2018年-德勤全球_AI_and_risk_management_29页_2mb

报告摘要

AI and Risk Management in Financial Services

Core Content

This document explores the integration of Artificial Intelligence (AI) into the Risk Management Framework (RMF) of financial services (FS) firms, emphasizing the need for a robust and adaptive approach to manage AI-related risks. It highlights that while AI offers significant potential for innovation and efficiency, it also introduces new challenges in risk identification, assessment, control, and monitoring.

Main Points

1. AI Overview

AI is a broad field that includes techniques like Machine Learning and Deep Learning, which enable systems to learn from data and make decisions.
AI applications are becoming increasingly popular in areas like speech recognition, natural language processing (NLP), and visual recognition.
AI is not a single technology but a collection of methods that simulate human intelligence, making it both powerful and complex.

2. Challenges to AI Adoption in FS

Differing views on AI application: AI use cases vary by sector, with some areas seeing more adoption than others.
Data availability and quality: AI relies heavily on large, high-quality datasets, which many FS firms lack due to legacy systems and data silos.
Transparency and accountability: The complexity of AI systems, especially those using deep learning, makes it difficult to audit, trace, or explain decisions, which can conflict with regulatory requirements like GDPR.
Human impact: AI adoption may affect organizational culture, talent strategy, and employee morale. It can also lead to job reassignment and skills gaps.

3. Embedding AI in Risk Management Framework

AI does not necessitate entirely new risk management processes, but existing ones must be enhanced to address AI-specific challenges.
The RMF lifecycle includes identify, assess, control, and monitor stages, each requiring adaptation to AI's unique characteristics.
AI solutions can introduce new or evolved risks, such as algorithmic bias, inaccuracy, and misuse, which must be carefully considered.

4. Risk Appetite and AI

A firm's risk appetite must align with its AI adoption strategy.
AI can affect the relative balance of risk components and the tools used to manage them.
Firms need to reassess their risk appetite for specific AI use cases and their broader implications on the organization.

Key Considerations for Each Stage of the RMF Lifecycle

Stage	Key AI Considerations
Identify	- Risks may evolve rapidly due to AI's learning nature. <br> - Reassess AI use cases as they learn and adapt. <br> - Consider both specific and general organizational impacts.
Assess	- Evaluate the level of risk exposure in AI applications. <br> - Account for AI's ability to make decisions based on patterns, not predefined rules.
Control	- Implement controls to reduce risks to an acceptable level. <br> - Ensure AI models align with the firm's risk appetite and values.
Monitor and Report	- Continuously assess the effectiveness of controls. <br> - Report on residual risk and remediation programs to governance bodies.

Example: AI in Insurance Policy Pricing

An AI model for property insurance pricing may use unstructured data to assess risk, including one-off local events.
Algorithmic bias and inference without causation are key risks in such models.
The model's decision drivers may not be clear, and one-off events could be misclassified as permanent risks.
AI models can be used in different contexts, and their risk implications must be understood at each stage of deployment.

Regulator Perspective

Regulators are increasingly aware of AI's benefits but are also concerned about potential risks and unintended consequences.
They are focused on transparency, fairness, and accountability in AI-driven decisions.
Firms must ensure that AI systems are explainable, auditable, and aligned with regulatory expectations.

Conclusion

AI is not a barrier to innovation but a tool that requires careful risk management.
Firms must adapt their RMF to address AI's unique characteristics, including its learning capabilities, complexity, and speed of evolution.
A scientific mindset and sandbox approach are essential for AI adoption, involving all three lines of defense.
Continuous engagement, training, and oversight are necessary to ensure effective AI governance and mitigation of risks.

Key Takeaways

AI adoption in FS is in its early stages and faces barriers such as data quality, transparency, and cultural resistance.
Effective risk management is critical for successful AI integration.
Regulatory expectations are evolving to include AI-specific considerations.
Ongoing reassessment and adaptation of risk practices are necessary to keep pace with AI's development.

Recommendations

Develop clear assessment criteria for AI use cases.
Ensure stakeholder understanding of AI's implications.
Revisit risk appetite statements to incorporate AI-related considerations.
Implement a comprehensive and continuous risk identification and management approach.
Foster a scientific mindset and cross-functional collaboration to support AI innovation and governance.