TLDR: The paper proposes a “proto-framework” for assessing the societal impact of AI systems, moving beyond the current risk-focused “responsible AI” paradigm. It operationalizes the concept of “freedom” (drawing on Kant, Sen, and others) into two dimensions: freedom as capability and freedom as opportunity. These dimensions are applied using the UN Sustainable Development Goals as thematic domains, with a structured evaluation process involving descriptive and numerical components, and a multi-stakeholder approach to gather diverse perspectives on AI’s aggregate societal effects.
In the ongoing global conversation about artificial intelligence (AI), the dominant approach to regulation often centers on ‘responsible AI,’ primarily focusing on mitigating risks. While crucial, this perspective can limit a comprehensive understanding of AI’s broader societal impact. A new research paper, “BEYOND RISK: A PROTO-FRAMEWORK FOR ASSESSING THE SOCIETAL IMPACT OF AI SYSTEMS” by Willem Fourie, proposes a novel proto-framework that shifts the focus from merely avoiding harm to proactively assessing how AI systems can enhance societal well-being by operationalizing the concept of freedom.
The Limitations of Current AI Regulation
Current AI regulatory frameworks, such as the European Union’s AI Act, are largely risk-based. They categorize AI systems by risk levels (unacceptable, high, limited, minimal) and impose compliance obligations aimed at preventing harm. While this is a necessary step, it often overlooks the potential positive contributions of AI. The paper argues that this risk-centric and AI system-centric approach, while well-intentioned, doesn’t fully capture the multifaceted nature of policymaking, which should also aim to enable flourishing and expand opportunities for citizens.
Drawing on Philosophical Foundations: Freedom as a Counterpart to Responsibility
To address this gap, the proto-framework introduces ‘freedom’ as a complementary concept to ‘responsibility.’ The paper delves into philosophical thought, starting with Immanuel Kant, who linked freedom to the active exercise of rationality and universal principles that serve both individual and collective well-being. Later thinkers like Max Weber and Hans Jonas expanded on this, with Jonas highlighting the ethical imperative of responsibility in the face of humanity’s unprecedented technological power, advocating for a ‘heuristic of fear’ to anticipate potential harms.
Building on these ideas, the framework operationalizes freedom through two key dimensions: ‘freedom as capability’ and ‘freedom as opportunity.’ Freedom as capability refers to the internal conditions that enable individuals and communities to act and engage meaningfully with the world. Freedom as opportunity, on the other hand, focuses on the absence of external constraints, evaluating whether AI systems expand or limit the choices available to people and institutions.
A Proto-Framework Aligned with Sustainable Development Goals
The proposed proto-framework concretizes these dimensions of freedom using the United Nations Sustainable Development Goals (SDGs). The SDGs, widely accepted global development objectives, provide a practical and legitimate proxy for categorizing AI’s impact. For ‘freedom as capability,’ the framework includes domains like prosperity, nutrition, health, education, water, energy, and housing. For ‘freedom as opportunity,’ it covers employment, innovation, socio-economic equality, gender equality, environment, government, and safety.
Each domain is evaluated using both descriptive components (identifying affected parties and the nature of impact) and numerical components (significance, scale, and likelihood of impact). This structured approach allows for a comprehensive assessment, moving beyond simple risk mitigation to understand the aggregate societal impact of AI systems.
Also Read:
- Designing AI for Human Well-being: A New Framework for Human-AI Interaction
- Navigating Trust and Regulation: A Global and GCC Perspective on AI in Education
A Collaborative Assessment Process
The framework is designed to be completed by diverse stakeholder groups: domain experts, system developers, and affected parties. This multi-stakeholder approach is crucial because societal impacts are complex and perceived differently depending on one’s position and experience. By comparing assessments across these groups, policymakers can identify areas of convergence (shared understanding) and divergence (potential blind spots or disproportionate effects), fostering a more nuanced and inclusive understanding of AI’s societal implications. The ultimate goal is to enrich policymaking processes by providing a structured, transparent, and comparable method for evaluating AI’s societal benefits alongside its risks, moving towards a more holistic approach to AI governance.
