A global review of 25 regulatory reports by the Economist Intelligence Unit (EIU) has identified ways that AI can go bad. Although the report doesn’t present how to avoid these traps, a future report from Mercator Advisory Group will. In the mean time, here is an excerpt from an article covering the EIU’s report:
‘Mr Sharma [Prag Sharma, Citi Innovation Labs senior vice president] said at the root of the risks is the inherent complexity of AI. Some AI models can look at millions or sometimes billions of parameters to reach a decision,’ Mr Sharma said. ‘Such models have a complexity that many organisations, including banks, have never seen before.’
The EIU report has listed key governance challenges and summarised regulatory guidance for banks using AI, including:
- Ethics and fairness: banks must develop AI models that are “ethical by design”. AI use cases and decisions should be monitored and reviewed, and data sources should be regularly evaluated to ensure the data remains representative;
- Explainability and traceability: steps taken to develop AI models must be documented in order to fully explain AI-based decisions to the individuals they impact;
- Data quality: bank-wide data governance standards must be established and applied to ensure data accuracy and integrity, and avoid bias; and
- Skills: banks must ensure the appropriate level of AI expertise across the business so they can build and maintain AI models, as well as oversee these models.
Commenting on the research, EIU editorial director Pete Swabey said: ‘AI is seen as a key competitive differentiator in the sector.’‘Our new study, drawing on the guidance given by regulators around the world, highlights the key governance challenges banks must address if they are to capitalise on the AI opportunity safely and ethically.’ ”
Overview by Tim Sloane, VP, Payments Innovation at Mercator Advisory Group