Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor!

Making sense of AI risks and opportunities

By Andrew Zarkowsky | Head of AI Underwriting, The Hartford

Artificial intelligence (AI) technology is moving quickly, with lots of different terminology emerging and a nearly equal amount of misunderstanding. Some forms of AI, such as computer vision, are already being used effectively to automate certain processes. Others, such as artificial general intelligence, are years from becoming productive tools. As more organizations inside and outside of insurance explore AI to improve their operational efficiency, it is important to keep in mind risks that accompany this remarkable innovation.

Pitfalls in implementing AI

Organizations that want to utilize AI, and those that may already be implementing it, should remain aware of several pitfalls. These include:

• Inadequate commitment of resources. AI cannot be implemented as a part-time project. Successful implementation of AI requires organizations to commit resources on both the technology side as well as the business side that have a deep understanding of those domains. Resources for AI projects cannot operate in silos.

• Technology immaturity. Depending on where an organization is in its technology journey and data management, it may be too soon to attempt to implement AI. An organization’s data infrastructure must be ready for AI. An analogy is trying to hook up a modern appliance to an antiquated electrical system. For the new appliance to work, the system must be configured to accommodate it.

• Roadblocks in achieving goals. Nowadays, it’s common to hear business leaders say, “Let’s use generative AI for that,” in reference to achieving certain goals. What they aren’t grasping is genAI is a tool, and not necessarily the only tool, that organizations can use. Rather than look first to a tool to achieve business goals, a better approach is to gain clarity on the goal and the purpose, then figure out the appropriate tool to use. The saying, “When all you have is a hammer, everything looks like a nail” applies here. Organizations that rush to use generative AI without first understanding whether it’s the right tool are likely to encounter roadblocks to reaching their goals.

• Putting the cart before the horse. Governance, risk management, training and change management are important elements in any AI project to establish trust and reliability. These should be brought in early in the process. A recommended approach is to bring these elements together and focus on learning how to use AI to solve a simple problem for the organization. Solving a major problem with AI is not a good first use of this technology.

• Lack of transparency for users. Developing the user interface for AI should occur from the beginning. In AI, as in other software, a lot goes on behind the scenes. An interface that makes AI processes as transparent as possible, without overwhelming the user, is vital. In heavily regulated industries, including insurance, there are requirements system users must meet. If users do not trust AI, they are not likely to use it, or they may elect to check and verify processes that the organization intended to automate – bypassing the organization’s objective for AI in the first place. Feedback from users is critical in AI development for any organization.

AI in underwriting

A popular use case for AI in insurance is to accelerate the underwriting process. Automating data gathering and analyzing certain information for making simple decisions – for example, is a risk within the insurance company’s appetite – can save human underwriters time. What underwriters do with that time will vary.

At The Hartford, we approach GenAI with disciplined urgency. That means testing, collecting data, and reaching conclusions based on evidence as fast as we can without sacrificing our discipline to be ethical, responsible and reliable. GenAI is just one tool in our AI toolkit, but a powerful one that allows us to tap into types of data that traditional models couldn’t effectively use.

Our focus is using AI responsibly in accordance with our core principles, with a goal of providing better service to our customers. We see AI as useful for certain tasks that are relatively easy for an underwriter to perform or repetitive. Our purpose in letting AI handle such tasks is to let our underwriters spend time focusing on higher-level risk decisions and shift their attention toward the customer.

What is possible with AI tools is changing almost every week. The Hartford is taking a measured approach to AI that fits within our culture. We want to proceed at a pace that’s appropriate for our business and our customers. For us, a focused approach is necessary due to how fast things are evolving with AI.

Mitigating risks

Risk in AI is both a positive and a negative. On one hand, AI can help reduce risks, such as cameras and AI-enhanced videos that can detect conditions that might cause accidents – potholes in roads or spills on restaurant floors. On the other hand, newer risks arise from the use of AI. This can include data exposure and a lack of human oversight, as well as risks associated with AI misuse, AI overselling business capabilities, and a company’s overreliance on AI. In addition, new risks can develop if threat actors use AI to find better ways to enhance their hacking schemes. For these reasons, compliance issues should be front and center for users of AI.

Speed and scale enabled by AI also pose risks as well as opportunities. An older business process might produce errors at a rate of five to 10 times per day. With AI speeding up such tasks, an organization might unknowingly generate those errors 100 times per day. The way to mitigate that kind of risk is to get back to effective governance and testing, and to bring risk management professionals into the AI development as early as possible.

Risk mitigation is not a brand-new discipline, and a completely novel mitigation approach is not necessary for AI. Many decades-old tools that work well in mitigating other process risks can be applied effectively to AI. Organizations that take a risk management toolset and become experienced at using it on AI will manage those new risks better. Working with others who have expertise in AI and risk management is a good way to navigate the risk and opportunities.

For more information, visit https://www.thehartford.com/insights/technology.

Andrew Zarkowsky is Head of AI Underwriting for The Hartford. Before taking on this role, he served as Technology Industry Practice Leader. He has more than two decades of experience in insurance, holding various leadership roles in underwriting. At The Hartford, he focuses on building responsible AI-enhanced underwriting tools to improve efficiencies and effectiveness. He is also responsible for assessing the insurance risk and risk management controls for creators and users of AI.