by Angus MacCaull

For coming on two years, I’ve been periodically rereading the same article about AI. “Insurance 2030—The impact of AI on the future of insurance” by three partners at McKinsey & Company details “a seismic, tech-driven shift” happening in our industry.

Here’s a one-sentence summary. Data coming in volumes we’ve never seen before from cheap sensors can be processed by AI to provide realtime risk assessment, pricing and adjusting.

There’s a lot to unpack in the McKinsey article. There are big assumptions about how sensors and risk work in the real world. There are also big assumptions about how business and government create value for people. I’ll write at length about these assumptions another time. For now, I’d like to look at the AI part. What is AI? How does it work?

Full disclosure: I’m not an AI expert. And even AI experts don’t try to cover their complex field in a few short paragraphs. Here’s my understanding of one aspect of artificial neural networks, which are sets of algorithms driving the current interest (and funding) in AI.

 

Suit with an artificial brain

 

I carry around a quote by marketing professor Edward McQuarrie of the Leavey School of Business: “Any major decision proceeds through a series of smaller subdecisions.” Artificial neural networks follow this principle. They process data by categorizing it in successive layers.

Each layer represents a pattern. For example, one layer in a network that does facial recognition in pictures might note all the images that have a circle—a head. A second layer might note all the circles that have three smaller circles inside—eyes and a mouth.

After a few more layers, the network decides which pictures have a face in them and which don’t. In order to make this decision, it made a bunch of smaller subdecisions about the features of faces.

We do this all the time ourselves. We look for patterns; then we look for patterns of patterns. It’s an activity that’s fundamental to insurance, whether in the back office or on the front line. But contemporary AI can do it a lot faster than us, with a lot more scope and granularity.