TLDR: A new study reveals that people exhibit a strong aversion to inequality in societal well-being, even more so than to personal risk, and this preference is consistent across the political spectrum. The research challenges the use of average life satisfaction as a policy metric, proposing nonlinear utility-based alternatives like Representative Life Satisfaction (RLS), which shows people are willing to sacrifice personal well-being for greater equality. These findings have significant implications for public policy and for designing AI systems that are aligned with fundamental human values of fairness.
A recent study delves into how individuals value well-being, revealing a widespread aversion to inequality that surprisingly transcends political divides. The research challenges conventional methods of measuring societal well-being and offers crucial insights for public policy and the development of ethical AI systems.
For years, life satisfaction has gained traction as a key metric for public policy, often used in place of traditional economic indicators like income or GDP. The idea is that life satisfaction captures a more subjective and holistic view of welfare. However, a core challenge arises: life satisfaction scores aren’t easily comparable or additive across individuals. A jump from 6 to 7 on a self-reported scale might not mean the same thing for everyone, and simply averaging these scores can implicitly endorse a utilitarian view where only the total ‘good’ matters, overlooking the well-being of the least fortunate.
To address this, researchers Crispin Cooper, Ana Friedrich, Tommaso Reggiani, and Wouter Poortinga conducted a stated preference experiment with 300 participants from a nationally representative UK sample. Participants evaluated life satisfaction outcomes for themselves and others under conditions of uncertainty, using an approach known as the “veil of uncertainty.” This method asks individuals to make decisions without knowing their own future position within a distribution, thereby revealing their true preferences regarding fairness and risk.
Key Findings on Risk and Inequality Aversion
The study found that a majority of participants displayed concave, or risk-averse, utility curves when evaluating well-being. This means people tend to value improvements in lower states of life satisfaction more highly than equivalent improvements in already high states. More strikingly, participants showed a stronger aversion to inequality in societal life satisfaction outcomes than to personal risk. In other words, people were more cautious when decisions affected the well-being of a population than when they affected only themselves.
Perhaps the most significant finding was that these preferences for fairness and aversion to inequality were unrelated to political alignment. This suggests a shared, fundamental normative stance on well-being that cuts across different ideological boundaries, offering a potential basis for consensus in a politically polarized world.
Implications for Public Policy
The research directly challenges the common practice of using average life satisfaction as a policy metric. Because individual utilities for life satisfaction are nonlinear, a simple average doesn’t accurately reflect collective human values. The study proposes the development of nonlinear, utility-based alternatives, such as Representative Life Satisfaction (RLS).
When the researchers applied their estimated utilities to UK life satisfaction data, they found that RLS scores were significantly lower than the mean life satisfaction. This implies that people, on average, would be willing to trade away a substantial amount of personal life satisfaction (up to 1.8 points on a 10-point scale) to achieve greater equality in well-being across the population. This highlights a strong public preference for equitable distributions over simply maximizing the average.
Also Read:
- Unveiling AI’s Inner World: How Language Models Express Preferences and Welfare
- AI’s Role in Decoding Emotions for Dispute Resolution
Guiding Value-Aligned AI Systems
The findings also have profound implications for the design and alignment of artificial intelligence systems. As AI increasingly makes decisions that impact human well-being, understanding these underlying human values is critical. The observed heightened interpersonal risk aversion suggests that AI systems operating on behalf of large populations should adopt a more precautionary stance to align with human intuitions about fairness and harm avoidance.
The independence of equity preferences from political identity could simplify the complex task of designing value-aligned AI, as it reduces the need to account for deep ideological divergences when modeling human values. However, the authors caution that AI systems should always operate under normative guardrails that cannot be violated, even in pursuit of aggregate gains, acknowledging that some ethical decisions are not suitable for purely numerical assessment.
This study provides both a methodological framework and empirical grounding to ensure that the values embedded in numerical calculations for policy and AI better reflect the preferences of the wider public. For more detailed information, you can read the full research paper here.
