Discussion about this post

User's avatar
Tim Boudreau's avatar

I’m a software engineer, and have written image fingerprinting algorithms.

Yes, there’s a financial component to why copyright enforcement works so well, while deepfake enforcement doesn’t.

But the engineering problem for the latter is MANY orders of magnitude more difficult, and is simply not solvable to anywhere near the degree of accuracy music identification is.

You can answer the question “does this content contain these relative musical pitches sustained for these durations, with these relative gaps between them” using simple, relatively computationally inexpensive Fourier analysis, and not be wrong most of the time.

We have nothing quite like Fourier analysis for *people*. What comes closest is facial recognition algorithms, using ratios of facial feature distances. But they are VASTLY less accurate.

With any system like this, you can choose to err on the side of false positives or false negatives. Both are harmful. So the system needs to be highly accurate or it will do more harm than good.

So, yes, there’s a financial/political aspect to it, but in the case of music, you have the convergence of a mature music industry with money to throw at the problem, *combined with* a problem that is technically executable at scale, executable with high accuracy, at fairly low cost. That is simply not the case - in terms of accuracy, cost, or the existence of an organized lobby of … people with faces … with organizations like ASCAP and BMI already in the business of delivering royalties for use of those faces. So underlying it all is the collective action problem - a Google or an Apple has no use or capacity for millions of *individuals* approaching them with complaints - they spend a lot of money black-holing that sort of thing.

Then, there’s the legal issue: copyright is well-defined. There’s no such equivalent for a face. The closest I can think of is in the 90s when Billy Joel trademarked his face (yes, really, that was a thing) - but trademark law is nowhere near robust enough, since it contains a lot of caveats about how a trademark is used and the domain it applies to; whereas copyright straightforwardly is, if you use it, you are infringing, with a few minimal carve-outs for journalism and fair-use of small sections.

So, the legal regime that would support doing “content ID for faces” doesn’t really exist - not with the kind of bright-line, unambiguous tests that copyright offers.

If you want to solve this, start the organization that will act as the ASCAP or BMI of faces, mitigate tons of humans to register their faces through it (through a promise of monetization) so the Apples and Google’s of the world can costume that database and automate the process. And lobby your legislators for law that makes enforcement practically possible. And pray for engineering advancements in facial recognition, which isn’t as good as the people making money off it would like you to believe.

VedJournal ✍️📜's avatar

Wonderful Read

Very unique and fabulous work

Thanks for sharing and keep WRITING 💫

29 more comments...

No posts

Ready for more?