I had the same thought. "Source Code" is a human readable and modifiable set of instructions that describe the execution of a program. There's obviously parts of an AI system that include literal code, usually a bunch of python scripts or whatever, to interact and build the thing, but most of it is on the one hand data, and on the other an artifact, the AI model and neither is source code really.

If you want to talk about the openness and accessibility of these systems I'd just ditch the "source" part and create some new criteria for what makes an AI model open.

There's code for training and inference that could be open-source. For the weights, I agree that open-source doesn't make sense as a category.

They're really a kind of database. Perhaps a better way to think about it is in terms of "commons". Consider how creative commons licenses are explicit about requirements like attribution, noncommercial, share-alike, etc.; that feels like a useful model for talking about weights.

I have heard government people talk about "the data is open-source" meaning it has public, no cost copy points to get data files e.g. csv or other.

Yeah, it's like an open-source jacket.

I don't really know what you're referring to....