Other parts of this series:
Not all that long ago, the most relevant question in the world of artificial intelligence (AI) was if this technology would make a significant impact on society.
We are far past that point. As the 2018 Technology Vision for Insurance found, the question now is not if but when.
The Tech Vision survey revealed four out of five insurance executives believe that in the next two years, AI will work next to humans in their organization as co-workers, collaborators and trusted advisors.
Such partnerships could have considerable operational benefits.
An Accenture client in life and health insurance, for example, expects to reduce handling times for certain claims to around 100 days to less than five seconds using machine learning, text analytics and optical character recognition. Results like that pique the interest of many life insurers.
Yet AI’s combination of capability and autonomy means it can’t be regarded as a simple software tool. AIs are making choices with profound consequences for customers, employees, and organizations. Like humans, AIs should “act” responsibly, explain their decisions, and work well with others.
Without conscientious shaping, AI can do significant damage to an insurer’s customers, employees, or brand.
For instance, a BBC investigation of the car insurance industry in the UK compiled price quotes to five leading AI-driven price comparison websites. The prices were pulled twice: once using the name of a white BBC producer, and again using the common British name “Muhammad Khan.” All five sites returned higher prices the second time. While the Association of British Auto Insurers has denied that the way its premiums are quoted is discriminatory, the controversy highlights the importance of transparency in AI decisions.
The Tech Vision survey reveals broad consensus around this conclusion. Ninety percent of surveyed insurance leaders said that it’s important for employees and customers to understand the general principles used by their organizations to make AI-based decisions. Yet 77 percent also agree that AI’s capabilities are advancing faster than their organization’s pace of adoption. In other words, AI’s capabilities are developing more quickly than insurers can make use of them.
This raises the specter of life insurers seeking to use tools they do not fully understand. It’s easy to imagine the risks of that.
The Tech Vision report identifies four key questions life insurers should answer as they plan to incorporate AI into their operations:
- How integral will AI be in your business going forward? If you’re not already on an AI journey, the time to start is now. Pick a single business challenge and assign one or two people to research how others have tackled it. Have the team build a plan for how your enterprise will deploy AI.
- Should AI-based decisions be explainable? If so, which ones? Evaluate each use of AI in your organization for ‘explainability’—that is, how transparent are the decisions the AI system is returning? Evaluate the situations where results are not explainable, but should be, and open discussions on how to make them more explainable.
- Are you ‘raising’ your AI with the right data? If you’re already working with AI, catalog all the data sources used in training and testing your systems. Regularly audit these data sources for bias. Similarly, are there bias conditions that the test data does not account for? Act to minimize bias in training data—consider synthetic data creation as a mechanism to achieve this.
- Do you want to go it alone? Early entrants in new industries often enjoy outsized advantages over latecomers. But participating in standards bodies is also an effective strategy to gain an early advantage. Consider joining an AI-focused consortium (e.g, Open AI, Partnership on AI to Benefit People) to get a head start.
AI, if deployed correctly, offers tremendous benefits for life insurers. Come back next week for a look at another technology with similar potential: extended reality.
In the meantime, the full Tech Vision for Insurance 2018 report is available here.
