AI cannot hold IP
By agreement of UNESCO (Recommendation on the Ethics of Artificial Intelligence) legal personality and rights cannot be held by an AI system itself. This is because the penalties of law are only effective against humans. An AI system independent of humans is the ultimate shell company, purely an available hiding place for corruption by the human agencies that set it in place. cf. my 2017 article with two law professors: Of, for, and by the people: the legal lacuna of synthetic persons. This was also the consensus of all the experts testifying at this January 2022 British event run by members of the British parliament: AI and Intellectual Property Rights: IPR protection for AI-created work. (Monday, 24th January 2022 | 5:30-7:00 PM London - global webinar) I was astonished at the level of consensus, here mostly from law experts from industry.
Who does hold IP can and should be agreed contractually
This means that the questions of who owns the IP comes down to some combination of
- the (legal) person who caused the art to be built,
- any rights they have assigned or agreed to the developers of the AI system, and
- any rights those developers assigned to – or
- were claimed by – any copyright holders of data used to train the system.
One thing any customer using AI to create art should demand is clarity from the system providers about how these four possible sources of ownership are to be divided. Note that these are not particularly hard or unusual legal questions; it's just that a lot of AI companies aren't used to any kind of legal compliance.
Note: Not every AI art system is derived by machine learning from data. For example: AARON is a system commissioned and developed by the artist Harold Cohen, who basically worked with programmers to derive rules for making art very much like his own. This work took decades, but anyone can make art through AARON now.
 |
| I took this picture of someone else's art. |
Comments
Where there is no human input or it is negligible, the work might not be copyrightable. The tricky bit is the degree of human intervention acceptable for copyrightability, and here again there is no agreement.
Recent US Copyright Office cases (see the case of Kristina Kashtanova) show the US regulator, for example, is setting the bar sufficiently high: they disregarded the efforts (which some may say are creative in themselves) required to construct a Midjourney prompt and then further efforts to sort through numerous less interesting outputs in search of more interesting.
Accordingly, if and when the regulator, and, more importantly, the higher courts will take the position that copyright does not subsist in a certain type of content, this deficiency cannot be cured by contract, thus eliminating the possibility to legally oppose commercial exploitation of identical content by others.