Fairness as Equal Concession: Critical Remarks on Fair AI

Area of Science:

• Computer Science
• Ethics
• Artificial Intelligence

Background:

• Existing research identifies obstacles to fair AI but lacks a systematic framework.
• Common fairness notions like 'treat like cases alike' are insufficient for AI applications.
• Current approaches fail to adequately address the complexities and trade-offs in AI decision-making.

Purpose of the Study:

• To develop a robust and systematic conception of fairness for AI research.
• To address the limitations of existing fairness definitions in the context of AI.
• To provide a framework that guides ethical considerations in the development of fair AI.

Main Methods:

• A comprehensive review of fair AI and philosophical literature.
• Identification of three key desiderata for a functional conception of AI fairness: meta-theory for tradeoffs, avoidance of impartial perspectives, and emphasis on context.
• Proposal of a new definition: fairness as appropriate concession in historical institutional decisions.

Main Results:

• The proposed conception of fairness meets the identified desiderata.
• A process-structure map is developed to organize ethical considerations in fair AI.
• The framework offers clarity for computer scientists and ethicists working on fair AI.

Conclusions:

• Fairness in AI requires a context-sensitive approach, acknowledging historical and institutional factors.
• A new definition of fairness as 'appropriate concession' offers a more effective framework for AI.
• This work provides a foundation for further technical and philosophical advancements in fair AI.