Sergey is a bioinformatics researcher in Saint Petersburg. He uses an AI tool to screen thousands of protein structure variants, identifies a candidate compound, and publishes a paper. His research draws on decades of prior work by biologists, chemists, and biochemists, work that was embedded in the training data of the AI he used. The compound might eventually cure a disease.
Who deserves credit?
And compensation?
This isn’t a hypothetical. It’s what research looks like now. When DeepMind’s AlphaFold predicted the three-dimensional structures of more than 200 million proteins, solving a 50-year problem in biology, it did so by learning from the work of tens of thousands of scientists who had spent careers characterizing individual proteins. Their contributions made AlphaFold possible. The question of what they are owed has no settled answer.
Existing Assumptions Are Wrong
The prevailing legal assumption is that AI output belongs to whoever owns the model or whoever prompted it. Under this framework, Sergey’s compound belongs to the cloud provider, or to the institution that paid for the subscription, or perhaps to no one because the patent office won’t grant patents for AI-generated inventions. None of these answers is right. They all ignore the most important contributors to the output: the scientists whose published work made the model capable of this prediction.