24 ms·

I disagree that "deepfakes" will be unverifiable. There is very robust forensic framework for analyzing videos for authenticity, and it often relies not just on

I disagree that "deepfakes" will be unverifiable. There is very robust forensic framework for analyzing videos for authenticity, and it often relies not just on the visual component but all the other vectors that open up as a result-EM waves being generated, audio analysis, etc. A deepfake that passes this analytical function would imply that we will not be seeing advancements in digital forensics due to AI, which does not make sense because there is no reason not to update forensic processes to keep up with changing technology.

wish5031 8y ago
That seems like a plausible argument, but there’s a huge portion of the population that’ll willingly believe any realistic-looking video they see, regardless of the evidence. That could have fatal consequences, e.g. as happened with the Planned Parenthood video in 2015. [0] [0] https://en.m.wikipedia.org/wiki/Planned_Parenthood_2015_undercover_videos_controversy https://en.m.wikipedia.org/wiki/Planned_Parenthood_2015_unde...
- fossuser 8y ago
  It doesn't even have to be realistic, people will believe outrageous claims or lies that conflict current video evidence because it fits a narrative they want to be true. It seems that trying to actually determine what's true as opposed to just attempting to frame issues in support of your team (or what you already currently believe to be true) is a relatively rare quality. I guess the silver lining is that while high quality fake videos may make things worse, there's already a pretty deep divide between people who will try and figure out what's true and people who don't. Higher quality fakes may only grab a small percentage of additional people.
- Consultant32452 8y ago
  I agree. People supported a war in Iraq based largely on cartoonish photoshops of mobile nuclear weapons labs in ice cream trucks.
ezoe 8y ago
If there is a program which output the authenticity of input, Can't we use that to further improve the NN? Eventually, given enough computation, trained NN reliably output videos even the said program can't bust. I see a dark future ahead of us.
- acjohnson55 8y ago
  No darker than the past. Although, to be fair, the past was pretty dark.
- oh_sigh 8y ago
  This is the ideas behind GANs and progress is reliant upon the generator and discriminator generally can't get too far ahead of one another

wish5031 8y ago

That seems like a plausible argument, but there’s a huge portion of the population that’ll willingly believe any realistic-looking video they see, regardless of the evidence. That could have fatal consequences, e.g. as happened with the Planned Parenthood video in 2015. [0] [0] https://en.m.wikipedia.org/wiki/Planned_Parenthood_2015_undercover_videos_controversy https://en.m.wikipedia.org/wiki/Planned_Parenthood_2015_unde...

fossuser 8y ago
It doesn't even have to be realistic, people will believe outrageous claims or lies that conflict current video evidence because it fits a narrative they want to be true. It seems that trying to actually determine what's true as opposed to just attempting to frame issues in support of your team (or what you already currently believe to be true) is a relatively rare quality. I guess the silver lining is that while high quality fake videos may make things worse, there's already a pretty deep divide between people who will try and figure out what's true and people who don't. Higher quality fakes may only grab a small percentage of additional people.
Consultant32452 8y ago
I agree. People supported a war in Iraq based largely on cartoonish photoshops of mobile nuclear weapons labs in ice cream trucks.

ezoe 8y ago

If there is a program which output the authenticity of input, Can't we use that to further improve the NN? Eventually, given enough computation, trained NN reliably output videos even the said program can't bust. I see a dark future ahead of us.

acjohnson55 8y ago
No darker than the past. Although, to be fair, the past was pretty dark.
oh_sigh 8y ago
This is the ideas behind GANs and progress is reliant upon the generator and discriminator generally can't get too far ahead of one another