Building Trustworthy AI

Context

1. Most FIIs are still working their way through fully integrating Gen 1 AI tools; some advanced Gen 5 tools like Large Action Models are already being tested by the larger banks

2. Regulators, supervisors are bringing about new laws, rules and standards for use of AI in FIIs e.g., NIST, EU AI Act, Rule 1033 etc

Knowledge Takeaways:

1. No matter what technology comes around or goes, Customer Protection will remain key for bankers and FinTechs in managing Digital Trust.

2. Trustworthy AI methods and tools are key to satisfying customers, vendors, Boards and investors

Action Steps

1. Digital Trust framework

2. Mapping between AI tools to known trust gaps

Simple taxonomy of AI tools

As AI technologies proliferate the landscape, its worth having a rough sketch of how sophisticated these tools are. Below is a rough taxonomy; notwithstanding the degree of overlap in these technologies. For example. While chatbots are generally Gen 1 tools, there are sophisticated variants of chatbots that are piped to LLM algorithms through APIs.

• Gen 1 was chatbots and automation

• Gen 2 was use of AI techniques in analytics and possibly risk management.

• Gen 3 is the use of LLM for middle and front office tasks

• Gen 4 will be Agent Based Models

• Gen 5 will be LAM models

New Legislation/Regulation coming about

• National Institute of Standard (NIST) released an AI Risk Management Framework (RMF) which the different regulatory bodies (like CFPB) will channel it through their own regulatory and supervisory programs.

• The EU AI Act in May 2024, approved an Act to define the AI ecosystem; assign risk categories and establishes norms within each category for cataloguing and managing risks.

• The Consumer Financial Protection Bureau's (CFPB) "Open Banking" rule, also known as the "Personal Financial Data Rights" rule, gives consumers more control over their financial data. The rule is based on Section 1033 of the Dodd-Frank Wall Street Reform and Consumer Protection Act. As of writing this note, the new administration of President Trump has not clarified the future role of CFPB.

Mapping AI tools to potential Trust Gaps

Its nigh impossible to outline every single use case and technology choice, for most regional banks and FinTechs, the following short table is a good starting point (v1.0) to get a grasp of the core trust issues customers face. Depending on the specific industry and use cases, we can always iterate to a better taxonomy of usecases, technologies and trust issues.