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.
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
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.
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.
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.
> 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.
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.
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.
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:
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.
It's not maximizing "fear response", I'm not sure where everyone in this thread is getting that from. It's maximizing the response of particular visual cortex neurons, in structures where they're shown to be recognizers of specific shapes or concepts.
I.e. it's evolving images that this neuron thinks look most like a monkey, or a person. It's not the only neuron making that judgment, and it's all super nonlinear, so ofc it looks strange and distorted.
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.
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, 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.
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?
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.