[toomanybeersies]

Speaking of the effect this would have on peoples memories, there's also the potential to use these tools to gaslight [1] someone.

An abuser could make images where a person was at an event they were never at, or with a person they never met.

> "You've totally met Steve before, here's a photo of you with him, how do you not remember?"

An abuser could even more effectively tear down someones reality than ever before. If they were having an affair with someone they just met, they could claim to be old school friends catching up, just insert them into an old photo.

Obviously, it's not all bad. There is the potential for this to be used for good as well, but I'm a pessimist.

[1] https://en.wikipedia.org/wiki/Gaslighting

[Bartweiss]

I mean, there's presumably a very short window for that before photo evidence becomes unconvincing to people. The first time a Senator gets "exposed" for some misdeed but proves the evidence is fake, "there's a photo of this" loses its punch. The surrealism of seeing a fake recreation of oneself might have some impact, but we handled ultra-realistic paintings alright.

It does touch on an interesting point, though: we've had roughly 100 years in which photo and audio recreations of events constitute "hard evidence" beyond our ability to fully falsify. It appears that within the next ~20 years we'll lose that reliability - footage of a politician making a dirty deal or a businessman engaging in conspiracy will become deniable not just as a misleading edit, but as outright fabrication.

What do we do at that point? Do smartphone videos get automatically hashed and uploaded to a blockchain somewhere, so that we can prove when the video came into being? Do we return to an 1850s sense of news, where claims effectively cease to be falsifiable except via personal experience? Are we ready for any of this?

[toomanybeersies]

In today's political climate, even with incontrovertible evidence, all you need to do is shout out at the top of your lungs: "FAKE NEWS", and it doesn't matter any more.

[kumarvvr]

Giving in to those who shout so is exactly their goal.

We, as technologists, should come up with solutions that allow such shouts to become meaningless.

[intended]

If you make shouts meaningless you make screams of help meaningless

[kumarvvr]

I meant shouts of deliberate misrepresentation.

[Bartweiss]

As national political news to a cynical, disinterested audience, fine.

But that's thinking too small: what happens when every recording of corporate misdeeds, every photo of a cheating spouse, even a child's baby pictures, lose their solidity as evidence? What happens to anonymous online conversations when "pics or it didn't happen" becomes "it didn't happen"?

[lordnacho]

The other side of that, detection of frauds, also benefits from current work in the ML/AI/etc.

Adversarial algos sound quite interesting. It might mean that we'd have to look at the evidence of fraud in a whole new way though, because the thing that gives away the fake might not be obvious.

[SideburnsOfDoom]

> I mean, there's presumably a very short window for that before photo evidence becomes unconvincing to people.

Conversely, the people who want to believe false things (and under the right circumstances that's most of us) will be easier to convince. What is truth anyway?

[ryacko]

That's not really true. Maybe fifty years when cameras were of a sufficient quality that touchups would be noticed compared to a photorealistic painting.

[genericacct]

My guess is that signatures will be forged via NN before pictures.

[Bartweiss]

I agree, but I wonder how much this matters? Are signatures still counted as a hard form of identity verification all that often?