomatoes that puts me off, and I will happily eat tomatoes once they've been prepared in one way or another and they no longer taste fresh. Maybe this is akin to the cilantro gene and we just have a predisposion against that taste?
> In a typical AI training run, the AI starts at a very high learning rate (also called “temperature”) and gradually “cools” down over time.
Temperature and learning rate are unrelated* in a typical (modern) AI setup. Temperature is a parameter that determines how "spiky" the probability distribution you're using is, or equivalently, how concentrated it is: a heated posterior is very spread-out (high entropy) while a cold one is dominated by a single narrow peak (low entropy).
*Except in the special case of simulated annealing, which is rarely used in modern AI optimization.
> So at the beginning, you might start in a random place and want to see if 500 miles away is more mountainous. But once your hard work has brought you 1000 feet from Everest’s summit, you don’t want to take a 500 mile jump and end up in New Delhi and have to start all over again.
Now this is a *different* algorithm (sounds like adaptive hill climbing, maybe).
Did you ever have an experience with psychedelics? I'm not saying you should because it can be dangerous but at moderate doses that don't send you into alien words or ego deaths it is really interesting to observe your filters (I believe from the default mode network of the brain, which is suppressed by many psychedelics) first hand.
What I mean with this: There are actually quite a lot of things that go through an unconscious filter before your conscious mind gets to process them. These things are not only the often portrayed visual artifacts (intense colors, moving / breathing textures etc.) but also *thoughts*.
That means there are actually thoughts that your filter does not allow you to process at all. Many of these filters are also created during our childhood. Thoughts that scare you, that you are constantly told are wrong or impossible or thought that just did not make sense or help you in any way.
With those filters suppressed, you can have these thoughts again. This can lead to very scary experiences ("bad trip") but also lead you into states of childish joy, wonder and interest about things that you normally do not (actually *can not*) notice.
When I would start feeling nauseous on long car trips as a kid, my parents gave me orange-flavored Dramamine. Associating it with nausea is likely why I hate artificial orange flavoring to this day (although I like real oranges).
I wonder if variation in learning rate/critical period plasticity could explain the genetic correlation between ASD and intelligence.
If genetic variants speed up activity-dependent patterning of primary sensory pathways, it might result a longer period of consistent sensory training inputs for areas involved in higher order integration, during their respective critical window. At the same time, if primary sensory windows close earlier it risks these pathways getting "trapped priors", with downstream brain regions getting comparatively poor data from that sense. My dated recollection is that fundamental functions of the brain involve critical windows that close much earlier (eg. retinal calcium transients, ocular dominance).
Could such variation sometimes result in higher order regions overweighting a particular sensory pathway, with possible impacts to cognitive performance, and other times result in improved performance when pathways integrate appropriately? It seems potentially consistent with the age of onset, sensory features and fixed interests of ASD.
I developed hate for shiitake mushrooms in my late 20s due to a terrible meal I made on a hiking trip. Even the smell of shiitakes put me off the food.
Fast forward to when I was 41, I decided to try something: I like all other mushrooms, so what if I pretend my priors didn't exist? My wife cooked a meal with shiitakes in it. I sat down to eat. I blanked my mind, told myself that I had never eaten shiitakes before, and was tasting them for the first time.
They tasted great and now I can eat them again.
I guess my point is that people should try updating on things they think are impossible to update on. Maybe they get lucky.
Often when kids have that tomato explosion experience, parents will keep reminding them of it and kind of forcibly making them internalize it. 'Remember that time when...'.
The family has visitors over and you're introduced as the kid who had a tomato explosion and now hates tomatoes.
Family photos are being shown and there's you next to a tomato plant...
People mostly have no memories of times before they're like 5, so if everyone would just drop the topic right after the incident, these things would not get written to ROM so hard.
I was the kid who hated Nutella until this year (I'm 33) when I just went ahead and put some in my mouth. It's just not very good.
The innovation path unfortunately would need to involve a reduction in the obsession with like for like growth in profits as the key indicators of success.
Thanks John Money. Surely we can put this to use to cure trans people immediately.
>One of my distant cousins won't eat tomatoes. His parents say when he was very young, he bit into a cherry tomato and it exploded into goo in his mouth, and he was so upset he wouldn't eat tomatoes from then on. Now he’s in his 30s and still hates them. Is this fairly described as a “critical window” for food preferences?
Don't we all have a story like this? I ate sriracha as a kid and got food poisoning. I now have a strong negative reaction to the smell of sriracha.
>So at the beginning, you might start in a random place and want to see if 500 miles away is more mountainous. But once your hard work has brought you 1000 feet from Everest’s summit, you don’t want to take a 500 mile jump and end up in New Delhi and have to start all over again.
It seems pretty universal that as things scale upward in mass (and accumulated learning/knowledge can be viewed as mass), they become less "maneuverable". Speedboats vs ocean liners, etc.
Maybe this relates to your language post from yesterday, where a child with a vocabulary of 10,000 words can more easily learn a new language than an adult with a vocabulary of 20,000 words (and an increased number of deep rules that must be discarded, like "adjectives of quantity come before adjectives of color".)
wanted to note that you are incorrectly using the term 'furry' to refer to people who are attracted to cartoon animals; not all furries are.
> the AI starts at a very high learning rate (also called “temperature”)
For modern models we're talking about completely different things when we talk about temperature and learning rate. I think the confusion here is probably due to the idea of a learning schedule being similar to simulated annealing.
I think there may also be a different subsystems learn differently aspect here. I’ve personally had food poisoning from a particular fast food restaurant and then not eaten at that entire chain for years. Similarly my girlfriend won’t go anywhere near vicinity of any kind due to a terrible experience in high school with a concerning quantity of coconut rum. I think we have a system that jumps further on bad food experiences and is much much more willing to trap into priors in a negative direction for very understandable survival reasons. E.g. berries that look/smell like this caused me to vomit all of my food up once, therefore I will never eat them again in case they are poisonous.
I think the only way to get past food aversions is to try the food again (even a tiny nibble) a bunch of times, and if those experiences are good, you start liking the food and get over the aversion. My guess is that the tomato thing is less about the specifics of the original experience, or the fact that it was in childhood, and more about your cousin (presumably) deciding each and every day for the subsequent decades to never try cherry tomatoes again. By contrast, you've presumably had lots of positive beach experiences between the scary one and today. (Exposure therapy is obviously related.)
(A useful parenting tip that I got from the book "Bringing Up Bebe" is that it's inordinately important to try to get a kid to eat even just one bite of each kind of food you're serving. After 5 meals in 5 different weeks where they grudgingly eat a single bite of a certain food X, you might find that they have started liking X and now gobble it up.)
I always figured these were something like a manifestation of chaos, in the mathematical sense. When you’re growing everything that comes later depends on what came before. Signals or their absence matter more. There’s not a “go back and redo the last step” feature in most of biology. The best thing to that is our cognition and I think that only comes after it develops right.
How does this programming relate to life long diet, behavior, skills? And how much more effort is required to retrain? We invest on a thesis. We spend 1.7T in US on food, 1.9T on healthcare cost of poor nutrition. You use to go to grocery and were thin, now you get diabetes. We have been trained to like ultraprocessed food even though it's bad for us. We don't invest in behavior change. We do invest in things people want. They want to live longer but are stuck eating the things programmed in their brain in their youth. What is the innovation path of least resistance to better health and behavior change?
This also explains why pedophiles aroe ften victims of abuse themselves when they were children.
"animals that can’t see out of one eye for the first few months of life permanently lose the ability to process that eye"
As someone with ONH that's much worse in one eye than the other, I'd like more information about this finding, if anyone has it on hand.
As a parent of an extremely premature child, I'm familiar with a few specific physical critical development windows. If you do not practice the swallowing reflex when you are an infant, you will struggle perhaps for the rest of your life (my son was intubated for months). Similarly, the development of your alveoli in your lungs seems to happen primarily in the first 5-7 years of life or so.
These examples don't relate to the idea of critical windows for developing preferences, but I thought I'd offer them as additional concrete examples.
For much of my life I was unable to eat any food located between two slices of bread (by definition, a sandwich), which my mind categorized as revolting and vomit-inducing, even though I like bread and food that comes on a single slice of bread (isn't that illogical?)
This was due to a bad experience in my childhood, not with food in itself, but because it was the food I had to eat during a very stressful period.
I'm not even sure I can eat one today.
There must be something unique about early life bad food experiences. Nothing else sticks so vividly and almost permanently.
> tomatos
> sexual animals
I think food disgust is likely considerably going to be different from sexual disgust in how its managed.
Like avoiding poison every day vs avoiding sex that may be the one time you pass on your genes outside your tribe
Childhood vs adulthood doesn't seem to be rocket science: children simply have less background to compare against. It is like wise old men vs. young naive ones, only the comparison is even worse.
Unlike AI - real intelligences have an ever ongoing comparison and update of input vs. experience. A simple example: if apples fall down - 100 past experiences of apples falling down won't be overwritten by one example of an apple falling up. A child, however, doesn't have 100 past experiences - maybe only has a couple so the fall-up outlier makes much more of an impression.
As for the ongoing fetish with "bad imprint" - the assumption being made is still that furries are just miswiring firing off random chance.
I say assumption because I do not think furries were a thing before television. Note that this is even though children have had stuffed animals for a long, long time; even animated animals in movies only constitute a relative handful of hours per year prior to the advent of television.
It is just as likely that it is the hundreds/thousands/tens of thousands of hours which children spend watching fake talking cartoon animals (basically people but that look like animals) on television that causes the miswiring. In this case, it is this prolonged conditioning in an entirely artificial experience as opposed to some neuron firing too strongly once during a child's viewing of a cute animal.
And if this is so, the problem isn't evolution - the problem is garbage being input into children at high rate:. GIGO. It is actually a testament to evolution that so few children are permanently miswired after such a regime of indoctrination that would make Guantanamo look like a week of spring break. If kids are watching 2000 hours of TV + internet videos a year from age 4 to 14 - that's 20000 hours of potentially bad info going in - and more importantly, is likely a greater source of input than any single other input including parents. Add in 1500 hours of schooling per year/12000 hours from 4 to 14 with a LibsofTikTok type teacher - holy crap.
It should only be a wonder why there aren't MORE furries.
Now consider how either a machine learning or neural network type AI would perform with this diet of input info.
People seriously think evolved intelligence is so easily reproduced? LOL
The two examples you gave could be highly salient/heavily imprinted because they are in areas highly important to reproductive fitness, ie reproduction itself and poison avoidance. Evolutionary pressure over generations is the training data for the genetic matrices that encode our neural matrices.
I think it's clear that formative stimuli like the exploding tomato happen and are not so uncommon. I don't share your puzzlement about why their effects are not more uniform.
* The brain is a large and complex system, and any given stimulus will hit different people differently.
* Processing of the stimulus in the years afterward can also greatly differ, based on subsequent formative stimuli, one's openness to new experience, etc.
It seems to like you're looking for a compact, unified, deterministic explanation of why X produces Y, whereas I don't expect such an explanation to exist in a large, somewhat chaotic system.
Just wanted to say I appreciated the visual pun in the image for this post.
I suspect the tomato story is a slight misunderstanding.
My pet theory as to why so many people hate raw tomatoes (quite common in my experience) is something to do with them being vegetables (savoury tasting) masquerading as fruits.
You bite into broccoli or a cucumber and it tastes green and leafy, as you'd expect (so kids don't expect to like it or eat it raw)- you expect a cherry tomato to taste like a cherry or a grape (sweet), and instead it tastes like a raw tomato (savoury, awful). So your cousin wasn't imprinted with the unpleasant experience of a tomato exploding in his mouth, but with the fact that this unpleasant savoury-tasting juice came out, rather than sweet fruity goodness.
There are attempts to find genetic or chemical explanations for the raw-tomato-skeptics among us: https://blogs.scientificamerican.com/cocktail-party-physics/tasters-choice-why-i-hate-raw-tomatoes-and-you-dont/ , but my explanation feels like it better matches my experience.
Counterexample to the tomato thing: I also thought I didn't like tomatoes since childhood, but that was a belief, not a taste. When someone convinced me to try one at age ~17, and it was good. Now I like tomatoes. Maybe the same would happen with your cousin if they tried one. Maybe the same would happen with me again if I tried a Cheeto (I won't though lol)
I haven't read this yet, but I greatly appreciate the image on the front page.
We'll learn a lot about possible critical windows in sexuality if and when specific cohorts of children hit puberty and exhibit widespread fetishes for face masks.
As someone that has tried tomatoes as an adult, your cousin made the right choice. "He still won't try tomatoes" is equivalent to "he still won't stick his hand in fire".
Do we know that all these childhood formative experiences are really the cause of anything, or if they are just the first time when something latent becomes salient?
Not "x made me gay" but "I realised i was Born This Way when I saw x".
This reminds me of training dogs. On the one hand they are incredibly trainable (for certain tasks). But they can carry one bad experience through their entire life.
So dogs having this incredible 'learning window' is the same trait that leads them to having 'cherry tomato-style' experiences inprint on them so strongly?
I remember it as if it were yesterday. As a young child I was walking to the shops with my Mum and she gave me a tomato to enjoy. I bit into it but a pip stuck painfully between my two front teeth and couldn't be dislodged without returning home. For a long time I wouldn't eat raw tomato, but now 70 years later I can eat cooked tomato or sliced tomato in salads. I still won't bite into a whole tomato.
Perhaps it is a matter of two influences colliding. The immediate strong event (e.g. the pip getting stuck) and a pre-existing strong aversion/desire (others messing with my teeth).
I didn’t like your ”just so” stories - kinks and fetisches come with incredible variations, so either you have to make up increasingly unlikely explanations, or you have to say that some fetisches can be rationally explained and others can’t, which seems strange. Whereas if you just believe it’s an unusual cross-wiring explained by childhood experiences, anything can fit that model.
Presumably the difference to tomatoes is that you’re _supposed_ to form sexual preferences at a young age and you’re mentally wired for it, but the mind being the messy thing it is it can easily get non-standard. This seems to include homosexuality, and I wouldn’t be the least bit surprised if it’s also how various trans things are formed.
Quoting ’Cryptonomicon’:
”I finally came clean to her about the stockings. It felt good. I'd been reading a lot about how the brain develops and had finally come to accept my stocking kink. It seems that when you are a certain age, somewhere between about two and five years, your mind just gels. The part of it that's responsible for sex becomes set into a pattern that you'll carry with you for the rest of your life. All of the gay people I've ever discussed it with have told me that they knew they were gay, or at least different, years before they even began thinking about sex, and all of them agree that gayness cannot be converted into straightness, or vice versa, no matter how hard you might try.
The part of your brain that handles sex frequently gets cross-wired into other, seemingly irrelevant areas at this age. This is when people pick up an orientation towards sexual dominance or submission, or when a lot of guys pick up highly specific kinks--say, rubber, feathers, or shoes. Some of them are unfortunate enough to get turned on by little kids, and those guys are essentially doomed from that point onwards--there is nothing to do except castrate them or lock them up. No therapy will unkink the brain once it has kinked.
So, all things considered, being turned on by black stockings wasn't such a bad sexual card to have been dealt. I laid this all out to Virginia during the trip home. I was surprised by how calmly she accepted it. I was too big of a jerk to realize that she was thinking about how it all applied to her.”
I doubt the exploding cherry tomato is the reason your cousin dislikes tomatoes. A chocolate truffle exploding in his mouth filling it with delicious goo is unlikely to have put him off chocolate.
It's common to find people who hate the taste of raw tomatoes (I'm one of them). I think it's like Coriander/Cilantro, where there's some compound that some people can taste and others can't.
I pride myself on being able to eat anything so spent many years developing a tolerance to raw tomato flavour, and so can now eat it without gagging but ugh, it's still awful. Even if I do sometimes voluntarily add it to burgers etc to maintain my tolerance. Many people have insisted that oh, I just haven't tried a GOOD tomato, try THIS organic heirloom homegrown flavoursome sweet cherry tomato... Nope they're all bad, sorry.
Re: the "bad childhood experiences cause lifelong specific food aversions", I'd put a personal anecdote on the table that I've never been able to make much sense of.
When I was a kid, I was a fairly picky eater (I guess most kids are, but I seem to have been worse than average - or so the adults told me). And I still vividly remember being totally (and irrationally) grossed out by a couple of specific foodstuffs, to the extent that I would throw up in my mouth if they were just brought near me.
Fast forward a few decades, and some of these are amongst my favourite foodstuffs.
The interesting thing is how the switch happened. In childhood, these foods were super yucky. And at some point during my teenage years, zap!, the aversion was suddenly gone. No laborious process of trying to re-condition myself. Just... seeing a plate of that food, and going "yum! let's have some of this!". If the aversion had not been as strong during childhood, this would not have been as surprising: but I really found these things revolting. And then all of a sudden... tasty!
Also interesting is that this did not happen with all food that my childhood brain considered suspect. A lot of cheese varieties are still off the map, particularly the more biologically active ones. But these are likely a special category, as we probably have some evolution-provided innate additional safeguard in our brains not to eat stuff that visibly has mold growing on it. A safeguard that I haven't been able to deactivate yet.
The term “temperature” usually wouldn’t be used for learning rates. Temperature, especially in the context of LLMs, is a parameter for softmax that determines how sharp you want the probability distribution of predictions to look (roughly, e.g., put 99.99% on the class that’s slightly better than any other or put approximately the same probability on everything almost regardless of the activations), controlling randomness of the output.
Having the learning rate (lr) higher at the beginning of the training and then decreasing it seems to speed up the training compared to having a small learning rate all along, but also, there are all sorts of lr schedulers, including those that decrease and increase the lr in a cycle (no one knows why this sometimes works better, and intuitions about mountains and hill-climbing don’t work well for spaces with millions of dimensions; gradient descent works from the derivative , not from looking around, and the you might be climbing weirdly combined counterintuitive mountains- https://www.lesswrong.com/posts/muLN8GRBdB8NLLX36/visible-loss-landscape-basins-don-t-correspond-to-distinct-algorithms).
People might call decreasing the learning rate “cooling”, but I think most wouldn’t use “temperature”.
There's a rich literature on learning of food preferences. The top level idea that you need to know is that omnivores are evolutionarily sensitized to the effects of food, for the very good reason that a varied diet implies regular encounters with foods that might kill you.
If you play the nasty game of using radiation to make a rat feel nauseous after ingesting some food, it will most likely avoid that food forever. Omnivores can't afford to give foods many second chances. Which is part of the story about tomatoes.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8455326/
Since sexual experiences and tendencies don't usually kill you, its reasonable to think that evolution will be more tolerant of them. An awful lot of the psychoanalytic stories about childhood experiences with big effects are just stories. They might matter a lot when they become stories that we tell ourselves and come to believe. but that's another topic.
If at age n you experience that some experience E is really bad, the probability per Laplace's rule of succession that the next E will be really bad is 1/n per year. Because n is small, 1/n is large. So you never do it again, and therefore can't update.
I think that's a large-ish part of this story.
I would assume that the reason why formative experiences can have such a drastic effect, like being against eating tomatoes, has almost nothing to do with age specifically and a lot more to do with having little else to judge the experience by.
It's entirely plausible that your thirty-something cousin could bite into a cheery tomato and get squicked out for the first time, but he would have three decades of experience with various species of tomato at that point, which is what lessens the impact of any new information. I would assume that, without experience with e.g. dragonfruit, he could look at the inside of one at any age *for the first time* and justifiably decide it's a horror too gross to deal with. We simply have less experience generally when we're younger, which is why these moments tend to happen in childhood.
You're the psychiatrist, I'm the bumpkin. You tell me if that's not accurate. All I have to go on is logic.
That's same in predictions as Freud just with more sophisticated mechanics
Just came to tell everyone that there's nothing wrong with pregnant Sonic, at least according to this song.
https://youtu.be/nkXz4UrPSuU?si=HhJMb9EToNlmsFA7
He aspirated tomato juice, and you didn't aspirate seawater? Just guessing, of course.
The high learning rate early in life likely does contribute to stronger imprinting/conditioning effects during childhood. Children's brains are still forming connections rapidly, so an intense experience can wire in certain associations more permanently.
Overcoming trapped priors later in life is difficult but not impossible. With enough contrary evidence and conscious effort, even adults can slowly weaken learnt associations. So while imprinting is stronger in childhood, neural plasticity continues through life.
i imagine childhood brain as less full of ruts and knotted pathways that adults have and swimming in BDNF—maybe the plasticity and new pathways that are triggered are done so in a way that is more potent bc there is a relative lack of others to compete with
Does valproic acid reopen critical windows in humans? I remember some excitement about this possibility about a decade ago. Was that just hype generated by a few underpowered studies?
"reacted in some profound way, like becoming a furry."
More likely ... let's back up to Freud, [paraphrased] "there are two kinds of deviation, deviation in aim, and deviation in objective." Is becoming a furry a deviation in objective: 'I tried to become a non-human ideation of an animal.' or is becoming a furry a failure in aim: 'I can't visualize a reasonably attainable human to human sexual model, I see more sexualization in animation than in humans.' I would think for a 'basement dwelling anime addict' the latter is the most likely case.
After writing this, I realized I have followed Freud into the trap of binary thinking, I do apologize for this.
> This still doesn’t explain the unpredictable nature of preference-changing events. My cousin had a bad experience with a tomato in childhood, but was that really his worst childhood experience?
I imagine it's probably something like "I didn't like tomatoes as a kid, so I didn't try them when I was older, so I didn't try them when I was older..." up to the present. Basically, he never had any reason to even gather more evidence about tomatoes being good or bad, so he never did, and never updated.
After your near-drowning, did you experience trepidation the first few times you went back to the beach?
These observations are so susceptible to selective memory and the pitfalls of post-hoc explanation.
I don't understand the logic here. How is a child who is updating stronger going to be more likely to end up in a trapped prior? If they're more willing to take 100-mile steps, why would we expect them to get trapped at a local minimum/maximum?
I think the implications of this theory on eating would be that children should be very adventurous with food, only to settle later in life on a narrow range of food preferences. That's opposite of reality.
And while there are some cases like that of a weird tomato aversion, that's an exception we shouldn't use to define the rule. My son had this issue with hot dogs, where he ate one while experiencing GI discomfort and decided that hot dogs make him sick. After a few months, he decided kielbasa and corn dogs are fine. I suspect he'll work his way toward the full hot dog experience again, so it definitely seems like a trapped prior. But if I accepted the hypothesis that children take larger step changes than adults, I wouldn't expect his priors to get trapped in the first place.
One other feedback loop to consider is Identity: as a young child I said I didn’t like mayonnaise and people found that surprising, so I became A Person Who Doesn’t Like Mayonnaise. This stuck around until after college when I realized that my preference wasn’t actually grounded in anything and I could switch my mayo preference and no one would care or give me a hard time for it.
For somatic, “feel like I need to vomit” reflexes this might be less of a factor, but for things like “I am American and so speak English” or “I am not a math person” or “I am a Democrat” these identity-based loops from childhood can be impactful and very hard to get out of, and they don’t happen as often to adults.
Naively, wouldn't it make sense for childhood priors to be stronger, just because they arrive without a lot of other signal to balance them out?
Although I wonder if there's something weird about food preference that makes it particularly rigid in children compared to other things. Anecdotally, I was an incredibly, obsessively picky child, and at least compared to how I was then I'm not at all picky now, and I can pinpoint, almost exactly, the age where that shifted for me (around 24) but I could not tell you why at all.
There's a surprising number of people here who dislike tomatoes, including me. Is this worth polling on? Maybe a question on the next ACX survey?
Personally, I think it's the remaining nightshade-like taste on fresh tomatoes that puts me off, and I will happily eat tomatoes once they've been prepared in one way or another and they no longer taste fresh. Maybe this is akin to the cilantro gene and we just have a predisposion against that taste?
> In a typical AI training run, the AI starts at a very high learning rate (also called “temperature”) and gradually “cools” down over time.
Temperature and learning rate are unrelated* in a typical (modern) AI setup. Temperature is a parameter that determines how "spiky" the probability distribution you're using is, or equivalently, how concentrated it is: a heated posterior is very spread-out (high entropy) while a cold one is dominated by a single narrow peak (low entropy).
*Except in the special case of simulated annealing, which is rarely used in modern AI optimization.
> So at the beginning, you might start in a random place and want to see if 500 miles away is more mountainous. But once your hard work has brought you 1000 feet from Everest’s summit, you don’t want to take a 500 mile jump and end up in New Delhi and have to start all over again.
Now this is a *different* algorithm (sounds like adaptive hill climbing, maybe).
Did you ever have an experience with psychedelics? I'm not saying you should because it can be dangerous but at moderate doses that don't send you into alien words or ego deaths it is really interesting to observe your filters (I believe from the default mode network of the brain, which is suppressed by many psychedelics) first hand.
What I mean with this: There are actually quite a lot of things that go through an unconscious filter before your conscious mind gets to process them. These things are not only the often portrayed visual artifacts (intense colors, moving / breathing textures etc.) but also *thoughts*.
That means there are actually thoughts that your filter does not allow you to process at all. Many of these filters are also created during our childhood. Thoughts that scare you, that you are constantly told are wrong or impossible or thought that just did not make sense or help you in any way.
With those filters suppressed, you can have these thoughts again. This can lead to very scary experiences ("bad trip") but also lead you into states of childish joy, wonder and interest about things that you normally do not (actually *can not*) notice.
When I would start feeling nauseous on long car trips as a kid, my parents gave me orange-flavored Dramamine. Associating it with nausea is likely why I hate artificial orange flavoring to this day (although I like real oranges).
I wonder if variation in learning rate/critical period plasticity could explain the genetic correlation between ASD and intelligence.
If genetic variants speed up activity-dependent patterning of primary sensory pathways, it might result a longer period of consistent sensory training inputs for areas involved in higher order integration, during their respective critical window. At the same time, if primary sensory windows close earlier it risks these pathways getting "trapped priors", with downstream brain regions getting comparatively poor data from that sense. My dated recollection is that fundamental functions of the brain involve critical windows that close much earlier (eg. retinal calcium transients, ocular dominance).
Could such variation sometimes result in higher order regions overweighting a particular sensory pathway, with possible impacts to cognitive performance, and other times result in improved performance when pathways integrate appropriately? It seems potentially consistent with the age of onset, sensory features and fixed interests of ASD.
> Now he’s in his 30s and still hates them.
I developed hate for shiitake mushrooms in my late 20s due to a terrible meal I made on a hiking trip. Even the smell of shiitakes put me off the food.
Fast forward to when I was 41, I decided to try something: I like all other mushrooms, so what if I pretend my priors didn't exist? My wife cooked a meal with shiitakes in it. I sat down to eat. I blanked my mind, told myself that I had never eaten shiitakes before, and was tasting them for the first time.
They tasted great and now I can eat them again.
I guess my point is that people should try updating on things they think are impossible to update on. Maybe they get lucky.
Often when kids have that tomato explosion experience, parents will keep reminding them of it and kind of forcibly making them internalize it. 'Remember that time when...'.
The family has visitors over and you're introduced as the kid who had a tomato explosion and now hates tomatoes.
Family photos are being shown and there's you next to a tomato plant...
People mostly have no memories of times before they're like 5, so if everyone would just drop the topic right after the incident, these things would not get written to ROM so hard.
I was the kid who hated Nutella until this year (I'm 33) when I just went ahead and put some in my mouth. It's just not very good.
The innovation path unfortunately would need to involve a reduction in the obsession with like for like growth in profits as the key indicators of success.
I have no clue how this can be achieved though..
"Not too sweet" something someone under 12 years old has never said, ever.
I used to put sugar on my Frosted Flakes, that apparently would kill an adult (poster on on here) from Europe or Asia.