Bristol-Myers Squibb turns to an AI startup to accelerate cancer research

By Casey Ross

“To be honest, I haven’t seen really good examples where all this data has been converted into actionable evidence,” said Pratik Shah at the Massachusetts Institute of Technology.

He said machine learning is useful for analyzing broad data sets and making high-level generalizations, but its utility diminishes as questions get narrower, and the data become scant and of lower quality.

“The real value of machine learning is for applications in health care without very narrow outcomes, which is what [pharmaceutical] companies usually and unfortunately go after,” Shah said. He added that it is especially important for pharma companies as well as other researchers to be transparent in their use of machine learning and the type and diversity of data they are relying on.

“There should be some accountability for what your algorithm is doing,” Shah said. “Data standards and benchmarking should be established before any actionable evidence is going to come out of this.”

Related Content