As I attend AINextCon Seattle this week and listen to the speakers from organizations such as Microsoft, Didi, Pinterest, or LinkedIn one theme comes out all the time: the more data they have from more users, the better the results.
Now let me be clear here. When I talk about more data from more users, I mean a lot more users and a lot more data than we are thinking about in the common clinic setting: millions of users, billions of data points, and tens of billions of relationships.
For example, Pinterest, (which I will loosely describe as a visual idea search engine, rather than a social media site) has 200 Million users per month, 100 billion pins and 2 billion boards.
The Pinterest AI software created and maintained by the 450 engineers then generates 10 billion pin recommendations a day ( filtered down from a data set of 100 billion possible recommendation a day), all done from the perspective of real time to the user.
Now, what makes this work isn’t just the talent of the engineers, but your decision to grant free access to your information which you provide on every engagement. You have done this once, upfront when you joined their site and agreed to their terms, and thats pretty much the end of it.
This makes me reflect on the current state of healthcare’s electronic health records (EHRs) and out privacy requirements, called HIPPA.
Back before medical school, i worked in the defense industry on several classified programs, and i can say without a doubt that the amount of time, money and energy a healthcare organization spends protecting little Johnny’s strep test results (or even the fact that he has a sore throat) meets or exceeds what defense industries do to protect out nations secrets.
The bureaucratic challenge of privacy within healthcare is the hurdle which the Amazons, Googles, and Microsofts of the world need to address, not the size and speed of their data sets.
Consider this scenario.
An 18 year old from University of Wisconsin comes home to Seattle for Christmas in the peak of flu season, and goes to the doctor with a fever, chills, and headache. The doctor seeing hime will be strongly considering the flu, especially since he has already seen 8 cases that day, until an alert pops on his EHR screen telling him or her to consider meningitis, because there is a case of confirmed meningitis from West Bend Wisconsin ( the home of University of Wisconsin) and 2 of his college friends (found out via status updates via Facebook). are in the ER with similar symptoms, one in Utah and one back in Virginia.
The AI challenge for this isn’t much harder than matching wedding dresses on Pinterest; the AI engineers and technology is there to do it.
What isn’t there right now is a critical mass of people sharing freely their health data because of privacy concerns.
Now, should we do this? That’s a separate, but very real question. (Most people will say no, until it’s their son who dies of meningitis. ).
But let’s not kid ourselves. We could do it.