AI-created faces now look so real, humans can't spot the difference
- Reference: 1645423151
- News link: https://www.theregister.co.uk/2022/02/21/ai_created_faces_fool_humans/
- Source link:
Two boffins – Sophie Nightingale from the Department of Psychology at the UK's Lancaster University and Hany Farid from Berkley's Electrical Engineering and Computer Sciences Department in California – studied human evaluations of both real photographs and AI-synthesized images, leading them to conclude nobody can reliably tell the difference anymore.
In one part of [1]the study – published in the Proceedings of the National Academy of Sciences USA – humans identified fake images on just 48.2 per cent of occasions.
[2]
In another part of the study, participants were given some training and feedback to help them spot the fakes. While that cohort did spot real humans 59 per cent of the time, their results plateaued at that point.
[3]
Faces used in the study. Click to enlarge
The third part of the study saw participants rate the faces as "trustworthy" on a scale of one to seven. Fake faces were rated as more trustworthy than the real ones.
"A smiling face is more likely to be rated as trustworthy, but 65.5 per cent of our real faces and 58.8 per cent of synthetic faces are smiling, so facial expression alone cannot explain why synthetic faces are rated as more trustworthy," wrote the researchers.
[4]
[5]
The fake images were formed using generative adversarial networks (GANs), a class of machine learning frameworks where two neural networks play a type of contest with one another until the network trains itself to create better content.
[6]File suffixes: Who needs them? Well, this guy did
[7]Journalist won't be prosecuted for pressing 'view source'
[8]Beware the big bang in the network room
[9]IT technician jailed for wiping school's and pupils' devices
[10]20 years of .NET: Reflecting on Microsoft's not-Java
The technique starts with a random array of pixels and iteratively learns to create a face. A discriminator, meanwhile, learns to detect the synthesized face after each iteration. If it succeeds, it penalizes the generator. Eventually, the discriminator can't tell the difference between real and syenethesised faces and – voila! – apparently neither can a human.
The final images used in the study included a diverse set of 400 real and 400 synthesized faces representing Black, South Asian, East Asian and White faces. Male and female faces were included – unlike previous studies that primarily used White male faces.
White faces were the least accurately classified, and male White faces were even less accurately classified than female White ones.
[11]
"We hypothesize that White faces are more difficult to classify because they are overrepresented in the StyleGAN2 training dataset and are therefore more realistic," explained the researchers.
The scientists said that while creating realistic faces is a success, it also creates potential problems such as nonconsensual intimate imagery (often misnamed as "revenge porn"), fraud, and disinformation campaigns as nefarious use cases of fake images. Such activities, they wrote, have "serious implications for individuals, societies, and democracies."
The authors suggested those developing such technologies should consider whether the benefits outweigh the risks – and if they don't, just don't create the tech. Perhaps after recognizing that tech with big downsides is irresistible to some, they then recommended parallel development of safeguards – including established guidelines that mitigate potential harm caused by synthetic media technologies.
[12]
There are currently ongoing efforts to improve detection of deepfakes and similar media, such as building prototype software capable of detecting images made with neural networks. A Michigan State University (MSU) and Facebook AI Research (FAIR) [13]collaboration last year even suggested the architecture of the neural network used to create the images.
But The Register recommends against taking any of Meta's deepfake debunking effort at … erm … face value. After all, its founder has been known to put out [14]images himself that will never ever ever ever leave the uncanny valley, thereby proving that, as narrow as that valley has become as a result of this study, it's here to stay. ®
Get our [15]Tech Resources
[1] https://www.pnas.org/content/119/8/e2120481119?fbclid=IwAR3Q-z0irdG4IuRuxJQYDA_xZ1amAsBF89dnfrE5PSEou7FloswFdBrW2EU
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YhNw20-QYd24F3dL9MfZLQAAAMA&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://regmedia.co.uk/2022/02/21/synthetic_versus_real_images.jpg
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YhNw20-QYd24F3dL9MfZLQAAAMA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YhNw20-QYd24F3dL9MfZLQAAAMA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2022/02/18/on_call/
[7] https://www.theregister.com/2022/02/15/missouri_html_hacking/
[8] https://www.theregister.com/2022/02/14/who_me/
[9] https://www.theregister.com/2022/02/14/adam_georgeson_jailed_it_technician_welland_park/
[10] https://www.theregister.com/2022/02/15/20_years_of_dotnet/
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YhNw20-QYd24F3dL9MfZLQAAAMA&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YhNw20-QYd24F3dL9MfZLQAAAMA&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://www.theregister.com/2021/06/17/facebook_deepfake_identification/
[14] https://www.theregister.com/2021/06/25/facebook_ceo_reveals_new_board/
[15] https://whitepapers.theregister.com/
S or R?
Real and ? I'm confused on a Monday morning.
Re: S or R?
Simulated? But what do the percentages mean? It probably says in the text, but it's monday AM here too.
Re: S or R?
Simulated? But what do the percentages mean?
It doesn't say it in the text (at this moment). Similarly puzzled I checked twice.
The caption of this image in [1]the original study however says:
The most (Top and Upper Middle) and least (Bottom and Lower Middle) accurately classified real (R) and synthetic (S) faces.
So the percentage is the number of people that rated the picture correctly as R or S.
Having a think about it though: what is it with the human species that we always end up making everything around us become/ generate a reflection of ourselves? I mean, there Shirley must be more useful challenges for that invested time/ tech/ money/ energy?
[1] https://www.pnas.org/content/119/8/e2120481119?fbclid=IwAR3Q-z0irdG4IuRuxJQYDA_xZ1amAsBF89dnfrE5PSEou7FloswFdBrW2EU
Re: S or R?
Nope, because nothing is more precious to us than our identity. It's damn instinct, ecen.
Fascinating
This is a bit of AI that fascinates me. Seeing all the fake faces the AI creates yet they look like a real live human. Sent one image to a group of friends asking if they knew the person. Obviously everyone said no, then the real that the face doesn't exist, just AI created. Appears I was the only one that found this interesting.
Re: Fascinating
So.. you're the only robot (sorry, synthetic person) among all your friends?
Re: Fascinating
I find it interesting. This is the first step to an AI that can create a talking, moving, acting face and body.
In just a few years, your favourite actor may never grow old and you can have a continuous sequence of "Raiders..." films without the need to "introduce" a replacement character in an attempt to secure the franchise as the original star moves slowly but surely towards the retirement home.
Now all they need is an AI to produced decent plots and dialogue and Hollywood can close shop.
"The authors suggested those developing such technologies should consider whether the benefits outweigh the risks – and if they don't, just don't create the tech."
This approach doesn't even really work long-term for WMDs, which are insanely difficult and expensive to research. I don't see how it could possibly work for something that only takes readily-available specialist training and off-the-shelf computer tech.
People who are concerned about fake news need to get a grip. It's not a technical problem, it's a political problem. Stop looking for technical solutions, start looking for political solutions.
"People who are concerned about fake news need to get a grip. It's not a technical problem, it's a political problem. Stop looking for technical solutions, start looking for political solutions."
Then there's no solution because political problems are icreasingly personal problems: matters of identity. And there are few things more vigorously defended than identity.
Good. Start feeding about a trillion to Clearview to poison their data.
Detail and perception
The highest resolution images in the data sets are 1024x1024 pixels. The typical facial image as reproduced in the paper, is just under 0.6 of a frame wide, so the facial image relied on is at best a mere 600 or so pixels wide. That's a pretty low resolution for adequate reproduction of small scale characteristics we use subliminally all the time to recognise individuals.
Furthermore, as a general rule, all studies that are based on human judgement suffer from an uncontrolled factor - the attentiveness of said subjects (the same problem that besets the Turing test). Plus, the paper does not disclose the breakdown (e.g. by ethnicity) of the subjects - a factor that could introduce bias (particularly in respect of the 'trustworthiness' judgement).
Consequently, although this is an interesting preliminary study, it would be unwise to assume it represents a universally applicable finding.
Trouble
I have trouble spotting fake faces.
The papers (real or online) are full of 'celebrities' that I don't recognise and I'm sure most of them are synthetic.
The one called Rylan(?) is obviously a fake. That painted on beard and the 'Banana Splits' teeth. Not going to fool anyone.
But for the rest everything is so instagram filtered and tweaked it's difficult to tell which is which.
...and if they don't, just don't create the tech...
Yeah, like this has ever worked.
I for one would like to get a fake face for myself which looks pretty much like me. Specifically for use during early morning video calls. I believe it would be beneficial for all participants: me not needing to pretend that I'm awake and the others not needing to pretend that they don't notice me pretending to be awake.