AI models find better medicines by embracing their own mistakes
By teaching computers to doubt their own predictions, researchers are uncovering powerful new drug combinations that traditional, high-confidence systems completely overlook.
Most artificial intelligence is trained to be a perfectionist, discarding any result it cannot verify with absolute certainty. However, a team from Brookhaven National Laboratory and Texas A&M University discovered that this rigid pursuit of accuracy actually blinds computers to the most innovative medical breakthroughs. By intentionally programming their models to embrace 'uncertainty quantification,' they allowed the AI to explore the murky, less-understood corners of chemical space.
There's more to this story — open the app to keep reading.
Continue Reading in App
1 more paragraphs · plus a 2-question quiz