We have created Frequently Asked Questions (FAQs) about the use of AI in financial services to create a resource for financial services stakeholders. These are designed for broad, non-technical audiences and share our insights on how AI is being used in financial services and to introduce key technological, market, and policy issues related to those applications.
We plan to issue additional FAQs periodically to enrich and extend the issues considered here.
AI FAQs
Key Concepts
Explainability in Credit Underwriting
Federated Machine Learning in Anti-Financial Crime Processes
The Data Science of Explainability
AI FAQs: Key Concepts
The first edition of our AI FAQs addresses a range of introductory questions about AI and machine learning.
- What are artificial intelligence (AI) and machine learning?
- How different are AI and machine learning from other forms of predictive modelling?
- How can we evaluate a specific use of AI and machine learning to understand relevant differences when compared to incumbent models?
- How are AI models different from other common forms of automated prediction?
- How are AI models and machine learning models developed?
- How do AI and machine learning models change after deployment?
- What are supervised learning, unsupervised learning and reinforcement learning?
- What is deep learning?
- How are AI and machine learning being used in financial services?
- What are the key policy debates about using AI in financial services?
- What forms of AI and machine learning are most commonly used in financial services? How do they work?
- What is the basis for believing that machine learning could improve credit underwriting?
AI FAQs: Explainability in Credit Underwriting
Our second edition considers more deeply issues and debates about model transparency and explainability and consider more deeply the implications of using machine learning for credit underwriting.
- What is model transparency? Why do we need it?
- What is model interpretability?
- What is model explainability?
- Why is model transparency especially important in the context of AI and machine learning models?
- What techniques can make machine learning models more transparent?
- How are post hoc explainability techniques being used in practice?
- What are global and local model explanations?
- How can machine learning be used in credit underwriting?
- Why is model transparency especially important in the context of machine learning models used for credit underwriting?
- Who needs information about how a credit underwriting model works?
- What potential risks are important to consider when lenders replace incumbent underwriting models with machine learning?
- What legal and regulatory frameworks apply to the use of machine learning credit underwriting models?
- What prudential frameworks apply to the use of machine learning credit underwriting models?
- What kind of statistical biases are most important with respect to underwriting models?
- What consumer protection frameworks apply to the use of machine learning credit underwriting models?
- What are the sources of discrimination or unfairness in underwriting models and other predictive models?
- What are the core concerns about adopting machine learning underwriting models from the perspective of fair lending and financial inclusion?
- How can we measure the fairness of a model?
AI FAQs: Federated Machine Learning in Anti-Financial Crime Processes
We have prepared an edition of our AI FAQs to consider the technological, market, and policy implications of using federated machine learning to improve risk identification across BSA/AML disciplines, including in customer onboarding where it may facilitate more accurate and inclusive customer due diligence.
- What is federated machine learning? How is it different from other forms of machine learning?
- How are federated models developed?
- What are the key features of current approaches to financial crimes compliance?
- What are the potential benefits of using federated machine learning to identify risks related to financial crimes?
- What are the potential confidentiality and privacy implications of using federated machine learning in anti-financial crime processes?
- What is the basis for believing that using federated machine learning to identify illicit financial activity has the potential to expand access to the financial system?
- What aspects of using federated learning to improve financial crimes compliance processes warrant further research?
AI FAQs: The Data Science of Explainability
This edition of our AI FAQs focuses on emerging techniques to explain complex models and builds on prior FAQs that covered the use of AI in financial services and the importance of model transparency and explainability in the context of machine learning credit underwriting models.
- How do current models for credit underwriting work? In what ways are machine learning models different? Are they necessarily more complex?
- What kinds of machine learning models are most relevant to credit underwriting?
- What factors can affect a model developer’s choice of machine learning algorithms?
- What are latent features? How do they affect model transparency?
- How can model developers improve the transparency of machine learning models?
- How can model developers build inherently interpretable machine learning models?
- What kinds of post hoc explainability techniques can be used to improve the model transparency?
- What characteristics differentiate post hoc explainability techniques?
- How can the capabilities and performance of explainability techniques be evaluated?
Related Publications
Machine Learning Explainability & Fairness: Insights from Consumer Lending
(April 2022)
This empirical white paper assesses the capabilities and limitations of available model diagnostic tools that are designed to help lenders manage machine learning underwriting models. It focuses on the performance of these tools in producing information relevant to adverse action and fair lending requirements.
Read MoreThe Use of Machine Learning for Credit Underwriting: Market & Data Science Context (Sept. 2021)
This report surveys market practice with respect to the use of machine learning underwriting models and provides an overview of the current questions, debates, and regulatory frameworks that are shaping adoption and use.
Read More