I agree with this post and with this passage in particular. I understand it’s not easy to come up with a fix, especially when keeping into account the differences among jurisdictions. Agreeing on some general principles on how to interpret such requirements in the future could already be very helpful.
In particular, does “skilled person” include skills such as
- raising capital (to pay for data access, licenses, computing power, data annotators);
- getting connections (to find people willing to sell data to you);
- dodging the law (fending off copyright lawsuits etc.).
Copied from comments on the previous draft, for ease of linking:
I think the point is that “a skilled person can recreate” isn’t true without the datasets. The sentence can be clarified by rephrasing “using the same or similar data” into “if able to gain access to the same or similar data”, or alternatively by weakening “substantially equivalent system” by defining it as a system which would be equivalent if trained on the same data but which gives different results for lack of access to the training data.
And shujisado said:
Person having ordinary skill in the art - Wikipedia
Since the term “skilled person” is used in the realm of patent law, I am also somewhat concerned about the appropriateness of using this term in OSAID.
If it is used as a legal term, it may have different meanings in different countries.
Followed by Matija Šuklje:
That is a good point, Sado-san. I actually was thinking about the “skilled person” the other day in the sense of the difference between why FOSS and OSAI and why for AI we require more than for sw.
The way I see it, with FOSS, if you have the source code – even without any additional documentation, in-code comments, or otherwise – the assumption is that this is enough for a “skilled person” to be able to “study” and “modify” the software. The source code is the absolute minimum needed, everything in addition to that makes it easier.
So what we should be aiming with the OSAI definition, is what is the minimum a “skilled person” requires to be able to (“use”), “study”, “modify” (, “share”) the AI system – nothing more and nothing less.
As such, I think the “skilled person” should be moved one paragraph higher to:
“The preferred form of making modifications for a machine-learning Open Source AI must include everything that a skilled person needs to recreate a substantially equivalent system, which includes:”