- From: Paola Di Maio <paoladimaio10@gmail.com>
- Date: Fri, 4 Nov 2022 18:48:01 +0800
- To: Dave Raggett <dsr@w3.org>
- Cc: W3C AIKR CG <public-aikr@w3.org>
Received on Friday, 4 November 2022 10:50:16 UTC
Dave, even the NN you point ot uses logic/KR to figure out there is a mismatch between the lip movement and the words. I think you are simply confirming the point that even ML needs KR PDM On Fri, Nov 4, 2022 at 6:30 PM Dave Raggett <dsr@w3.org> wrote: > > > On 3 Nov 2022, at 21:32, Paola Di Maio <paoladimaio10@gmail.com> wrote: > > *how to use KR to identify Deepfakes (short answer: by using logic)* > > > That isn’t accurate, as it is possible to train neural networks to spot > deep fakes by looking for clues that suggest machine generated images and > video frames, see: > > > https://hai.stanford.edu/news/using-ai-detect-seemingly-perfect-deep-fake-videos > > The article notes that this will get harder over time as the generators > get better. > > Logic isn’t needed for this and the KR is implicit in the network > architecture and training data. > > Dave Raggett <dsr@w3.org> > > > >
Received on Friday, 4 November 2022 10:50:16 UTC