Are there any FOSS licenses that explicitly allow the use of the software as AI training data with no restrictions, and explicitly disclaim any copyright interest in the output of the AI?
I'm effectively looking for the opposite of this.
Open Source Stack Exchange is a question and answer site for people organizing, marketing or licensing open source development projects. It only takes a minute to sign up.Sign up to join this community
There surely are FOSS licenses that explicitly allow the use of the software as AI training data with no restrictions. For a start, consider every license that allows the set of all possible uses with no restrictions (e.g., ultra-permissive ones like the WTFPL and CC0).
Looking at the question you link, it's either the case that
a particular AI training routine abstracts away the copyrightable components of a piece of code sufficiently to be not infringing, or
a particular AI training routine does not do so and its output therefore is a derivative work of related input(s).
In the event that case #2 holds, you want to grant permission for the use of your software as training input to allow the redistribution of the AI output.
Probably this is fine to allow for a FOSS license (assuming you also grant freedoms for usual uses by human users/developers), but it will probably, in effect, allow ultra-permissive use of your code in all cases anyway. Suppose a particular model, trained on your software as input, reproduces it verbatim (or nearly verbatim). Your stated intent is that this should be allowed and you wish to disclaim any copyright interest in this output (i.e., your code itself). Therefore anyone may use this output without restriction by, effectively, running your code through a machine-learning-powered identify function.
I'd say ultra-permissive licenses are not merely sufficient for your specifications but are also the only set of rights you could effectively grant.