AI Research
Jun 11, 2026
Research reveals new methods to enhance predictions of human preferences using utility models
Jun 11, 2026
AI Summary
A recent study highlights the limitations of traditional pairwise comparisons in random utility models (RUMs) for predicting human preferences. By incorporating rankings of three alternatives, researchers demonstrate a more effective approach to understanding correlations in choices, which could improve decision-making in various fields, including AI and consumer behavior.
- L. L. Thurstone's 1927 paper laid the foundation for random utility models (RUMs), which quantify human preferences and aid in predictions about choices in various scenarios.
- RUMs are commonly used in government and industry to forecast decisions in counterfactual situations, such as transportation routes during road closures or allocation of funds.
- A recent paper presented at an international conference by researchers from MIT and Nanyang Technological University found that traditional pairwise comparisons limit the ability to identify correlations between preferences.
- The study suggests that asking individuals to rank three options instead of two can reveal important correlations, improving the accuracy of preference estimations.
- The researchers developed algorithms to efficiently extract preference information from larger datasets, indicating that fewer experiments are needed than previously thought.
- This research is expected to enhance the effectiveness of RUMs in various applications, including the optimization of AI models and consumer recommendations.
preference predictionrandom utility modelsmit researchmachine learningdata analysis