Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Responsibility in a Multi-Value Strategic Setting

Timothy Parker, Umberto Grandi, Emiliano Lorini

TL;DR

A model for responsibility attribution in a multi-agent, multi-value setting is presented and it is shown that non-dominated regret-minimising strategies reliably minimise an agent's expected degree of responsibility.

Abstract

Responsibility is a key notion in multi-agent systems and in creating safe, reliable and ethical AI. However, most previous work on responsibility has only considered responsibility for single outcomes. In this paper we present a model for responsibility attribution in a multi-agent, multi-value setting. We also expand our model to cover responsibility anticipation, demonstrating how considerations of responsibility can help an agent to select strategies that are in line with its values. In particular we show that non-dominated regret-minimising strategies reliably minimise an agent's expected degree of responsibility.

Responsibility in a Multi-Value Strategic Setting

TL;DR

A model for responsibility attribution in a multi-agent, multi-value setting is presented and it is shown that non-dominated regret-minimising strategies reliably minimise an agent's expected degree of responsibility.

Abstract

Responsibility is a key notion in multi-agent systems and in creating safe, reliable and ethical AI. However, most previous work on responsibility has only considered responsibility for single outcomes. In this paper we present a model for responsibility attribution in a multi-agent, multi-value setting. We also expand our model to cover responsibility anticipation, demonstrating how considerations of responsibility can help an agent to select strategies that are in line with its values. In particular we show that non-dominated regret-minimising strategies reliably minimise an agent's expected degree of responsibility.

Paper Structure

This paper contains 12 sections, 8 theorems, 4 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Model
  4. Values, Goals and Moral Action Systems
  5. Responsibility Attribution
  6. Multi-Value Responsibility
  7. Excuses
  8. Anticipating Regret and Responsibility
  9. Regret Anticipation
  10. Responsibility Anticipation
  11. Conclusion and Future Work

Key Result

theorem 1

Passive responsibility satisfies consistency and completeness.

Figures (1)

  • Figure 1: A visual summary of the main results of this chapter. The two circles indicate that regret-minimising strategies are exactly passive resbonsibility-minimising strategies (Theorem \ref{['thm:passiveequivalence']}) and that non-dominated strategies are exactly inexcusable passive responsibility-minimising strategies (Theorem \ref{['thm:inexcusableequivalence']}). The intersection of these two circles is always non-empty (Theorem \ref{['thm:alwaysoverlap']}).

Theorems & Definitions (29)

  • definition 1: Multiagent Transition System
  • definition 2: Moral Action System
  • definition 3: Weak Dominance
  • definition 4: Non-Dominated
  • definition 5: Liability
  • definition 6: Responsibility via Strategy
  • definition 7: Passive Responsibility
  • theorem 1
  • proof
  • definition 8: Weak Excuse
  • ...and 19 more