德勤-人工智能与风险管理报告（英文）-2019.4-29页_2mb

报告摘要

Deloitte: AI in Financial Services and Risk Management

Core Content

This document from Deloitte explores the implications of Artificial Intelligence (AI) for financial services (FS) firms, particularly in the context of risk management and regulatory compliance. It outlines the challenges and opportunities associated with AI adoption, emphasizing the need for FS firms to adapt their existing risk management frameworks to accommodate the unique characteristics of AI.

Main Points

1. Executive Summary

AI is increasingly being adopted in financial services, but its implementation faces several challenges.
FS firms are cautious about AI due to concerns over data quality, transparency, and regulatory risks.
Effective risk management is essential for successful AI adoption, not a barrier to it.
The biggest challenge is not new risks but the difficulty in identifying and managing existing risks in unfamiliar ways due to AI's complexity and speed of evolution.

2. A Brief Overview of AI

AI refers to the development of computer systems capable of performing tasks requiring human intelligence.
Key AI techniques include:
- Machine Learning: Systems that improve through exposure to data without explicit programming.
- Deep Learning: A subset of machine learning that excels in tasks like speech and image recognition, often seen as "black boxes" due to complexity.
- Speech Recognition and NLP: Ability to understand and generate human language.
- Visual Recognition: Analyzing images to identify objects, scenes, and activities.

3. Challenges to Widespread AI Adoption in Financial Services

Differing views on AI application: AI use cases vary significantly across sectors.
Data availability and quality: A major obstacle due to reliance on high-quality, large datasets.
Transparency and compliance: AI's complexity and opacity challenge regulatory compliance and explainability, especially under GDPR.
Understanding AI: Key stakeholders need to grasp AI's implications and risks, including how to mitigate them.
Human impact: AI adoption may change job roles, require new talent, and affect staff morale and retention.

4. Embedding AI in Your Risk Management Framework

AI adoption requires a scientific mindset, with trial and error, continuous testing, and a "sandbox" approach.
AI should be integrated into the existing Risk Management Framework (RMF), rather than requiring a completely new system.
The RMF lifecycle includes identify, assess, control, and monitor stages, all of which need to be adapted for AI.

Key Risk Considerations

Risk Category	Sub Category	Key Considerations
Model Risk	Algorithm Risk - Bias	· Inherent bias in data or evolving datasets complicates risk identification.
	Algorithm Risk - Inaccuracy	· Incorrect algorithms or poor data quality can lead to inaccurate outcomes.
	Algorithm Risk - Feedback	· Undetected feedback in learning systems may compromise accuracy.
	Algorithm Risk - Misuse	· Poor understanding of AI limitations may lead to misinterpretation of outputs.
Technology Risk	Information & Cyber Security	· Reliance on open-source components can introduce security vulnerabilities.
	Change Management	· Changes in upstream systems may lead to unforeseen consequences.
	IT Operations	· Legacy systems may not support AI's big data needs.
Regulatory & Compliance Risk	Data Protection	· Evolving AI solutions may increase breach risks under data protection laws.
	Regulatory Compliance	· Difficulty in explaining AI decisions to regulators.
Conduct Risk	Culture	· Perceived ethical and regulatory concerns may hinder AI adoption.
	Product Innovation	· Risk of developing products that don't meet customer needs.
People Risk	Roles & Responsibilities	· Unclear roles and responsibilities may increase risk of errors.
	Recruitment & Skills	· Insufficient in-house skills and over-reliance on a few experts pose risks.
Market Risk	Concentration Risk	· Over-reliance on a few AI vendors increases systemic risk.
Supplier Risk	Liability Allocation	· "Black box" algorithms may obscure liability in case of damages.

Risk Appetite and AI

AI can inherently increase or decrease certain types of risks (e.g., model risk).
Firms must revisit their risk appetite to incorporate AI-specific considerations.
Clear assessment criteria are needed to evaluate AI use cases and their conduct risk implications.

Risk Management Framework Lifecycle

Identify: Understand and monitor risks that AI may introduce, including its evolving nature and impact on organizational culture and human capital.
Assess: Define and embed a risk assessment process to evaluate the level of exposure.
Control: Implement controls to mitigate risks to an acceptable level.
Monitor and Report: Continuously assess the effectiveness of controls and report on residual risk and remediation efforts.

Conclusion

AI is not a new concept, but its application in financial services is still in its early stages.
FS firms need to develop a comprehensive understanding of AI, its risks, and how to integrate it into their risk management practices.
Effective risk management is key to successful AI adoption and innovation.
Regulators are increasingly focused on AI's implications, especially in terms of transparency, fairness, and compliance.

Key Takeaways

AI adoption in FS is complex and requires adaptation of existing risk management practices.
Transparency, data quality, and regulatory alignment are critical.
Firms should consider both the technical and cultural implications of AI.
A scientific mindset and continuous monitoring are essential for managing AI risks.

展开完整摘要