Original post: What Can Fetish Research Tell Us About AI?

Table Of Contents:

1: Alternative Theories Of Fetishes
2: Comments Including Testable Predictions
3: Comments That Were Very Angry About My Introductory Paragraph
4: Commenters Describing Their Own Fetishes
5: Other Comments

1: Alternative Theories Of Fetishes

Erusian writes:

If you take the full evopsych route (which this article implicitly does) then fetishes are best explained not as misfires of the procreative impulse but as part of the wider definition of sexual activity and display. The idea that sex is literally just PIV intercourse is not true of any complex and social species. In all such species you see social roles and rituals around sex. And these are adaptive in that the competitions increase prosociality and role fitness.

I find it hard to justify the misfire hypothesis actually since so much of sexual and pre-sexual activity is obviously not literally penetrative sex and so much of what's 'normally' attractive is not related to that. Lingerie, for example, seems completely unjustifiable in such a framework except as a niche fetish. But it's actually pretty universal. I understand lingerie as a sociosexual signal and that explains it pretty neatly. But if we're being trained on seeking PIV intercourse solely or its directly associated traits then you have to walk a pretty long way to explain such 'universal' fetishes that are common even among virgins.

I don’t see the connection from “the wider definition of sexual activity and display” to “some people literally can’t have an erotic experience unless their partner is dressed head to toe in black leather”.

I agree it makes total sense that some things that are closely related to sex (eg lingerie) can get sexual valence in and of themselves through something like classical conditioning. But that doesn’t explain why some things not that closely related to sex (at least for most people) can get sexual valence even greater than the actual sex act.

Giles English (extremely relevant blog) writes:

> “Here are some zero-evidence just-so-story speculations for how various fetishes might form”

I think that the BDSM fetishes listed are most parsimoniously explained as *super stimulation* affecting those wired for D/s mating strategies. (Just to be clear, I'm talking primal monkey stuff here, not modernity, and "natural" isn't the same as moral.)

On the dominant side of the slash, being the dominant mate seems adaptive. The male dominant benefits from effective mate guarding, the dominant female is assured access to resources etc. So, I think some people are wired to find being dominant sexy, resulting in a whole load of obvious kinks that emphasise or simulate this.

On the submissive side of the slash, picking a high dominance mate also seems adaptive. An - using the term as shorthand - "alpha" male or female both have improved access to resources, provide improved protection to offspring, and produce children with a better chance of reproducing etc.

So then you have kinks stemming from solutions to the problem of courting an alpha when you are not an alpha: things that emphasise submissiveness, and maybe sometimes things that ping the "sneaky fucker" strategy such as cuckolding and sissification. And you have kinks that make a partner *seem* more alpha. And both of these are super stimulation.

Neike Taika-Tessaro writes:

As a submissive, data point: Shame doesn't play into it at all for me. If you ask my friends who's most comfortable talking about sex, it's me. I bring it up as a casual topic like other people bring up video games they've been playing (but, yes, I do try to make sure the people I'm talking to don't mind, sex is a squick topic for a lot of people).

For me, it's the 'thrill' of being helpless and at someone's mercy.

The best way to think of this may be of being on a roller-coaster. You can't get off, you're at the whim of the ups and downs and all those curves throwing your body around, but while it's hitting some buttons in your brain that are making your body respond with stress (raising your heartrate, etc) is, on some level, really exhilerating!

Being restrained (be it just by bodily force, or by ropes, etc) is a lot like that. There's a deep sense of risk, it requires a lot of trust (and forgiveness, and communication, which new submissives sometimes don't realise), but it's very rewarding.

Steve Byrnes writes:

Here's a really simple theory:

PHYSIOLOGICAL AROUSAL—REGARDLESS OF THE SOURCE—CAN CONTRIBUTE TO SEXUAL AROUSAL

That seems true to me, and seems to explain (or at least contribute to the explanation of) four of your list items—spanking, sadomasochism, urine/scat, and bondage/domination/submission.

Proves too much. What are the most physiologically-arousing events young children experience? It must include things like stubbing their toe, touching a hot stove, or going really fast on a bike. But approximately nobody has toe-stubbing, stove-touching, or biking fetishes.

On the other hand, things like latex/rubber, cartoon animals, and uniforms, which don’t cause physiological arousal, do cause fetishes.

I agree that physiological arousal is part of the puzzle, but I continue to think the best explanation is that fetishes are produced by things that are like sex along some axis. Physiologically-arousing is one way a thing can be like sex, but it’s most likely to produce fetishes if it’s physiologically-arousing in the most sex-like ways. For example, spanking is physically arousing and involves another person applying rhythmic pressure close to your genitals; stubbing your toe is physically arousing without that addition. Spanking produces fetishes and toe-stubbing doesn’t.

There was also a long debate over what qualifies as a fetish, which you can see here. Many people pointed out that by traditional definitions, things like oral sex should qualify.

My impression is that oral sex was viewed as a bizarre perverted act, similar to other fetishes, until the mid-20th century, when it caught on. I think this is part of a general pattern where anything that’s common enough becomes universal (or at least there are compounding gains from commonness). I think this is the same process that homosexuality and transgender are going through now; as it becomes well-known and not-weird, lots of people who would never have been into it a century ago find that they have whatever mental raw materials predispose them to it. A hundred years ago, it might have felt obvious that oral sex was a fetish; a hundred years from now, it might feel obvious that BDSM isn’t.

This is speculative and I’m not a sex historian, so take all of it with a grain of salt.

2: Comments Including Testable Predictions

Gwern asks:

The [claim that spanking fetishes might be associated with childhood spanking] seems to have one clearly testable prediction: as spanking rates in Western parenting have drastically plummeted over the past century, AFAIK, there should be a ~20-30 year lag where spanking fetishes also plummet. Is there?

Aella, who conducts fetish surveys, says she has the answer. Go to her comment for the full statistics, but people who report being spanked as children say they find spanking more erotic.

Is this causal? Maybe. Alternate explanations: people who found spanking erotic committed more mischief to get spanked more often; parents who found spanking erotic were more likely to spank their kids and fetishes are partly genetic; parents are more likely to spank oldest kids, and oldest kids have more fetishes. None of these seem as convincing to me as the simple causal story.

Aella finds that spanking fetishists are no older (on average) than non-fetishists. Since spanking rates were higher in the past, we might consider this evidence against the theory (using “time” as a non-confounded proxy variable for spanking exposure). But having fetishes was also less common in the past. I hope Aella is able to analyze some of this in more depth!

A friend on Discord quotes an NYMag article:

Enema fetishists, for whom the ultimate erotic act is to be splayed across someone’s lap with a rubber hose in their rectum, are rarer than they used to be, says Lehne, but those that do exist tend to be older Jewish men of Eastern European descent whose mothers used enemas to force the issue when their little ones didn’t poo on cue.

I found this enlightening when I read it, but forgot it was on Discord; trying to find it for this Highlights post, I searched first the ACX comments, then the subreddit, then Twitter, then places further afield. End result is that now “enema” is in my search history for every social media site. I hope this doesn’t affect what ads I get.

I predict (maybe postdict?) that there will be some effect from people who experienced the trigger as children, plus some other effect from people who just think about it or see it on porn.

3: Comments That Were Very Angry About My Introductory Paragraph

The first paragraph of my post was:

Arguing about gender is like taking OxyContin. There can be good reasons to do it. But most people don’t do it for the good reasons. And even if you start doing it for good reasons, you might get addicted and ruin your life. Walk through San Francisco if you want to see people who ruined their lives with opioids; browse Substack to get a visceral appreciation of the dangers of arguing about gender.

I meant this mostly as a joke. But some people got really angry about it. And it seems unfair to deflect their anger by pointing out it was a joke, when I feel the joke has a core of truth. So I’ll commit to the bit and try to defend myself here.

The paragraph starts by admitting that some people do it for good reasons. If I were someone who believed I was a brave whistleblower speaking out against predatory marketing, I would have just assumed I was one of the people with good reasons. “The wicked flee when no man pursueth”.

…but fine, these people seem to be taking this pretty seriously, and deserve a longer and more serious response.

First and least relevantly, I disagree with them on the object-level question. I assume their concerns are about puberty blockers - drugs which are given to transgender minors to prevent them from going through their birth-sex puberty (ie natal men getting deeper voices and chest hair, natal women menstruating, etc). I’m not a child/adolescent psychiatrist and I don’t prescribe hormones, so I’m not an expert in this topic and this should be considered my amateur opinion only (although my impression is that the APA, AMA, and various other guideline-setting organizations agree with me). But I think these are overall good, for a few reasons:

• The effects of birth-sex puberty are irreversible and will make it much harder to transition in the future. The effects of puberty-blockers are mostly reversible, and preserve the option to either transition or return to birth-sex in the future. Like all drugs there are potential side effects, some of which are irreversible, but in the case of puberty blockers these seem mild and comparable to other psychiatric interventions. I think the precautionary principle supports having confused children who don’t know what they want do the reversible rather than the irreversible thing.

• The biggest studies suggest that about 98% of children who take puberty blockers do later go on to transition (nothing in real life is 98%, so I assume something is wrong with this study, but things do seem to lean towards a vast majority continuing). An optimistic interpretation is that the screening process is very good and they’re only given to people who really want them; a pessimistic interpretation is that they push children further onto the transgender path. I don’t think whatever “pushing” doctors can do is enough to produce these kinds of numbers - compare the success rate of doctors/parents trying to push kids away from transgender! - so I lean towards the optimistic interpretation. That makes it even clearer that we should do the reversible thing (which helps 98% of people and reversibly harms 2% of people) and not the irreversible thing (which helps 2% of people and irreversibly harms 98% of people).

• As a pseudo-libertarian, in difficult decisions I prefer the option which preserves individual choice. It becomes more complicated when there are children involved. But it becomes less complicated again when the child spontaneously requests something, their parents agree, their doctors agree, and all medical guideline-making organizations agree. So now the calculus becomes “deny people the right to make decisions about their own bodies in a way which irreversibly harms 98% of them and helps 2%” vs. “allow people to make decisions about their own bodies in a way which helps 98% of them and reversibly harms 2%”.

• I do think it’s suspicious and bad that everyone is suddenly becoming transgender, and I support efforts to figure out why and stop it at the root, in some way which will prevent so many kids from wanting to be transgender. But it seems cruel to fail to figure that out, let lots of kids become horribly depressed about their gender, and deny them access to treatment. By all means, figure out that smoking causes cancer, but if you haven’t figured that out and cancer rates are still high, don’t restrict access to chemo. I realize these goals are sort of in competition, in the sense that allowing people to transition raises the visibility of transgender which might contribute to transgender being more common, but faced with the extremely-well-established finding that denying care directly hurts transgender people, vs. the very conjectural hope that maybe it would, in the distant future, decrease the rate of gender dysphoria, I once again go with clear evidence over vague conjecture, and letting people control their own bodies vs. not doing that.

If she means some childhood intervention stronger than this, I probably oppose it, although I’d have to look at each thing individually to be sure.

But none of this is especially relevant to the current debate, since my paragraph deliberately didn’t single out either side as worse than the other. It just said lots of people seemed too addicted to arguing about this.

Let’s say the skeptics are completely right. About 1500 kids get puberty blockers each year in the US, but probably some cases are unrecorded, and probably the numbers will increase over time, so let’s say 5,000 kids. We’ll assume it’s inappropriate for half of these kids, and they end up sterile and mentally ill without having been helped in any way.

This is going to sound offensive, but as far as “bad US medical policies” go, 2,500 children having their lives low-key ruined is nothing. I can think of a dozen US medical policies that are much worse than that! I wrote here about how bad IRB policies probably kill about 50,000 people per year! The failure to allow human challenge trials for COVID vaccines probably killed about 10,000 people; the decision to delay the vaccine an extra few weeks to influence the 2020 election probably killed about 1,000. Inappropriate prescribing of antipsychotics causes 1800 deaths in the UK each year (the US number is probably closer to 10,000). Laws about organ donation incentives are “responsible for millions of needless deaths”. Even if you only care about children, there was the whole FDA fish oil story. Even if you only care about sterilization, Paul Ehrlich is still around! I’ve tried so hard to raise awareness of some of these issues, and although I’m deeply grateful for the five people who take them seriously, it’s a massively uphill battle.

But as soon as anyone brings up gender, the awareness raises itself. Millions of people who have never thought about IRBs spend a substantial portion of their lives having strong opinions on gender. Several presidential candidates are centering their entire campaigns around gender. Richard Hanania, whose many flaws have never included a lack of self-awareness, freely admits that I Hate Pronouns More Than Genocide. The rest of the world may hold that philosophy only implicitly, but they hold it nevertheless.

So yes, I think this is because arguing about gender is addictive. I say this as someone whose many flaws also do not include lack of self-awareness, and who’s spent years fighting the addiction and mostly winning.

I have also had this particular pleasure, and of course I sympathize with this person, but I also think her statement is literally correct as written. Again, not an expert, not a trauma-focused therapist, etc, but my amateur opinion is you gotta stop re-enacting your trauma. Some transgender activist cyberbullies you - many such cases! - and then you spend the rest of your life trying to own trans activists to prove that they were wrong and you were right and the world is safe again. IOU a post fleshing out this theory in more details sometime in the next few months. But for now, search your feelings, you know it to be true.

I don’t think I’m desperate. I think I’ve seen a lot of people go crazy over this and am trying to warn those who aren’t too far gone. I freely admit that sometimes you should go crazy about confronting injustice - John Brown ended up dead but his abolitionism was worthwhile. But if you’re going to sell your soul, ask how much you’re getting for it.

I don’t just mean “you could save orders of magnitude more people working on IRBs without offending anyone”. I also mean: if you’re going to change the world by focusing on trans issues, change the world by focusing on trans issues. Richard Hanania, again, has many flaws - but the guy clearly has a 28-step plan to end wokeness forever and is on, like, step 16 or something by now. Agree or disagree, you’ve got to respect the grind. Everyone else mostly seems to be making angry tweets and taking reactive potshots at the other side.

According to Graham’s Wikipedia page:

In interviews in 2022 and 2023, Linehan said the debate over transgender issues had "consumed his life": it had lost him work, made him financially destitute, and ended his marriage

This is not completely unlike the life outcomes of my opioid addiction patients.

I never asked any of these people to care about my half-joking introductory paragraph to an essay on AI and fetishes. But as The Last Psychiatrist says, “If you’re reading it, it’s for you.”

4: Commenters Describing Their Own Fetishes

Tiffany writes:

For what it's worth, on the specific physical action of bondage (mostly thinking of things like people being tied up here), when I'm engaging with it, I can feel my brain going down the same mental paths as grasping, like, with my arms and hands. That is, viscerally speaking, I would want to tie up a partner for essentially the same reasons I would want to hold them, and vice versa; the bonds are just a tool with which to accomplish something similar to holding. The analogy to the other fetishes listed here seems fairly obvious.

Anand writes:

I accidentally developed a bit of a foot fetish earlier this year, because of AI.

Midjourney 5 had not come out, and so we were using 4 to generate some characters for our app.

We needed some full-body images in transparent PNGs. The faces were very realistic, and even the hands generally had a good number of fingers, but it was really hard to get a full body shot without the feet or legs being cropped. Even though the prompt had "full body" and "head to toe" and a few other adjectives tossed on at the end, 9 times out of 10 it would end up cropping around the knees or just have the upper torso.

But the rest was really good — consistent characters and outfits, just missing the bottom. After a few hours of experimentation and re-rolling we could generally end up with the ideal full-body image, including feet. I developed a whole toolbox of tricks, from changing the aspect ratio to be really tall to blending in an example pose as a second image. One of the tricks to get this was to get really specific about the feet. What kind of shoes and socks? Mentioning feet a few times with emphasis. Things like that. My language was one that I imagine a foot-fetishist’s google search history. "Tall woman with feet showing, high heels, extra feet please with an order of foot." Or extreme detail like "orange nikes with shoelaces and socks and shadows under the shoes" That would generally convince the AI that the feet should also be shown. Not all the time, but some combination helped. I dreamt of feet, and kept trying to come up with more synonyms to get my desires across to this stable diffusion model.

Eventually after enough fiddling around a render would eventually come in with feet and we would be happy. After a few weeks of this process, anytime I finally saw some feet in a render (or shoes, rather) was a delightful feeling.

Then Midjourney 5 came out and started behaving a lot better. Feet were pretty trivial to generate (and hands reliably had 5 fingers) and the excitement of seeing them quickly wore off.

5: Other Comments

Jeffrey Soreff writes:

> “But concepts like “man” and “woman” are learned during childhood as patterns of neural connections"

Hmm - this casts doubt on whether the practice of systematically preventing children from seeing nude men and women is ... optimal. Oh well, not my problem...

Seems plausible. The more you keep kids in the dark about what normal sex is, the more they have to speculate, get things weirdly wrong, and then end up crystallizing those wrong guesses as fetishes.

My other crazy theory along these lines is that the modern emphasis on hiding gender - both obvious manifestations like parents who refuse to gender their children, less obvious manifestations like the parents who think it’s old-fashioned to get pink things for their daughter, and universal things like women mostly not wearing dresses - prevents kids from getting enough gender cues to develop a model of gender and increases the chance that they become trans (because it’s less obvious to their System I what gender they are). I think this is pretty unlikely, but still plausible enough to deserve some study. I don’t know how you’d study it though; most kids who are less exposed to clear gender cues are in environments that are more liberal in other ways too.

David Roman (blog) writes:

I find it remarkable that, even though he’s somebody who has written somewhat disparagingly of Slavoj Zizek in the past, the study of fetishes has led Scott to arrive at an understanding of sexuality not unlike Zizek’s. The way Zizek sees it, sexuality has a universal surplus, a capacity to overflow the entire field of human experience so that everything, from eating to excretion, from beating up our fellow man (or getting beaten up by him) to the exercise of power, can acquire a sexual connotation. And this is not a sign of its preponderance, but one of a certain structural faultiness: sexuality strives outward and overflows the adjoining domains precisely because it cannot find satisfaction in itself, because it never attains its goal or never-ending reproduction and because – as Alexander argues – sexuality is continuously thwarted by evolution – condoms, porn, etc – so, in Deleuzian terms, “perversion enters the stage as an inherent reversal of this ‘normal’ relationship between the asexual, literal sense and the sexual co-sense.” In perversion, even light perversion such as the one expressed by foot fetishes, sexuality becomes one desexualized object among others. To put it in an even more Zizekian fashion, I have to quote Zizek himself (in “The Plague of Fantasies,” 2009): “This link between sexualization and failure is of the same nature as the link between matter and space curvature in Einstein: matter is not a positive substance whose density curves space, it is tied to the curvature of space. By analogy, one should also 'desubstantialize' sexuality: sexuality is not a kind of traumatic substantial Thing, which the subject cannot attain directly; it is nothing but the formal structure of failure which, in principle, can 'contaminate' any activity. So, again, when we are engaged in an activity which fails to attain its goal directly, and gets caught in a repetitive vicious cycle, this activity is automatically sexualized - a rather vulgar everyday example: if, instead of simply shaking my friend's hand, I were to squeeze his palm repeatedly for no apparent reason, this repetitive gesture would undoubtedly be experienced by him or her as sexualized in an obscene way.”

Peter Gerdes (blog) writes:

I think a really important implication of this is that, contra a fundamental plank in AI alignment risk arguments, it's not the case that we should expect greater intelligence to mean greater coherence.

I mean, in some sense any set of actions can be seen as optimizing a preference function but what we mean by coherence is that it looks like maximizing a simple function (maximize paperclips not appreciate the weird beauty of paperclips that are uncommon as a paperclip hipster). Fetishes don't look like paperclip maximization.

Steve Byrnes writes:

> “Evolution controls the genetic code but not the reinforcement environment. Humans have the option of training AIs directly, a much higher bandwidth and less lossy communication channel.”

I don't think the comparison is clear-cut.

Yes, in reinforcement learning, humans have the option of rewarding the AIs for doing certain behaviors, which seems nice, and is genuinely something that evolution cannot do, except when those behaviors can be operationalized into relatively simple detector code that evolution can put in the brainstem etc.

On the other hand, "rewarding the AIs for doing certain behaviors" is a terrible idea, from an AI alignment perspective. We want to reward the AI for doing the right thing FOR THE RIGHT REASON, not just for doing the right thing full stop. Otherwise we reward the AI for being deceptive.

And in that sense, it seems to me that the genome is in some respects beyond the state-of-the-art of ML reinforcement learning alignment approaches. In particular, the genome sets us up such that some THOUGHTS are rewarding and others are not—not behaviors. There's interpretability right in the foundation, I think.

My own main technical AI alignment research interest is to figure out the nuts and bolts of how the genome makes people (sometimes) nice to each other — https://www.alignmentforum.org/posts/qusBXzCpxijTudvBB/my-agi-safety-research-2022-review-23-plans#2__Second_half_of_2022__1_3___My_main_research_project .

I think good theories of sexual attraction may be somewhat related, and I would definitely be interested in them, although I haven't spent any time looking into that topic so far.