[jaster]

I find the article and your comment strongly resonate with a comment I made some time ago on an article about GAN-generated faces:

> I guess very soon we will be able to generate "super-attractive" (as in "superstimuli") faces for virtual personas, according to targeted demographics and purpose (advertisement, youtube videos for kids, political messages and so on).

(original comment here https://news.ycombinator.com/item?id=18310355)

Here it's not so much about the face than about the behavior and interactions, but I think the same idea hold true.

It seems to me in the near future we may well be faced with a constant exercise in self-control in the face of the multiplication of such projects.

[whymauri]

We can already do this with non-human primates.

https://science.sciencemag.org/content/364/6439/eaav9436

I remember joking that this paper puts us marginally closer to the Men in Black memory wiping device, but in the context of this thread that joke seems a bit dark.

[jaster]

Thanks for the article!

I think it's even worse actually, because this is not something that needs to be done explicitly or on purpose.

Simply training AIs with a target goal of maximizing engagement could lead to models discovering and exploiting superstimuli or superstimuli-like bias in humans.

[etiam]

With the reservation that I only skimmed the article, it seems like what they produced was visual stimuli resulting in patterns of neural activation/non-activation at a rather limited number of sample points in a higher order visual area (macaque V4).

Not to take away from their results, such as they are, but it is very expected that visual inputs should have specific and fairly predictable effects on V4, and one could probably have designed such patterns manually from known perceptual psychology in a few iterations, with feedback from the recording electrodes for the details.

It is not at all obvious to me, and indeed not even plausible, that they'd be able to control arbitrarily chosen neurons this way.

[pmoriarty]

That link is broken for me.

You wouldn't have an article title that I could search for, would you?

[Arnavion]

"Neural population control via deep image synthesis" - Pouya Bashivan, Kohitij Kar, James J. DiCarlo

[bitwize]

We don't need computers to do that. Look at anime. The large eyes and cute faces are definitely optimized to evoke an emotional response in the viewer. And it appears to have worked all too well, if the existence of waifu culture is any indicator.

GANs may find as-yet-uncovered maxima in the same problem space, but superstimulus-level attractiveness has already been achieved.

[CodeGlitch]

Yes my emotional response is to vomit into my mouth :) As a straight male, anime "cute girls" is just grotesque to me. Different strokes for different folks,I guess.

[Cactus2018]

We don't even have to go with GAN generated faces:

https://thechive.com/2015/09/13/heres-what-the-average-perso...

[AntiImperialist]

For just faces, there's this website, if anyone is unaware: https://thispersondoesnotexist.com/

[candiddevmike]

Is it just me, or does anyone get deeply disturbed with some of the generated ML images? Parts of them look so bizarre due to the image composition, my brain simply cannot comprehend that part of the image. It's a very awful feeling, like my brain acknowledges that what I'm seeing is terribly unnatural. I don't experience this when looking at artwork, or some random scribbling, but with ML images it seems like they're generated in such a way that it defies any kind of recognition.

[robocat]

Presumably threat actors are working on generic algorithms to optimise the production of images by the ML system such that they have the absolute worst affect on viewers of those images.

Optimising for minor effects like blindness to areas, up to major effects like epileptic seizures.

[gwillen]

I often describe the sensation of trying to read ML-generated text as "feeling like I'm having a stroke". Phrasing not original to me obviously, but I like it. Although I've never had a stroke myself, so I can't say how accurate that is.

[TheMightyLlama]

I wonder what the opposite of that is. Is there text that’s so well written that it’s like looking at silk flowin through the air?