😖 AI Hallucinations Could Help Scientists
LLMs sometimes generate unreliable or false information— hallucinate. CEO of OpenAI Sam Altman said that the problem of AI hallucinations will be solved in the next couple of years. But some scientists consider them useful.
Of course, we are not talking about cases when AI models fabricate or distort facts. But often, the models generate unexpected hypotheses that are not obvious to humans, and such creativity helps scientists.
"The public thinks it's all bad. But it's actually giving scientists new ideas. It's giving them the chance to explore ideas they might not have thought about otherwise," Amy McGovern, director of the Institute for Climate AI Research, explains.
Examples:
🧪 AI hallucinations have accelerated research into new antibiotics, according to MIT bioengineering professor James Collins.
🧪 Last year's Nobel Prize laureate in chemistry, David Baker, credits AI's bursts of imagination as a crucial element in creating proteins never before seen in nature. He was inspired in part by the principle behind Google's DeepDream, a model that morphs existing images into psychedelia.
🧪 Researchers at the University of Texas use "hallucination methods" to train robots to navigate complex environments. This approach generates training data without the need of expert demonstration or trial and error.
🧪 AI "fantasies" have helped UCLA researchers develop a new catheter that significantly reduces the risk of bacterial infections. Hospitals will soon start purchasing it.
🧪 Meteorologists at the University of Oklahoma use the rich AI imaginings to discover unexpected factors that can drive extreme events like deadly heat waves.
More on the topic:
🧬 Can AI Actually Create? Yuval Noah Harari Responds.
