Data Science and Human Centered Design
At the 2019 NCCI Annual Issues Symposium, Jim Guszcza, US Chief Data Scientist with Deloitte Analytics discussed the promise of artificial intelligence and data analytics holds for the workers’ compensation industry.
The fearful narrative about artificial intelligence are that computers will eventually become smarter than people and will take over the world. There is also fear that AI could eliminate close to half of employment in the United States.
In reality, AI is is allowing computers to do things that humans used to do. Any program that replaces computer function could be considered AI. An example of this is an ATM machine. There are actually more bank tellers now that there were before ATM machines as the need for less tellers in a location allowed banks to open more branches so they could better interact with the community.
One problem with AI is that if you train the computer using past outcomes you could train in past bias. You can also train in bias into the model if you do not adequately consider all potential variables in the training.
An advantage of AI is that it can filter out “noise” which is irrelevant information that can distract from the decision making process.
It is important that future AI design is human centered. AI will provide better outcomes if they are designed to work with human psychology.
AI can also lead to what he referred to as automated stupidity. People become overly reliant on the automation and make bad decisions because of that. An example is people not paying as close attention when driving because their car has lane assist and auto stopping. This technology can actually make drivers less safe if they do not use it correctly.
When it comes to the insurance industry one question is can underwriting be computerized? In a very limited case the answer is yes. Simple underwriting such as personal auto insurance could be underwritten using AI, and in fact it usually is as you can purchase such coverage without human interaction. However, more complex lines of coverage require human judgment. AI can assist this judgment but it does not replace it.
Data analytics alone cannot change outcomes. They can point us in the right direction but they are not the complete solution. In order for analytics to be impactful they must be followed by the correct human judgments, decisions and behavioral change. This is a very important consideration when it comes to the use of predictive analytics in the workers’ compensation claims setting.
“Nudge” is another AI tool that can drive change. Nudge detects unwanted activity and nudges the person to change that behavior. In the workers’ compensation setting, wearable sensors could detect unsafe behaviors and “nudge” the worker to change the behavior thus preventing an injury.