[TeMPOraL]

More generally: we need a legal framework that makes people deploying automation responsible for what the automation does to the same extent as they would be if they hired people to do the same work. "It's a false positive in an automated system" should never be acceptable justification for invalid legal action, nor should it be used as extenuating circumstances. Either you're prepared to pay for the mistakes of your algorithm, or you should not be deploying the algorithm at all.

[rectang]

If you knew that your algorithm produced false positives and you deployed it anyway, then you had fraudulent intent for the subset of automatically generated takedown notices which were false positives.

[jaywalk]

That's a nice theoretical definition of "fraudulent intent" but you're not going to get a court to agree with it, I don't think.

[rectang]

Say that the false positive rate is 3 in 4, so that 75% of takedown notices are invalid. Do we have fraudulent intent? I dare you to tell me that we don't.

OK, we've established that delegating to an algorithm doesn't provide an impenetrable shield. Now it's just a question of how irresponsible you have to be for the court to go against you.

Now, is there any difference between the injured parties who are affected by your false positives when that rate is 1% versus 75%? For the subset of assessments affecting the injured party, your false positive rate is effectively 100%. Why should they bear your burden?

An algorithm can be used to identify prospects for takedown notices. If you choose not to vet those prospects, then any mistakes are on you. If you can drive down the false-positive rate low enough, you might choose to accept the costs every time you falsely accuse someone and turn the algorithm loose anyway. But if you can't absorb the costs of the algorithm's mistakes, don't rely on the algorithm.

[yjftsjthsd-h]

Really? You run something knowing that it's going to file invalid claims; how is that not intent to commit fraud?

[jaywalk]

You put out a product knowing that it can fail in X (very long) time or Y (highly unlikely) scenario. Does that mean you intentionally put out a faulty product?

"This process is inherently subjective, so creating perfect software while properly protecting our copyrights would be impossible. We determined that the false positive rate would be quite low, and had no intention of pursuing action against any invalid claims."