AI (Artificial Intelligence) is not here an may never be. If your software salesperson, and that is all they are, starts spouting 'AI,' quietly show her out of the building, she is trying to sell you a bill of goods she doesn't understand. On EMRandHIPAA this morning, Anne Zieger posted "Are Healthcare Data Streams Rich Enough To Support AI?" This is a little akin to asking if we should have a salary cap for Quidditch teams.
The thing that Anne is calling AI is actually statistics. If you want advanced statistics calculated, you call an actuary. Those are some smart individuals, but they aren't capable of making Lt. Cmdr. Data, or the HAL-9000 a reality. In fact, this is not a valid application of statistics, even. Like the fictional study of psychohistory, Asmiov's premise was that 'you can't apply it to the individual meaningfully. ' The same applies with statistics, so the assertion that any amount of health data from an individual will give a system the ability to predict (chronic) illness is spurious. Acute illness is a different thing. The more nails you hammer the more likely you are to hit your thumb. Unless you hammer nails for a living. ...or you have a nail gun. I think I just proved that you can't apply statistics to the individual. So yes, before you yell at me, if you smoke, quit. Smoking is a scourge on society and causes lung cancer and lots of other problems in populations. That doesn't mean you are going to get lung cancer if you smoke.
"What about IBM's Watson?" you ask. IBM won't call it AI even, because it isn't. It is a 'question and answer engine.' You could do something very similar at home with Microsoft's Speech Recognition SDK that called out to Bing to get you a list of answers. Is that really remarkable? Maybe it is pretty cool, but there are lots of search engines and lots of speech recognition software our there. Sure, that's not all Watson does, but we didn't spend hundreds of man years and millions of dollars to get similar functionality either.
"I hear about Machine Learning! That has to be AI!" you cry. Well, not so much. Here at Sentia Health, we created a little algorithm that told a doctor what his or her top 25 diagnoses were for the past six weeks. This is just a database query, but the machine 'learned' what the doctors were doing at any given time. This allowed the influenza diagnosis to rise to the top of the list during flu season so the doctors could document all flu patients the same, allowing one click documentation in the Electronic Medical Records system (EMR). It appears that the machine learns, and if you only take the top answer, you will get a different one at different times. We've been doing that for decades and Larry Ellison gave the masses that ability in 1979. So no, again, nothing new here. Nothing to see. These are not the 'droids we are looking for. Move along.
So the software salesmen and saleswomen of the world are trying to sell us a bill of goods. They have co-opted terms they don't understand for the purpose of parting you, dear reader, from your money. This is not fairly harmless marketing like calling blue 'cerulean', red 'scarlet' or yellow 'sunshine.' This is misleading you into thinking your IT provider has done something amazing, something akin to GPS, something you can't get anywhere else, something they haven't done.
Always remember: Beware of Greeks bearing gifts, don't take any wooden nickels, buy low and sell high, and above all follow the money. In this case, the money will lead you back to a salesperson who does not have your best interests at heart.
No comments:
Post a Comment