Artificial intelligence?

I grew up during the beginning of the digital age.  I remember when the first personal computers came out.  I was always fascinated by a computer game called ELIZA, an early natural language processing game in which the computer played the role of a psychotherapist and the user would ask for help with his or her problems of the day.  Here is an example of how a conversation might go:

Little did I know that ELIZA was one of the early forerunners of artificial intelligence.  I was reminded of ELIZA (or at least I felt like I was talking to it again) just the other day when I was calling my health insurance provider (name withheld to protect the innocent) in order to submit a recent claim.  My provider coverage changed slightly at the beginning of the calendar year, and I had a claim to submit for service that occurred during the second week of December.  In this particular case, I paid out of pocket with the expectation that I would submit a claim later.

The website and app (which I usually use for these kinds of transactions) weren't able to assist me, so I called the service line directly.  As is usually the case, I was talking to an artificially intelligent virtual assistant.  The conversation went something like this (the first several questions were to verify my identity, which I am leaving out for obvious reasons):

Virtual Assistant: How can I help you today?

Me: I need to submit a claim.

Virtual Assistant: Okay.  You need to submit a claim.  Is that correct?

Me: Yes.

Virtual Assistant: What was the date of your service?

Me: December 6, 2024.

Virtual Assistant: Did you say "December 6, 2024"?

Me: Yes.

Virtual Assistant: One moment please.  I'm sorry, there is not a claim for that service.

Me: Correct.  That is why I need to submit the claim.

Virtual Assistant: I'm sorry.  I didn't get that.  How can I help you today?

Me: I need to submit a new claim.

Virtual Assistant: Okay.  You need to submit a claim.  Is that correct?

Me: Yes.

Virtual Assistant: What was the date of your service?

Me: December 6, 2024.

Virtual Assistant: Did you say "December 6, 2024"?

Me: Yes.

Virtual Assistant: One moment please.  I'm sorry, there is not a claim for that service.

Me: Can I speak to customer service?

Virtual Assistant: I can help you.  Would you like me to repeat the menu?

Me: Hang up.

Now, I realize that my issues weren't necessarily straightforward.  However, there was no easy way to request to speak with a living, breathing human being who would know how to help me.  I ended up going through a member of our HR team to help me submit my claim.  

I know that artificial intelligence has come a very long way since ELIZA, and I know that my health insurance provider's virtual assistant wasn't a true representation of what artificial intelligence can really do.  However, my experience did raise a few questions in my mind that inspired me to do a little research.  I read about something known as the Winograd schema challenge, a test of artificial intelligence developed by Hector Levesque in 2011 (and named after fellow computer scientist Terry Winograd), supposedly to improve upon the classic Turing test, which was originally developed by the computer scientist Alan Turing to test whether a computer could exhibit intelligent behavior equivalent to a human.   These tests are essentially questions that computers are terrible at answering, and here is a great example:

A large ball crashed through a table because it was made of steel.  What was made of steel?

The question would be easy for a human to answer (the ball crashed through the table because the ball was made of steel), but apparently artificial intelligence would have difficulty answering the question.  I guess we need to add one more question to the Winograd schema challenge - "What was the date of your service?"  I am reminded of an essay that Nicholas Carr wrote on October 5, 2012 entitled "The ennui of the intelligent machine" (I can't find it online, but it included in his book, Utopia is Creepy, a collection of his essays, articles, and blog posts).  Carr was writing about neural networks and how Google was using artificial intelligence for some of its speech recognition and image recognition tasks.  He quoted one of Google's engineers, Jeff Dean, who said, "We are seeing better than human-level performance in some visual tasks" (specifically identifying house numbers on satellite images for Google's Street View application) and how Dean thought that the real Turing test has less to do with what we think of intelligence and perhaps more to a computer's inability to experience boredom.  As you can imagine, some of these tasks are fairly mundane.  

Dean went on to say, "It's probably that the task is not very exciting, and a computer never gets tired."  Carr added that it takes real intelligence to get bored.  He added, "Forget the Turing Test.  We'll know that computers are really smart when computers start getting bored.  If you assign a computer an overwhelmingly tedious task like spotting house numbers in video images, and then you come back a couple of hours later to find the computer checking its Facebook feed or surfing porn, you'll know that artificial intelligence has truly arrived."

What is clear is that artificial intelligence is most definitely here to change how we work and how we live.  As I talked about in a recent post (see "Six tests (plus one challenge) for physicians and health care leaders..."), artificial intelligence will be integral to just about every facet of health care in the future.  Health care leaders (and physicians) will need to embrace it or be left behind.