Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Societal Adaptation to Advanced AI

Jamie Bernardi, Gabriel Mukobi, Hilary Greaves, Lennart Heim, Markus Anderljung

TL;DR

As advanced AI diffuses, capability-modifying governance becomes increasingly infeasible and safeguards are not failproof. The paper proposes a structured adaptation framework that classifies interventions into Avoidance, Defence, and Remedy, applied along a causal chain from development to impact, and demonstrates it with three threat domains: election manipulation, AI-enabled cyberterrorism, and loss of AI control. It introduces a three-step resilience cycle—identify/forecast/assess risks, identify/evaluate adaptive responses, and implement/measure effectiveness—and outlines practical recommendations for governments, industry, and third parties. The work provides a concrete pathway to harness beneficial AI diffusion while mitigating systemic harms, emphasizing proactive risk identification, cross-sector coordination, and sustained investment in societal adaptation.

Abstract

Existing strategies for managing risks from advanced AI systems often focus on affecting what AI systems are developed and how they diffuse. However, this approach becomes less feasible as the number of developers of advanced AI grows, and impedes beneficial use-cases as well as harmful ones. In response, we urge a complementary approach: increasing societal adaptation to advanced AI, that is, reducing the expected negative impacts from a given level of diffusion of a given AI capability. We introduce a conceptual framework which helps identify adaptive interventions that avoid, defend against and remedy potentially harmful uses of AI systems, illustrated with examples in election manipulation, cyberterrorism, and loss of control to AI decision-makers. We discuss a three-step cycle that society can implement to adapt to AI. Increasing society's ability to implement this cycle builds its resilience to advanced AI. We conclude with concrete recommendations for governments, industry, and third-parties.

Societal Adaptation to Advanced AI

TL;DR

As advanced AI diffuses, capability-modifying governance becomes increasingly infeasible and safeguards are not failproof. The paper proposes a structured adaptation framework that classifies interventions into Avoidance, Defence, and Remedy, applied along a causal chain from development to impact, and demonstrates it with three threat domains: election manipulation, AI-enabled cyberterrorism, and loss of AI control. It introduces a three-step resilience cycle—identify/forecast/assess risks, identify/evaluate adaptive responses, and implement/measure effectiveness—and outlines practical recommendations for governments, industry, and third parties. The work provides a concrete pathway to harness beneficial AI diffusion while mitigating systemic harms, emphasizing proactive risk identification, cross-sector coordination, and sustained investment in societal adaptation.

Abstract

Existing strategies for managing risks from advanced AI systems often focus on affecting what AI systems are developed and how they diffuse. However, this approach becomes less feasible as the number of developers of advanced AI grows, and impedes beneficial use-cases as well as harmful ones. In response, we urge a complementary approach: increasing societal adaptation to advanced AI, that is, reducing the expected negative impacts from a given level of diffusion of a given AI capability. We introduce a conceptual framework which helps identify adaptive interventions that avoid, defend against and remedy potentially harmful uses of AI systems, illustrated with examples in election manipulation, cyberterrorism, and loss of control to AI decision-makers. We discuss a three-step cycle that society can implement to adapt to AI. Increasing society's ability to implement this cycle builds its resilience to advanced AI. We conclude with concrete recommendations for governments, industry, and third-parties.
Paper Structure (34 sections, 2 figures, 1 table)

This paper contains 34 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. The Need for Societal AI Adaptation
  3. Capability-Modifying Approaches Will Become Less Effective Over Time
  4. Increased Diffusion Makes Capability-Modifying Interventions Less Feasible
  5. Safeguards Are Not Failsafe
  6. Adaptive Approaches Aid Beneficial Diffusion
  7. Adaptation Can Enable Beneficial Use
  8. Adaptation Can Enable Open Development
  9. A Framework for AI Adaptation
  10. The Causal Chain to Negative Impacts of AI
  11. Interventions to Reduce Negative Impacts
  12. Capability-Modifying Interventions
  13. Adaptation Interventions
  14. Examples of Adapting to AI Risks
  15. Election Manipulation with Generative AI
  16. ...and 19 more sections

Figures (2)

  • Figure 1: A simplified causal pathway to an AI system causing negative impacts and how various types of intervention can reduce them. The focus of this paper is on the latter three interventions: adaptation interventions.
  • Figure 2: The three-step adaptation cycle that must be implemented to successfully adapt to advanced AI. Resilience is society’s capacity to perform this loop.