Engineering Trustworthy AI: A Developer Guide for Empirical Risk Minimization

TL;DR

This paper discusses how key requirements for trustworthy AI can be translated into design choices for the components of ERM, to provide actionable guidance for building AI systems that meet emerging standards for trustworthiness of AI.

Abstract

AI systems increasingly shape critical decisions across personal and societal domains. While empirical risk minimization (ERM) drives much of the AI success, it typically prioritizes accuracy over trustworthiness, often resulting in biases, opacity, and other adverse effects. This paper discusses how key requirements for trustworthy AI can be translated into design choices for the components of ERM. We hope to provide actionable guidance for building AI systems that meet emerging standards for trustworthiness of AI.

Figures (12)

• Figure 1: Trustworthy AI adds new design criteria for based methods. Besides small computational complexity and data requirements, these methods must be sufficiently explainable, privacy-friendly, fair and robust.
• Figure 2: uses the average incurred on a to approximate the (or expected ).
• Figure 3: We can diagnose a ML method by comparing its with its . Ideally both are on the same level as a (or benchmark error level).
• Figure 4: Equivalence between and penalization.
• Figure 5: -based methods are defined by design choices for , and . This paper discusses design choices that facilitate KRs for trustworthy AI.

Theorems & Definitions (1)

• Definition 1