Actuaries Versus Artificial Intelligence: What Do Actuaries Do? What Will They Do?
The room was packed. I’d guess almost 1,000 people came to hear thought leaders James Guszcza, FCAS, of Deloitte and David Ingram, FSA, of Willis Towers Watson talk about data science and behavioral science at the CAS Annual Meeting in Anaheim.
They were talking, at least it sounded to me as I considered it and went back through my notes, about what it means to be an actuary today.
That’s a topic a lot of us think about as our profession seems encroached upon by artificial intelligence (AI).
Once artificial intelligence, whatever it is, (Guszcza noted that the definition of AI is erratically drawn), gets cranking, it will be machines scrubbing, collating, analyzing and concluding — yes, telling us — what we humans should think.
As actuaries we have always assumed that to the greatest brain goes the truth. In actuary versus AI, it is AI that will always win. It is smarter. It is faster. It never sleeps.
Now chess should be the ultimate bummer in the battle of man versus machine. Watson beat Kasparov. Some time ago. By more than a little bit.
But the world of chess, Guszcza pointed out, has moved to a higher plane, above man, yes, but above machines, too.
The best chess in the world today is played not by a man and not by a machine, but by a team — a team of computers and people working together.
It’s called freestyle chess, though I’ve also read reference to it as centaur chess. Computers memorize and categorize thousands and thousands of moves and games. People apply soft skills — hard to describe but abundant and important. Together the computer and the human make a better team than either one on its own.
If you think about it, that larger phenomenon — humans create a tool that outshines them, then harness and leverage it — is as old as invention itself. Humans tamed the equine, and a skilled horseman can outrace Usain Bolt. Speed began to depend on the skill of the trainer and rider, not on the physical prowess of the individual. Cars are faster and more powerful (in horsepower) than horses, so we had to learn to drive.
It is elsewhere too: The well-tended loom outproduces the most facile weaver. The steam drill outlasts John Henry. (The “steel-driving man” folk hero would beat the machine but dies from the stress.)
It is even in the actuarial world. Forty years ago, before the desktop computer, I’m told, actuaries worked in shifts. The early shift calculated estimates. The late shift double-checked the work.
All of those jobs were swept away by the computer, but the number of casualty actuaries keeps growing — from less than 1,000 in 1977 to nearly 8,000 today. We’ve done such a good job of harnessing the machines that demand for the humans is running ahead of supply.
What did we do right?
We created systems that made our work more valuable.
Take loss reserving for an example: We build several models to estimate ultimate losses (chain ladder, Bornhuetter-Ferguson, Cape Cod); learn the strengths and weaknesses of each; and use that knowledge in addition to everything else we know about the claims environment to select an estimate.
We didn’t invent this multimodel approach. Weather forecasters consult multiple models. (It was the European model that forecast Superstorm Sandy’s fateful left turn in 2012.)
But if AI is fast approaching, knowledge of the model may become as outdated as dressage at the Indianapolis Speedway.
And we aren’t the only ones who will have to change. Good doctors, Guszcza said, excel at pattern recognition. Patients present a set of symptoms and doctors diagnose. They are, in their way, like old-fashioned chess players, learning a massive set of symptoms, illnesses and prescriptions and applying that knowledge deftly. But that skill will shrivel in importance in the age of Watson.
