Former ARPA-H Director’s Startup Takes on AI’s ‘Dumb Problems’ in Biotech

A new startup founded by a former director of the Advanced Research Projects Agency for Health (ARPA-H) is pivoting away from flashy, complex AI models to solve what he calls the field’s “dumb problems” in biology and medicine. The venture, highlighted in this edition of STAT’s AI Prognosis, focuses on the unglamorous but critical bottlenecks where artificial intelligence meets bio. These include fixing broken data pipelines, correcting flawed lab workflows, and automating tedious manual tasks that routinely stall research. The core issue, according to the founder, is that many AI tools in healthcare fail not because the algorithms are weak, but because the underlying data is messy or poorly organized. His approach targets these basic, foundational errors rather than chasing high-profile breakthroughs. In a separate development covered in the same report, new accuracy data on AI scribes—software that transcribes doctor-patient visits—shows these tools are improving but still require close human oversight to avoid clinical errors. The startup’s strategy signals a broader industry shift: instead of asking AI to solve impossible problems, investors and researchers are now paying for systems that simply do the boring, repetitive work correctly.