How to Be Helpful to Multiple People at Once

Area of Science:

• Decision-making psychology
• Social welfare theory
• Human-computer interaction

Background:

• Decision-makers often face recipients with diverse preferences.
• Determining how to allocate resources or aid equitably is a complex challenge.
• Understanding human judgment in these scenarios is crucial for designing fair systems.

Purpose of the Study:

• To quantitatively model what people consider desirable behavior in decision-making scenarios with varied preferences.
• To identify which social welfare metrics best explain human choices.
• To inform the design of ethical AI and assistive technologies.

Main Methods:

• Developed a quantitative model to assess preferences using metrics like maximax, maxsum, maximin, and inequality aversion (IA).
• Conducted drink-choosing tasks where participants indicated their preferred outcomes.
• Varied experimental conditions, including participant roles and task complexity (multi-step decisions).

Main Results:

• Participant behavior was best explained by the maximin metric (maximizing the well-being of the least well-off).
• Maxsum (maximizing total group utility) and IA metrics also showed some consistency with participant behavior.
• Maximin behavior remained consistent across multi-step tasks, indicating a non-shortsighted approach.

Conclusions:

• Human decision-making in scenarios with diverse preferences leans towards ensuring the well-being of the most disadvantaged.
• Findings provide a basis for developing AI and robotic systems that align with human values of fairness.
• The study helps define desirable behaviors for future leaders and automated decision-making systems.