How AI Can Help Researchers Fail

Artificial intelligence has supercharged many elements of the research operation: Scientists can write more papers, analyze more data more quickly and run test more efficiently. In their rush to innovate and make breakthroughs, researchers should also use AI to make mistakes, says Venu Govindaraju the senior vice president for research, innovation and economic development and a distinguished professor at the University at Buffalo. 

“Given that AI is going to bring so much efficiency, instead of just five or 10 hypotheses, let’s say we should test 100 hypotheses because we should test the ones which are risky,” he said in a recent episode of The Key, Inside Higher Ed’s news and analysis podcast. 

Scientist should consult AI to come up with the best hypotheses to test, but then also use its power to recreate the surprise factor, he said.  “Many of the scientific breakthroughs happened because of serendipity, because of some unexpected collisions of people and ideas and forums and so on, and we have to give that a chance,” he said. “If we simply follow the hypotheses that an AI platform is proposing, then chances are we will improve science and make it faster. But we’ll probably miss out on a whole bunch of other ideas, which could also bloom.”

Scientists gain knowledge through understanding what doesn’t work and why, he said, acknowledging that intentional failure requires changes in the research ecosystem. At the renewal stage of grants, “we should not just look for those ideas that succeeded, but also fund and promote some bold initiatives,” he said. “In a rewarding system, I cannot work unless I am funded by the agency. So the agencies have to be deliberate in ensuring that they are rewarding some of these other pathways as well. So as long as that is happening, the best ideas will keep moving forward and will keep advancing.”

Listen to the full episode here.