 By Matthew Edwards, Senior Director and Innovation Lead at WTW; and Arlen Galicia Carreon, Associate Director at WTW.
AI for code – the next big milestone
The use of generative AI for coding for in-house applications is set to be the next big thing in 2024 as the industry realises just how powerful the latest models have become and insurers find ways to leverage this power. In a recent conversation, a non-executive director in a major UK insurance firm revealed that they had already started using generative AI for a coding project to translate all the code from the insurer’s entire legacy box of business into their preferred code to sit more efficiently with their newer main block of business.
When looking at exactly how these technologies can positively impact our day-to-day work, the writing of computer code is a prime example of a core application of AI. For example, an AI coding system can help generate and test code, as well as assist in the debug process which many developers struggle with. AI can also significantly help to improve documentation and adherence to coding best practice.
AI technologies can also facilitate code translation, such as transforming an Excel macro file into an open-source code like Python or R, with the endgame of fitting such applications into a better governed process. There are many other applications of generative AI that can help the insurance industry, such as report drafting, checking the consistency of reports in large groups or compliance with group or professional standards, and process automation that requires collation and large numbers of documents to be inspected.
Insurance firms are also undertaking competitions internally to see who can come up with the best generative AI use case, such as feeding generative AI an insurer’s complete collection of training and underwriting manuals to create an expert ‘Bot’. This approach also benefits from avoiding the risk of any external interaction, which is sensible for insurers in 2024 that are considering how best to use generative AI, while a better understanding and a level of control are still being established.
AI regulation on the rise
The opportunities of AI do not come without risks, which means implementing AI must be approached with care. As AI becomes progressively more integrated into insurance industry practices, regulatory oversight is also on the rise. This means insurers need to make sure that their AI practices comply with relevant regulations.
With such a heavy reliance on data, protecting data privacy and maintaining ethical standards are crucial. For this reason, insurers will need to comply with data protection regulations and handle personal or sensitive data ethically when using AI.
There is also the risk of bias unfairness. AI models can unintentionally learn and produce biases presented in the training data, leading to unfair outcomes. As a result, a continuous monitoring for bias is essential, alongside a commitment for transparency and fairness in their AI applications.
A key question for regulators will be the extent to which their focus is on the internal use of AI by an insurer, as opposed to concentrating on the company’s actual outputs generated by AI. With the main focus of regulators to date having been on the outputs (for instance, whether premiums are fair and non-discriminatory), the hope shared by many insurers is that this approach will persist.
A further problem arises with transparency. All model users, stakeholders and regulators ideally require their models to be transparent. But this is not possible with generative AI, which is typically based around neural networks with a hundred or more labyrinthine layers, each containing thousands of ‘nodes’ (in effect, robotic neurons). So how can we learn to cope without transparency? Alternative criteria will need to be defined to allow use while retaining confidence in that use.
The AI takeover - redefining insurance
All too often, the insurance industry approaches risk from a one-sided perspective, only seeing the negative side. While this is a natural human instinct and typical of chief risk officers concerned with everything that could possibly go wrong, real-world risks tend to be two-tailed. That is to say, insurers also need to think about the commercial risks of being slow to harness the powers that generative AI offers and hence being left behind.
Looking ahead, the insurance industry is likely to accelerate the pace at which AI and human expertise are integrated. Insurers that invest in the necessary resources and capabilities to ensure the benefits of AI are effectively harnessed, while being mindful of its limitations and potential challenges, will be best equipped to thrive in this new era of insurance innovation.
Generative AI will be profoundly transformative and far more so than analytics and machine learning were predicted to be 10 years ago. Until very recently, industry leaders were sceptical as to how such tools could safely add value to their business. Given the record speed at which these tools are evolving, coupled with an increasing awareness of the technology’s scope and transformative potential, we should be flipping the default question from ‘show me how generative AI can help in this part of the value chain’ to ‘explain to me why you’re not using generative AI here’.
|
