News for hackers. Stuff that matters.
Re:“AI promised to revolutionize radiology but so far its fa... by duxup on Monday June 07, 2021 @11:38AM (#27423842)
I recall stories of IBM Watson's failures were focused around how they sold it as just dumping data into the machine and wonders coming out.

Meanwhile actual implementation customers weren't ready / were frustrated with how much prepping the data was required, how time consuming it was, and in a lot of ways how each situation was a sort of research project of its own.

It seems like any successful AI system will require the team working with the data to be experts in the actual data, or in this case experts in radiology ... and take a long time to really find out good outcomes / processes, if there are any to be found.

Add the fact that the medical industry is super iterative / science takes a long time to really figure out ... that's a big job.

Re:“AI promised to revolutionize radiology but so far its fa... by visarga on Monday June 07, 2021 @12:24PM (#27424388)
There's no free ride, ML is data centric, you got to get close and personal with the data and its quality. That means 90% of our time is spent on data prep and evaluations.

Getting to know the weak points of your dataset takes a lot of effort and custom tool building. Speaking from experience.