1,000
1.00@quantizor As @lcar pointed out below, computers have aided research substantially for a long time. The achievement, design and a lot of actual hard intellectual work is still fairly credited to the authors of the software - as was the case for all software that came before. If the only criterion is "humans couldn't do it", you could (with a squint) say that any of logistic regression/genome assembly algorithms/<literally tons of other algs> outperform humans in science.
@quantizor That AI can do better-than-human research independently. At least to the extent humans can do it independently, e.g. probably no problem if it tasks a human with carrying out tasks in the lab as long as it is able to provide quite detailed instructions and feedback to said human.
But computers have always aided research and performed tasks that humans either couldn't do or would have taken an eternity to do. So I don't think that alone counts. I'd say the question being asked here is about AI either conducting a significant portion of total research output or becoming as generally capable or productive as the average human researcher. Would be good to have some clarification though.
Deepmind just walking into a room full of material scientists / microbiologists and telling everyone they can go home because it finished all the science.
Considering that Deepmind has already finished folding every protein known to man and successfully discovered 1000x as many new crystals that can be formed as the combined output of materials scientists in all human history it has clearly made scientific discoveries in some fields that are far beyond human capabilities. Could you be more specific about the question?
As in, will some/most/all new scientific knowledge be generated by AI in 2030? With or without human input? Or will AI be better than humans at conducting some/most/all types of research? Or will there exist an AI that is better at scientific research than the average researcher in a number of fields??