FDA can help solve clinical AI reliability issues

As 2025 begins, we are living in a Groundhog Day, where distrust of clinical AI is widespread and the consequences can be dire.

About two years ago, one of us, Enes, co-published an original study that showed something was wrong. The data that the FDA publicly shares after approving a clinical AI tool shows little evidence that it is effective. Recently, another review of the same published overview revealed similar results in a deeper dataset. The STAT research report, along with academic analysis, found that commercially available clinical AI tools actually perform poorly despite being brought to market.

Unsurprisingly, in this context, polls reveal that clinicians recognize the potential of AI but are cautious about its use. The reality is that if the status quo persists for too long, we risk entering an AI winter where innovation stagnates and patients suffer. If companies are unable to convert innovation into paying customers, venture capital funding will decline.