[Paper Review] The Role of Cooperation in Responsible AI Development

Amanda Askell, Miles Brundage+1 more|arXiv (Cornell University)|Jul 10, 2019

Innovation, Sustainability, Human-Machine SystemsSocial Sciences72 references46 citations

TL;DR

The paper argues that competitive pressures can undermine responsible AI development and identifies five factors and four strategies to foster industry cooperation for safer, more socially beneficial AI.

ABSTRACT

In this paper, we argue that competitive pressures could incentivize AI companies to underinvest in ensuring their systems are safe, secure, and have a positive social impact. Ensuring that AI systems are developed responsibly may therefore require preventing and solving collective action problems between companies. We note that there are several key factors that improve the prospects for cooperation in collective action problems. We use this to identify strategies to improve the prospects for industry cooperation on the responsible development of AI.

Motivation & Objective

• Explain why competition can undermine responsible AI development and introduce the concept as a collective action problem.
• Identify factors that improve cooperation among AI developers in pursuing responsible development.
• Discuss potential strategies to overcome collective action problems and promote safer, more socially beneficial AI.

Proposed method

• The authors conceptualize responsible AI development and its costs and benefits.
• They analyze incentives under competition and how first-mover dynamics influence safety investments.
• They define five cooperation-facilitating factors: High Trust, Shared Upside, Low Exposure, Low Advantage, and Shared Downside.
• They propose four actionable strategies: correct harmful misconceptions, collaborate on shared challenges, increase appropriate oversight, and incentivize adherence to ethics and safety standards.

Experimental results

Research questions

• RQ1 How does competitive pressure affect incentives to invest in responsible AI development?
• RQ2 What factors increase the likelihood of cooperation among AI developers in responsible AI development?
• RQ3 What strategies can reduce collective action problems and promote safer, more beneficial AI?
• RQ4 Under what conditions could regulation, liability, or market forces better align incentives with social welfare?

Key findings

• Competitive pressures can create a collective action problem that leads to underinvestment in safety, security, and social impact.
• Five factors improve cooperation prospects: High Trust, Shared Upside, Low Exposure, Low Advantage, and Shared Downside.
• Four strategies to enhance cooperation: correct misconceptions, collaborate on common research/engineering challenges, increase oversight where appropriate, and incentivize ethical/safety adherence.
• The analysis discusses how market forces, liability laws, and regulation interact with competition to shape responsible AI incentives.