• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Scientists reconstruct visual data from brain scans using AI

EviLore

EviLore

Expansive Ellipses
Staff Member
Full paper: https://www.biorxiv.org/content/10.1101/2022.11.18.517004v2.full.pdf

kOAYa1Y.jpg


The top row of images was presented to the test subjects. Stable Diffusion constructed the bottom images from MRI data of the subjects’ brain activity.

The future will be wild.
 
RiccochetJ

RiccochetJ

Member
That's really cool but also terrifying at the same time.

Reminds me of a 1992 movie called Fortress with Christopher Lambert and Kurtwood Smith where in this prison people will get punished for dreaming but also the warden can see what you're dreaming. I remember really liking the movie but have no clue if it holds up these days:

 
N

NickFire

Member
I'm a little stunned by the concept of this. So much possible utility. So much potential harm if used wrong.

The possible use in criminal law alone would be insane. Could the tech be deemed reliable enough for use at all? Would it's use violate constitutional rights? If deemed sufficiently reliable, would it warrant a constitutional amendment to ensure it can be used? Is using it giving too much power to algorithmic manipulation allowing innocent people to be framed? But what about the wrongfully accused who could be exonerated by it? Would their rights be violated by pretending the technology doesn't exist? These debates would be off the charts!

And my oh my oh my oh my. Imagine a consumer level device with this tech? Spouse / partner keeps coming home late, deleted texts, hidden money. The potential use in personal relationships would be unlimited.
 
M

Moneal

Member
NickFire said:
I'm a little stunned by the concept of this. So much possible utility. So much potential harm if used wrong.

The possible use in criminal law alone would be insane. Could the tech be deemed reliable enough for use at all? Would it's use violate constitutional rights? If deemed sufficiently reliable, would it warrant a constitutional amendment to ensure it can be used? Is using it giving too much power to algorithmic manipulation allowing innocent people to be framed? But what about the wrongfully accused who could be exonerated by it? Would their rights be violated by pretending the technology doesn't exist? These debates would be off the charts!

And my oh my oh my oh my. Imagine a consumer level device with this tech? Spouse / partner keeps coming home late, deleted texts, hidden money. The potential use in personal relationships would be unlimited.
Click to expand...
It violates the 4th amendment if not done voluntarily, but so does most of the shit homeland security currently does.
 
N

NickFire

Member
Moneal said:
It violates the 4th amendment if not done voluntarily, but so does most of the shit homeland security currently does.
Click to expand...
Right now I would tentatively agree with you, but no way I'm taking a final position when the technology is still an infant. I just cannot foreclose the possibility that it would be treated like making a suspect look into their phone to let police open it if the tech was conclusively sound, or removing video footage from a suspect's residence (etc.).
 
sankt-Antonio

sankt-Antonio

Member
Imo this reads more like them wanting more money for future research. I bet somewhere in the details one could find out that the image was not created on the fly by recostructing MRI scans (and only them), but there was some pre existing dataset used that makes this all possible and kind of lame. Wouldn't be the first time.

Edit: Yeah. They have datasets that show Brainactivity when viewing a certain picture. Then they check the MRI pattern against the dataset. Lame. They cant generate the image (any random image) from new MRI data. At least that's what I gathered from a quick read.
 
Last edited:
Davesky

Davesky

Member
So we can almost confirm the existence of near death experiences and the afterlife but we still can’t cure cancer. Great world to be living in.

I hope they use AI to reconstruct someone’s experience on a DMT trip.
 
Last edited:
22•22

22•22

Doesnt need recognition
nightmare-slain said:
If brain implants ever become a thing then we're fucked.
Click to expand...

We're already there with our current system. We're already fucked if they choose so. Tip toe. Normalization has to set in before you take it to the next step. Lol. This is a fucking rabbit hole and I'm not going there. Thanks for posting EviLore EviLore

Edit: I recently relistenend a interview that went into AI and "Smart Cities" etc. So interesting and scary. Wish I could share but if I remember correctly it took place in the second hour which is paywalled. Let me see.

Found it!



Sorry if offtopic. 🙈
 
Last edited:
N

NickFire

Member
Davesky said:
So we can almost confirm the existence of near death experiences and the afterlife but we still can’t cure cancer. Great world to be living in.

I hope they use AI to reconstruct someone’s experience on a DMT trip.
Click to expand...
Maybe tech like this could one day help treat / prevent cancer???

My thought process is they would be learning more and more about how the brain functions. And I've always understood that cancer was a mutation that the body did not kill off before it took root. So maybe more data about how the brain detects and attacks potential cancers could be harvested and used to invent new approaches to treating cancer.
 
H4ze

H4ze

Member
I wish we could live like back in the 90's :(
This is a bit too terrifying, some serious George Orwell shit.
 
EviLore

EviLore

Expansive Ellipses
Staff Member
sankt-Antonio said:
Imo this reads more like them wanting more money for future research. I bet somewhere in the details one could find out that the image was not created on the fly by recostructing MRI scans (and only them), but there was some pre existing dataset used that makes this all possible and kind of lame. Wouldn't be the first time.

Edit: Yeah. They have datasets that show Brainactivity when viewing a certain picture. Then they check the MRI pattern against the dataset. Lame. They cant generate the image (any random image) from new MRI data. At least that's what I gathered from a quick read.
Click to expand...
Incorrect, they used Stable Diffusion. It’s a very generalized AI imaging model that can produce virtually any image imaginable.
 
IDKFA

IDKFA

Member
I read the whole 8 page essay (I didn't look into the references) and I still don't really see what the purpose of this is?

I understand the goal for "artificial systems that see and recognize the world as human visual systems do", but what is the benefit of this?
 
EviLore

EviLore

Expansive Ellipses
Staff Member
IDKFA said:
I read the whole 8 page essay (I didn't look into the references) and I still don't really see what the purpose of this is?

I understand the goal for "artificial systems that see and recognize the world as human visual systems do", but what is the benefit of this?
Click to expand...
Humans aren’t great at understanding how the brain works. We can identify regions and see when they light up, basically. AI models, on the other hand, can already see almost exactly what the brain sees. The potential applications are endless. Look at what Neuralink is researching. Brain computer interfaces, sight/motor function restoration, treatments for depression, mind control.
 
TonyK

TonyK

Member
Maiden Voyage said:
what do u believe?
Click to expand...
In Crom.

I don't know, it simply sounds internet bullshit to me. Even if it was real, I can't believe it will work as people are assuming it works, like a brain scanner that print your memories, brain doesn't work in that way. I assume, if it's real, that they are using the eye as a camera and what they are scanning is the optical information, not the neural information, your memory.

But I don't know shit, maybe en five years everyone has a brain printer in home.
 
sankt-Antonio

sankt-Antonio

Member
EviLore said:
Incorrect, they used Stable Diffusion. It’s a very generalized AI imaging model that can produce virtually any image imaginable.
Click to expand...
The image is created by SD but it’s working off of a pool of known pictures from the database, it’s still a neural network. Its basically what DALL-E usees.
They all need a good database to generate stuff. Having MRI brain functions coupled with pictures in the database and than extracting pictures off of it is the same as DALL-E generating a new picture as a combination from its database of knows „things“. It’s cool but not groundbreaking and it wouldn’t know what to do with the MRI data from someone looking at a picture not already in the database. The dataset used in this study is mega small and specific.

They are never going to get the dataset size of DALL-E with irl human MRI scans needed to recreate thought. Only based on a logistical point of view.
 
C

CGNoire

Member
Final Fantasy: Spirits Within Dreams Recording Incoming.

.....Im still not convinced there isnt some fudging in this report to make it look more impressive than it is. Its happened too many times lately for me ro take this at face value. Something seems off. Maybe Im Paranoid or in this case not Paranoid enough ;).
 
Last edited:
G

GloveSlap

Member
I used to read Weekly World News (fake tabloid) as a kid in the early 90s. One story that always stuck with me was that "someone" invented a VCR that could record your dreams. It stuck with me because it would be amazing of course, but also because it will almost certainly happen at some point.
 
You must log in or register to reply here.
Top Bottom