BMA: supervision of AI/ML systems to be risk-based
More than half of insurance groups and the broader commercial market plan on using artificial intelligence/machine learning within the next five years, the Bermuda Monetary Authority has reported.
The information is included in the Bermuda Insurance Sector Artificial Intelligence and Machine Learning Survey – 2022 Report.
In the survey, conducted late last year, insurance groups reported that 68 per cent were using AI/ML systems. The BMA said this outcome was expected because of the scale of business operating internationally.
Nevertheless, it said, the opposite was true for small commercial insurers. Furthermore, more than half of the large commercial companies did not currently use AI/ML technologies.
Overall, 38 per cent of respondents currently use AI and ML systems in their operations and only a limited number of the long-term and property and casualty insurers presently use AI/ML systems.
Among the 62 per cent of respondents that did not use AI/ML technologies, 23 per cent indicated that they planned to adopt AI/ML in the next five years or less.
The BMA said: “These survey results suggest that 54 per cent of insurance groups and the broader commercial market plan on using AI/ML within the next five years, allowing the authority and the Bermuda insurance market sufficient time to develop a creditable and fit-for-purpose AI/ML framework.”
The BMA said the top challenges and obstacles preventing the adoption and usage of AI/ML systems included:
• AI/ML systems not being critical to current business offerings
• Presently, no viable business case
• Lack of requisite skills and expertise to implement these technologies
• Limited budget
The survey found that the areas of concern insurers had when considering adopting AI/ML systems included explainability, auditability, modelling challenges, system security, transparency and consistency of outputs.
Among the use cases, cybersecurity threat detection topped the list of applications for companies using AI/ML models.
The BMA said 66 per cent of respondents who use AI/ML had no written digital strategy. Nevertheless, the majority of respondents with a deliberate digital strategy for AI and ML had the strategy set and written at the board level (73 per cent).
The report said: “An encouraging insight from a governance perspective is that the current controls for AI/ML digital strategies are set at all three lines of defence, including the operational and enterprise risk management levels and compliance function/internal audit.”
As the degree of market adoption of AI/ML systems varies for the Bermuda market given the various sub sectors – insurance groups, large commercial, small commercial, captives, special purpose insurers, intermediaries and innovative companies – the BMA said it must adopt a risk-based approach to regulating and supervising AI/ML systems.
Each sub sector is expected to have varying levels of adoption and impact on their respective risk profiles; therefore, the BMA is not looking to implement a one-size-fits-all approach.
The BMA said typical AI/ML capabilities included speech, image and video recognition, natural language processing, conversational agent capacity (eg chatbots), predictive modelling, augmented creativity, smart automation, advanced simulation, and complex analytics and predictions.