Open access and AI
Open Access was supposed to be the result of a movement for equality in research. Researchers should have equal access to the scientific knowledge irrelevant of their origin or work place conditions. Scientific knowledge should not be walled in order to keep out the underprivileged.
That was a great idea and some of the people I respect the most took part in the movement. For the record, I still totally support open access.
The problem arises with the use of AI as a research tool. Clearly, in the years to come, having access to a good AI model will be fundamental for researchers. Those models are trained on all existing knowledge and can help whoever has access to them. This means that we moved “access” from the document to the model.
Currently, access to those models is impossible for the underprivileged, and I seriously doubt it wll ever be “open access” for all researchers. Thus, I believe that allowing them to train on all existing knowledge is creating the same, if not worse, gap that we had before open access to publications.
I see two possible solutions:
- Restrict open access with very clear non-commercial/non-derivative work closes.
- Develop a system where if my paper was used for training, I have access to the model.
The second option has the viral/contamination element the GPL had and I assume it will be rejected.