The Financial Technology Association (FTA) champions the power of technology-centered financial services and advocates for the modernization of financial regulation to support inclusion and responsible innovation. FTA’s “Just the Facts” series aims to inform financial technology policy discussions to safeguard consumers and advance the development of trusted, digital financial markets and services.
FTA members are developing, refining, and deploying artificial intelligence (AI) for a variety of beneficial purposes in the financial services sector.
- AI is the application of data science to create machine algorithms that “can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.”
- It allows financial institutions to quickly innovate and implement effective, better tailored programs, while increasing the reach of their products and services.
- AI and algorithms have been used for decades in financial services, governed by a comprehensive model-focused compliance and regulatory regime.
- Currently, AI is being used to remove embedded bias from credit underwriting, improve some financial offerings and compliance, prevent fraud and illicit finance, and reduce risk.
- Fintech use of AI fosters competition in financial services by enabling the deployment of smarter, more cost-effective services and increasing financial inclusion through robust analysis of traditional data sources.
Most stakeholder concerns about AI use in lending focus on explainability and data.
- Using explainable, transparent models that are not black boxes is the responsible approach to generating predictions that affect credit availability, notably in underwriting. Sufficient explainability will provide users insight into the data variables in order to perform robust fair lending analysis, to generate consumer notices, and to audit the model’s performance against its training objectives.
- Rigorous and comprehensive data integrity, protection, and privacy protocols are critical to ensuring that data inputs into AI models are appropriate and compliant with applicable laws.
Machine Learning is a way for AI to gain ‘knowledge,’ where data scientists use datasets to create and ‘train’ a model, which can help financial firms use more granular data to reduce or eliminate existing bias, improve service while increasing efficiency, and boost consistency in fair lending and other compliance-related outcomes, ultimately empowering consumers to make smarter financial decisions.
- AI/ML can vastly improve on national credit scores (which are algorithms) by building modern algorithms that use more data inputs (the AI) and then training those algorithms (the ML) to look for more inclusive underwriting results, all while maintaining risk within regulatory parameters.
- AI is helping predict cash flow, which gives individuals better insights for financial decisions.
- AI/ML is used to improve consumer protections by detecting and responding to potential attempts to compromise consumers’ online accounts and steal their sensitive financial information.
- AI/ML is also increasingly used by financial technology companies to monitor for financial crime, allowing institutions to better target illicit activity that may be perpetrated by bad actors and submit more accurate and streamlined suspicious activity reports (SAR).
- Some institutions are also using AI for regulatory analysis and operational elements like employee expenses.
Structured datasets can prove helpful for training AI/ML models, while large language models can also incorporate unstructured data.
- Training datasets help the AI/ML model ‘learn’ to accomplish its intended use, and help the developer and deployer validate the accuracy of the model’s outputs (e.g., census tract data and anonymized credit reporting data for AI/ML underwriting models).
- The AI/ML model’s accuracy is measured against its intended use, for example, to meaningfully increase approval rates within a financial institution’s current portfolio without increasing the risk to the firm.
- ‘Large Language Models’ are trained on large bodies of textual content and can be used to process structured and unstructured text in a large variety of financial services use cases (e.g., automating customer service errands and creating conversational interfaces (‘chatbots’)).
- Like traditional financial firms, fintechs are exploring various applications of AI for operational purposes in addition to consumer-facing uses, subject to numerous laws, regulations, and supervisory oversight.
- Regulators domestically (NYDFS) and internationally (MAS & ECB) are also leveraging AI.
AI/ML usage is well regulated in the financial services sector, and these requirements set the standard for how existing law can apply to traditional use cases transformed by algorithms. However, policymakers can do more to advance consumer- and compliance-centric utilization of AI/ML in financial services.
- Federal consumer financial law requires notices to consumers about their data to inform them that their data is being used to make decisions.
- To combat bias and fraud, policymakers should establish clear baseline principles for providers and minimum expectations that those standards must meet.
- Permissible uses of consumer data are outlined in the Fair Credit Reporting Act.
- In accordance with Section 5 of the White House AI Executive Order, policymakers should look at existing law and public datasets, and focus on addressing gaps so that (i) applicable law is clear and (ii) the best datasets are not monopolized for the benefit of a few.
- Bans on discrimination and a basis for establishing anti-bias approaches flow from the Equal Credit Opportunity Act.
- Policymakers should support new credit underwriting/scoring systems that improve on FICO and other legacy approaches with embedded bias.
-
Strong guidelines for credit scoring transparency should be established to create a level playing field for all Americans.
- Banking agency third party risk management guidance requires financial service providers to monitor and manage certain activities of third-party vendors, including models acquired through them.
- Policymakers should clarify that transparent AI/ML is an accepted tool for use in financial services and for boosting regulatory compliance in accordance with Section 7 of the White House AI Executive Order.
- Banking agency model risk management guidance provides a framework for financial institutions and vendors to manage models, primarily through testing and validation.
- Policymakers should increase tech literacy/talent at financial regulatory agencies in accordance with recent GAO recommendations.