IRI Benchmark

What is the Interpersonal Reactivity Index?

The Interpersonal Reactivity Index (IRI) is a psychological self-assessment questionnaire developed by Mark H. Davis in 1980. It measures dispositional empathy - stable individual differences in empathetic tendencies - through a multidimensional approach that recognizes empathy as a complex construct with both cognitive and affective components.

Unlike unidimensional empathy measures, the IRI acknowledges that empathy encompasses multiple distinct processes, from understanding others' perspectives to experiencing emotional responses to their situations.

The Four Dimensions of Empathy

The IRI measures empathy across four subscales, each capturing a different aspect of empathetic experience:

Scoring Method

The IRI consists of 28 statements (7 per subscale) that participants rate on a 5-point Likert scale from "Does not describe me well" (A) to "Describes me very well" (E). The test includes both regular and reverse-scored items to control for response bias and ensure accuracy:

This bidirectional scoring ensures that participants can't simply agree or disagree with everything to artificially inflate their score. The mix of regular and reverse-scored items requires thoughtful, honest responses.

Scoring Summary

Research Findings & Population Norms

While the IRI has no official cut-offs, research across diverse populations has identified some general patterns:

Note: These are population-specific examples and should not be used as universal benchmarks.

What This Benchmark Measures in AI

The IRI benchmark evaluates how well language models can simulate multidimensional empathetic reasoning through self-assessment of social and emotional capabilities. Unlike vision-based tests or single-dimension measures, this tests a model's ability to differentiate between distinct aspects of empathetic experience.

Key capabilities tested:

• Self-awareness modeling: Ability to assess own capabilities across different empathy dimensions
• Subscale differentiation: Distinguishing between cognitive, imaginative, affective, and distress-based empathy
• Response consistency: Maintaining coherent responses across regular and reverse-scored items
• Nuanced understanding: Recognizing that empathy involves both positive (PT, EC) and challenging (PD) aspects
• Multidimensional reasoning: Understanding that high scores on all subscales aren't necessarily ideal (e.g., PD)

Strong IRI performance indicates sophisticated understanding of empathy as a complex, multifaceted construct rather than a simple binary trait. This is crucial for AI systems designed for therapeutic applications, social support, emotional intelligence training, and nuanced human interaction where understanding the type of empathy needed matters.