Since humans are allowed to learn from and mimic intellectual property and make money from it, how smart does the AI have to be for us to give it the same rights?
This is where I land too. I do recognize the economic impact will take my livelihood too eventually.
ML algorithms are transforming the input to a latent space & learning the parameters that minimize the loss with in that space. Copyright prevents copying from the source material. But the act of mimicking style or repeating a concept mentioned elsewhere isn't protected. Sure, we'd like references in academia, but as long as the material doesn't significantly resemble the source material it has been allowed.
We are able to do it artificially now and that allows it to be done at unprecedented scale.
Society grants ownership to incentives creation. Not out of respect for Justice or fairness as your argument implies. AI doesn’t need an incentive to create and the marginal cost to create goes down almost infinitely once the AI training cost has been recouped.
It will be interesting to see a case trying to argue competition.
On one hand, a static training with a cutoff date and no perfect memory of training data makes for a hard case that the underlying use of the copyrighted material constitutes a competition with a newspaper (where the value is in accuracy archival information or very recent information).
But on the other, the eventual product's extended applications are certainly capable of being a significant competitor.
It just seems weird to see that argument floated when the actual usage is so tenuously connected to the competition.
If the NYT was suing an individual who read it for years, improving their own writing style and vocabulary, who then went on to write articles for a competing newspaper, that would clearly be laughed out of court.
There's probably going to be a significant financial incentive in light of ongoing and upcoming litigation to nailing down what's actually going on in the NN and how much is a stochastic regurgitation and how much is building world models.
Why doesn’t it benefit society when it is the basis for compensation for large and important creative industries? Maybe it lasts too long given that the life of the author plus decades results in protection lasting longer than necessary to incentivize authors to create works.
There's a lot protected by IP law that has real impact on people's lives, like textbooks or anything medical, and that we're in theory capable of unlimited distribution for better-than-free (in that the resources currently being put into preventing distribution exceed the resources that would be needed for distribution) but instead restrict to a comparatively tiny number of people in order for profit to be extracted.
I don't know whether or not other methods of creating an "incentive", like public funding, will work as full replacements - but I think it's worth acknowledging how much is being left on the table. An immense amount of waste and even death could be prevented if there's a viable way to tear down significant portions of IP law.
> Last month, the Associated Press became one of the first news organizations to strike a licensing deal with OpenAI, but the terms of the deal were not disclosed.
I would give a lot to be a fly on the wall during that negotiation.
11 comments
[ 2.8 ms ] story [ 20.3 ms ] threadIt needs about as much rights as a PostgreSQL database.
On one hand, a static training with a cutoff date and no perfect memory of training data makes for a hard case that the underlying use of the copyrighted material constitutes a competition with a newspaper (where the value is in accuracy archival information or very recent information).
But on the other, the eventual product's extended applications are certainly capable of being a significant competitor.
It just seems weird to see that argument floated when the actual usage is so tenuously connected to the competition.
If the NYT was suing an individual who read it for years, improving their own writing style and vocabulary, who then went on to write articles for a competing newspaper, that would clearly be laughed out of court.
There's probably going to be a significant financial incentive in light of ongoing and upcoming litigation to nailing down what's actually going on in the NN and how much is a stochastic regurgitation and how much is building world models.
I don't know whether or not other methods of creating an "incentive", like public funding, will work as full replacements - but I think it's worth acknowledging how much is being left on the table. An immense amount of waste and even death could be prevented if there's a viable way to tear down significant portions of IP law.
I would give a lot to be a fly on the wall during that negotiation.