3 ms·
AI image generation advances rapidly. Can we still tell if a picture is fake?
- xyzsparetimexyz 4y agoSurely adversarial neural networks trained to detect fakes are the answer here?
- mattdesl 4y agoIf a network can successfully detect a fake, then it will be incorporated into GANs to produce images that are no longer detectable as fake by the discriminator. This is the whole basis for generative adversarial neural networks.
- bergenty 4y agoUntil the fakes are so good that it will be impossible to keep them out of the training set which means the ability to distinguish them disappears.
- freddealmeida 4y agoYes we can tell.
- schwartzworld 4y agoCould we tell if an image was fake before? There are plenty of cases where people have been fooled by images, even predating computers and photoshop: https://www.latimes.com/archives/la-xpm-2003-apr-07-oe-russ7-story.html https://www.latimes.com/archives/la-xpm-2003-apr-07-oe-russ7... In many cases, the way people learn that images were faked is by someone recognizing details that only they could have known. For example, in the case where Oprah's head was put on Anne Margeret's body on the cover of TV Guide, the fake was discovered because someone recognized the dress, not because of a formal analysis of the photo in some way.
- hackeraccount 4y agoImages can lend weight to an argument but I don't think they've ever been over whelming - that said the weight they lend has certainly gone down over time. I think this skirts around some larger truth that the greater weight a source - photographic images , science, religion - has the more the incentive there is to manipulate it and so the less weight it carries. The incentive - fundamentally the ability to change reality - is so great that it's hard to imagine it being otherwise.
- O__________O 4y agoIf image provenance ever becomes significant enough issue, simply be solved by: (a) requiring creators/systems cryptographically sign works; (b) systems used for viewing content will either block or warn users of lack of a signature; (c) etc. Major companies including Intel, Adobe, etc - are already working on this too.