In talking with someone about “preferred form for modification” over the weekend at FOSDEM, the FSF (now sort-of-OSI?) four freedoms came up. They’re not bad, but they’re extremely developer-focused in their “use case”. This is not a new observation, of course, but the changed technical and social context of AI seems to be bringing it to the fore that different users have different variations on the values and why open is so important to them.
Here’s a very quick, rough cut of what I proposed instead as key freedoms, and who those matter for. These are not exclusive (eg in many cases developers also care about replication; businesses obviously frequently care about modification, etc.) but compared to the original four freedoms, convey that different stakeholders have different needs—all of which have been served, but not explicitly called out as metrics, by the FOSS movement over the years.
- modification (foremost, for developers): We like to play with things, and often can make them better in the process. Enough said.
- replication (foremost, for scientists): It’s not really science if you can’t replicate it, and you can’t replicate it if it isn’t really yours. High overlap with modification and transparency, but something other constituencies can often live without.
- transparency (foremost, for governments): you can’t effectively regulate what you can’t understand, and it’s never OK for something that needs to be regulated to be opaque. (Obviously we allow this all the time but as we’re all reminded this week that leads to all kinds of malignancies.)
- cost of re-use (foremost, for business): This is perhaps the least unique, but it’s important to remember that statistically businesses very rarely modify open source. They pick and choose what they use, and compose them in almost entirely unique ways, but that’s a property of architectural design and cost, not of ability to modify.
These of course get you to mostly the same place as the traditional four freedoms. But importantly for the discussion of “open in AI”, replication and transparency require a focus on data, while for many businesses (and certainly for most developers) weights are sufficient and may well be preferred in many/most cases.
One could imagine other use cases and priorities, of course. But I wanted to bang this out quick and there was a nice symmetry to having four. Leave more on the various socials :)