In this presentation Professor Stephen Roberts explore the use, abuse and pitfalls of using GenAI in insurance.
The impact of AI on employment and economic structures
In the presentation, Stephen highlights the transformative potential of AI across various job sectors. Analysts predict that AI will impact 40-60% of jobs globally, not by eliminating them but by automating specific tasks within those roles. This mirrors historical technological advancements, such as the changes in medical professions over the past century. Advanced economies will face the highest exposure to AI due to their technological readiness, but this also introduces a risk of increased technological fragility. He emphasises the need for checks and balances to manage these changes responsibly. The uneven distribution of AI capabilities across the world exacerbates these challenges, with significant disparities between countries.
Skepticism and understanding of AI in the insurance sector
Despite the widespread adoption of AI in finance and insurance, there remains a significant level of skepticism and concern among practitioners. Around a quarter of early adopters worry about their lack of understanding of AI's inner workings. This is particularly pertinent as AI systems grow more complex, often exceeding human comprehension. The analogy of negotiating with an alien underscores the unfamiliarity and potential risks associated with relying on AI systems. The speaker stresses the importance of transparency and a deep understanding of AI models to ensure they operate as intended and to mitigate unforeseen consequences.
The role of synthetic data in AI development
The increasing use of synthetic data in training AI models is a critical development in the insurance industry. Soon it is estimated that 60% of data used for AI training will be synthetic. This approach is necessary to address the scarcity of real-world data, particularly for rare events or large loss scenarios. However, synthetic data must accurately reflect real-world statistics to be useful. Stephen notes that current challenges include ensuring the plausibility and reliability of synthetic data, which must be derived from realistic foundations to be effective. This highlights the ongoing need for rigorous validation and testing of AI systems trained on synthetic data.
Generative AI: opportunities and risks
Generative AI, while powerful, presents both exciting opportunities and significant risks. Stephen describes generative AI models as "stochastic parrots," capable of producing plausible outputs based on their training data but prone to errors when extrapolating beyond known boundaries. Examples such as misidentified images and incorrect mathematical calculations illustrate these limitations. The potential for AI to generate convincing but incorrect information poses risks, particularly in critical applications like insurance underwriting. The speaker advocates for the use of constrained, narrow AI models with trusted data sources to mitigate these risks. Additionally, the speaker draws parallels to past technological skepticism, suggesting that while generative AI can enhance productivity, it must be approached with caution and critical oversight.
Embracing AI with caution
The presentation concludes with a call for a balanced approach to AI adoption. Professor Roberts acknowledges the transformative potential of AI but urges rigorous evaluation and skepticism. By embracing AI cautiously and critically, businesses can leverage its benefits while minimizing risks. This approach echoes the advice of Dennis Gabor: "We cannot predict the future, but we can engineer it." Be passionate yet critical of AI, ensuring that its integration into business processes is both effective and responsible. This philosophy underscores the need for continuous measurement, validation, and improvement in AI applications, particularly in complex and high-stakes industries like insurance.
Sign Up to TINsights
Where we share our latest blogs, industry reports and insights
