Who can you trust?

Did Robert Heinlein have the answer to deepfake photos and video?

Who can you trust?

That’s a pretty good song by Morcheeba, but it’s also an increasingly apt question for our age.  “Photographic proof” is ceasing to be proof.

We are already beginning to see readily created AI based “deepfakes” that call any audio, video, or still image into question.  This one of Joe Biden attacking Trans people is just the beginning.

The voice is perfect.  The lip movement not quite, but that will come soon. 

And this is an AI-generated picture from a collection of AI-generated photos on the nonexistent 1963 Burning Man festival.

This kind of thing will mark a sea change in terms of what counts as dispositive evidence.  And that will be a problem.

In recent years, a common Internet refrain is “pics (or video) or it didn’t happen.”  Those were seen as proof.  And generally they were.

Of course, photos and moving images have been faked for years.  Famously, a film clip of Hitler celebrating France’s surrender in 1940 was spliced and looped to make it look as if Hitler was “dancing a jig” in excitement, when in fact it was a deliberate fake:  “John Grierson, a Scottish film maker and director of the Canadian Information and Propaganda Service, saw an opportunity. He and Stuart Legg, an English filmmaker, realized that the clip of Hitler’s step could be looped, repeated, and turned into a little dance step that would make the Fuhrer look ridiculous. They worked their film wizardry so well that many people thought Hitler had indeed danced a celebratory jig.”

Robert Capa’s famous “falling soldier” photo from the Spanish Civil War was also probably faked.  If so, it joins a list of famous faked war photos, from the U.S. Civil War on up.

But while we accept that photos in high-stakes areas where resources are available – like war propaganda – might be fake, we generally assume that ordinary pictures, from cameras, phones, security video, and the like are genuine.  It’s too hard to fake a picture for ordinary purposes.

Well, it was too hard.  Now it’s easy.  Anyone with a computer and an Internet connection can produce persuasive fakes, and the quality is only going to improve over time.  Probably sooner than later.

So what do we do?  How do we prove that things happened, or didn’t happen, if we can’t trust audio, video, and photographs?

I think the solution is to go back to the future.

Remaining in the image space, it’s electronic images that are most questionable.  A half-measure relies on the fact that pictures and movies made with film are harder to fake, and the existence of negatives that can be examined is some kind of guarantee of genuineness.  That, of course, will be fakable too, in time probably even cheaply, but it’s a steeper hurdle of fakery.

There will also be proposals for, and perhaps the creation of, “incorruptible” systems based on encryption and so forth.

But that, too, will be overcome.  (People like to say that “encryption always wins,” but that’s not its historical record, as ULTRA and MAGIC illustrate.)  So we need to go even further back to see our testimonial future.

Prior to the 20^th Century, photographic evidence was seldom available; before the late 19^th century it basically wasn’t available at all.  So what did courts, including the court of public opinion, do?  They relied on first-person testimony.  We may be forced to rely largely on that again.

There are problems.  Witnesses can be unreliable, poor observers, or simply suborned.  Law school evidence classes often open with staged crimes designed to demonstrate the generally poor recall of eyewitnesses.  (My former colleague Neil Cohen used to have someone run in and shoot him with a blank-firing prop gun, then run out, and quiz students on the characteristics of the shooter.)  Nonetheless, human witnesses have a few advantages, one of which is that human beings – whose evolution was driven in no small part by learning how to tell when other humans were lying – are better at evaluating the credibility of people than of machines.

For important matters, we might also rely on specially trained observers, and there’s actually a science-fictional precedent for that.  It’s an institution that plays a minor role in Robert Heinlein’s Stranger in a Strange Land, the “Fair Witness.”  The Heinlein Society’s concordance describes a Fair Witness as a “Person rigorously trained to observe, remember, and report without prejudice, distortion, lapses in memory, or personal involvement.”  The character Anne, one of Jubal Harshaw’s staff, is a Fair Witness.

That role plays a minor part in the book, so we don’t get a lot of detail, but we know that Fair Witnesses are designed to be neutral, and exact, as this passage illustrates:

Jubal to Jill: "Even Cavendish did not--at least he won't say so. You know how Fair Witnesses behave."

Jill: "Well...no, I don't. I've never met one."

Jubal to Jill: "So? ANNE!"

Anne was on the springboard; she turned her head. Jubal called out, "That house on the hilltop--can you see what color they've painted it?"

Anne looked, then answered, "It's white on this side."

Jubal went on to Jill: "You see? It doesn't occur to Anne to infer that the other side is white, too. All the King's horses couldn't force her to commit herself...unless she went there and looked--and even then she wouldn't assume that it stayed white after she left."

Fair Witnesses also have robes that they wear when on duty, which may grant them certain unspecified privileges and immunities as they observe, Heinlein implies.

Specially trained and reliable witnesses would certainly be a help.  But, of course, they’re humans and thus fallible and corruptible.  (Heinlein, a creature of his times, was a pretty big believer in institutions and professionalism.  The past decade has largely served as a refutation of both.  And even in his day, the institutions and professions were less trustworthy than we thought; it was just harder to find out when they were lying, sort of a meta-case of what I’m writing about here.)

Technology might help some, as it will probably soon be able to tell if people are lying via brain scans with high reliability.  (I doubt it will be able to tell if they’re just wrong, though).  And that technology offers its own set of – very troubling -- problems that go way beyond this essay.

The other alternative is just to muddle through.  The legal system, and public opinion, managed to function adequately without dispositive proof via pictures and film, and so can probably do so again.  It won’t be as easy to produce evidence that convicts or exonerates someone clearly, but we’ll get by.  (And even now, how many cases or controversies are dispositively answered by imagery?  Some, but not most).

Like the song says, we’ll have to muddle through somehow.  We always do.

Share any thoughts in the comments, and if you like this piece please subscribe, or consider upgrading to a paid subscription.  Thanks!