7 ms·

Trippy images were designed by AI to super-stimulate monkey neurons

68 points by amatus 7y ago

adolph 7y ago
I wonder how this technique could be used to develop specific-person optimized user interfaces. Of course, it’ll probably be first used in ad-tech.
- pavel_lishin 7y ago
  "I will buy whatever you're selling to make this stimulus stop, please."
  - andai 7y ago
    Resume viewing. Resume viewing.
monkeyHELL 7y ago
JFC. This thing is a monkey nightmare machine. https://3c1703fe8d.site.internapcdn.net/newman/gfx/news/hires/2019/2-thesetrippyi.jpg https://3c1703fe8d.site.internapcdn.net/newman/gfx/news/hire... For sure, those are very likely images of fear. If you wanna call them “highly stimulating” well... That’s probably not untrue. But the only thing that the monkey is likely finding “stimulating” about familiar people in surgical masks is an association with pain and terror. Note the odd expressionless faces standing over them, holding utensils. Pretty god damned ominous. Not at all images one might call fun.
- b_tterc_p 7y ago
  What are the paper faces?
  - max76 7y ago
    Facemasks?
- Darkphibre 7y ago
  You seem to be anthropomorphizing the neuron. This work is pretty innovative: We've used this technique to generate similar images of computer neural networks to try to figure out what a given neuron or layer is actually learning: http://yosinski.com/deepvis http://yosinski.com/deepvis Now, we're monitoring a biological neuron with the same technique, and coming up with similar feature selection. That both validates the approach we've been taking with deep neural networks, AND it provides us heretofore-unimaginable access to individual neuronal activation materials. Exciting stuff!
  - BoiledCabbage 7y ago
    > You seem to be anthropomorphizing the neuron. And respectfully, you seem to be ignoring the fact that the neuron is part of a living sentient creature. That comment epitomizes the cliche of an engineer/scientist so excited by the technology that they are completely blind to its real world implications.
  - comex 7y ago
    The images in this study were generated by a regular GAN, though. The biological neuron responses were only used to optimize the input parameters to the GAN. So I'm not sure if it really counts as independent feature selection.
anon1m0us 7y ago
This is sick. I can't believe as human beings we still do this to other conscious creatures with feelings. It's emotionally traumatizing. It's wrong.
- mruts 7y ago
  Maybe we should just test things on people instead?
  - anon1m0us 7y ago
    Yes, if the human chooses to be the subject of the test, then yes. Monkeys don't give consent.
    - mruts 7y ago
      Neither do cows, pigs, or chickens. Should I feel sorry for them as well?
      - anon1m0us 7y ago
        Yes.
- nartz 7y ago
  I'm not sure you are interpreting this correctly - its not like they were torturing the monkey. This is almost the same as studies where they show humans a set of images and ask them which they like the most. Except in this case, they just show them the images and try to measure it. Most of these comments are blowing this out of proportion.
  - crishoj 7y ago
    Subjecting a being to a machine which automatically evolves stimuli to maximise fear response is not torture?
    - 7y ago
max76 7y ago
Why was this study done on monkeys instead of apes (closer to humans) or mice (cheaper)?
pavel_lishin 7y ago
> To find out which sights specific neurons in monkeys "like" best, researchers designed an algorithm, called XDREAM, that generated images that made neurons fire more than any natural images the researchers tested. I know they put "like" in quotes, but damn. Imagine this from a scifi angle: > "The aliens from Tau Ceti found which stimuli specific humans "like" best by trying them out and seeing which ones made humans scream the most.
- mturmon 7y ago
  The methods and results are scary/foreboding. It's like they are automating and extending the methods used by, say, horror movie creators. Consider the horror-movie hockey mask, which also has the deep, dark eye sockets seen on these images.
- ASalazarMX 7y ago
  Fortunately the brain has astounding plasticity. It it's only images, we would get used to them.
  - pavel_lishin 7y ago
    Sticks and stones may break my bones, but words can penetrate to the core of my soul, leaving wounds that never heal. https://www.theverge.com/2019/2/25/18229714/cognizant-facebook-content-moderator-interviews-trauma-working-conditions-arizona https://www.theverge.com/2019/2/25/18229714/cognizant-facebo...
  - dwaltrip 7y ago
    Try telling that to those who do content review for YouTube, Facebook, etc. People have suffered PTSD from jobs like that. Saying it's just images isn't saying anything at all. Mental harm is just as real as bodily harm, even if it is harder to examine.
floatingatoll 7y ago
I was afraid of this when I saw the headline, because it means we are finally at risk developing a Parrot attack. If it’s not apparent yet why digital displays need anti-seizure filters in hardware before it’s too late for humankind, read this and be very afraid: http://www.infinityplus.co.uk/stories/blit.htm http://www.infinityplus.co.uk/stories/blit.htm
Eliezer 7y ago
Remember when it was becoming increasingly difficult to tell the difference between real headlines and Onion headlines? Well, now it's becoming harder to tell the difference between reality and SCP Foundation logs.
ve55 7y ago
This paper pretty closely resembles one that I saw a few months ago: Neural Population Control via Deep Image Synthesis: https://www.biorxiv.org/content/10.1101/461525v1 https://www.biorxiv.org/content/10.1101/461525v1, which is included among the first few references. They do quite a bit of building off of it, but some of the images in the full paper are definitely as uncanny to look at as the ones included in the article about it above.
nartz 7y ago
I wonder if there are more 'universal' images that would be able to have the same effect across populations? I could imagine this type of thing going in art galleries, evoking different tailored emotions.
- kadoban 7y ago
  That would be an interesting application. I assume first we'll see it used in advertising though. More money and looser morals there.
bluejay2387 7y ago
I find this both disturbing and fascinating from the perspective that this is essentially an attempt at developing a saliency map of a living organism... so instead of XAI it would be XI?
00117 7y ago
How do all the commenters here know that the images are eliciting a fear response? The referenced study measures firing rates of visual neurons. It does not discuss ultimate emotional reactions.

68 points by amatus 7y ago

I wonder how this technique could be used to develop specific-person optimized user interfaces. Of course, it’ll probably be first used in ad-tech.

pavel_lishin 7y ago
"I will buy whatever you're selling to make this stimulus stop, please."
- andai 7y ago
  Resume viewing. Resume viewing.

monkeyHELL 7y ago

JFC. This thing is a monkey nightmare machine. https://3c1703fe8d.site.internapcdn.net/newman/gfx/news/hires/2019/2-thesetrippyi.jpg https://3c1703fe8d.site.internapcdn.net/newman/gfx/news/hire... For sure, those are very likely images of fear. If you wanna call them “highly stimulating” well... That’s probably not untrue. But the only thing that the monkey is likely finding “stimulating” about familiar people in surgical masks is an association with pain and terror. Note the odd expressionless faces standing over them, holding utensils. Pretty god damned ominous. Not at all images one might call fun.

b_tterc_p 7y ago
What are the paper faces?
- max76 7y ago
  Facemasks?
Darkphibre 7y ago
You seem to be anthropomorphizing the neuron. This work is pretty innovative: We've used this technique to generate similar images of computer neural networks to try to figure out what a given neuron or layer is actually learning: http://yosinski.com/deepvis http://yosinski.com/deepvis Now, we're monitoring a biological neuron with the same technique, and coming up with similar feature selection. That both validates the approach we've been taking with deep neural networks, AND it provides us heretofore-unimaginable access to individual neuronal activation materials. Exciting stuff!
- BoiledCabbage 7y ago
  > You seem to be anthropomorphizing the neuron. And respectfully, you seem to be ignoring the fact that the neuron is part of a living sentient creature. That comment epitomizes the cliche of an engineer/scientist so excited by the technology that they are completely blind to its real world implications.
- comex 7y ago
  The images in this study were generated by a regular GAN, though. The biological neuron responses were only used to optimize the input parameters to the GAN. So I'm not sure if it really counts as independent feature selection.

anon1m0us 7y ago

This is sick. I can't believe as human beings we still do this to other conscious creatures with feelings. It's emotionally traumatizing. It's wrong.

mruts 7y ago
Maybe we should just test things on people instead?
- anon1m0us 7y ago
  Yes, if the human chooses to be the subject of the test, then yes. Monkeys don't give consent.
  - mruts 7y ago
    Neither do cows, pigs, or chickens. Should I feel sorry for them as well?
    - anon1m0us 7y ago
      Yes.
nartz 7y ago
I'm not sure you are interpreting this correctly - its not like they were torturing the monkey. This is almost the same as studies where they show humans a set of images and ask them which they like the most. Except in this case, they just show them the images and try to measure it. Most of these comments are blowing this out of proportion.
- crishoj 7y ago
  Subjecting a being to a machine which automatically evolves stimuli to maximise fear response is not torture?
  - 7y ago

max76 7y ago

Why was this study done on monkeys instead of apes (closer to humans) or mice (cheaper)?

pavel_lishin 7y ago

> To find out which sights specific neurons in monkeys "like" best, researchers designed an algorithm, called XDREAM, that generated images that made neurons fire more than any natural images the researchers tested. I know they put "like" in quotes, but damn. Imagine this from a scifi angle: > "The aliens from Tau Ceti found which stimuli specific humans "like" best by trying them out and seeing which ones made humans scream the most.

mturmon 7y ago
The methods and results are scary/foreboding. It's like they are automating and extending the methods used by, say, horror movie creators. Consider the horror-movie hockey mask, which also has the deep, dark eye sockets seen on these images.
ASalazarMX 7y ago
Fortunately the brain has astounding plasticity. It it's only images, we would get used to them.
- pavel_lishin 7y ago
  Sticks and stones may break my bones, but words can penetrate to the core of my soul, leaving wounds that never heal. https://www.theverge.com/2019/2/25/18229714/cognizant-facebook-content-moderator-interviews-trauma-working-conditions-arizona https://www.theverge.com/2019/2/25/18229714/cognizant-facebo...
- dwaltrip 7y ago
  Try telling that to those who do content review for YouTube, Facebook, etc. People have suffered PTSD from jobs like that. Saying it's just images isn't saying anything at all. Mental harm is just as real as bodily harm, even if it is harder to examine.

floatingatoll 7y ago

I was afraid of this when I saw the headline, because it means we are finally at risk developing a Parrot attack. If it’s not apparent yet why digital displays need anti-seizure filters in hardware before it’s too late for humankind, read this and be very afraid: http://www.infinityplus.co.uk/stories/blit.htm http://www.infinityplus.co.uk/stories/blit.htm

Eliezer 7y ago

Remember when it was becoming increasingly difficult to tell the difference between real headlines and Onion headlines? Well, now it's becoming harder to tell the difference between reality and SCP Foundation logs.

ve55 7y ago

This paper pretty closely resembles one that I saw a few months ago: Neural Population Control via Deep Image Synthesis: https://www.biorxiv.org/content/10.1101/461525v1 https://www.biorxiv.org/content/10.1101/461525v1, which is included among the first few references. They do quite a bit of building off of it, but some of the images in the full paper are definitely as uncanny to look at as the ones included in the article about it above.

nartz 7y ago

I wonder if there are more 'universal' images that would be able to have the same effect across populations? I could imagine this type of thing going in art galleries, evoking different tailored emotions.

kadoban 7y ago
That would be an interesting application. I assume first we'll see it used in advertising though. More money and looser morals there.

bluejay2387 7y ago

I find this both disturbing and fascinating from the perspective that this is essentially an attempt at developing a saliency map of a living organism... so instead of XAI it would be XI?

00117 7y ago

How do all the commenters here know that the images are eliciting a fear response? The referenced study measures firing rates of visual neurons. It does not discuss ultimate emotional reactions.