Cornell: IER AI Revolutionizes Decision-Making

Cornell's AI for Informed Decisions

Cornell University has developed the "Interactive Explainable Ranking" (IER), an artificial intelligence tool designed to structure human decision-making. The goal is to highlight the gaps between the values declared by users and their actual choices.

The IER generates rankings based on micro-evaluations of options according to criteria that the user can adjust. This system has been validated through studies and received recognition at the ACM CHI conference. Available for free online, the tool is, however, only accessible in English.

A Personalized Decision Workshop

The IER functions as a true decision workshop. Users can submit a series of options to evaluate, whether they pertain to professional decisions, housing choices, or project proposals. The interface offers pre-established examples or allows users to import their own data via local files such as CSVs.

A concrete example is provided to illustrate its functionality: choosing a cat to adopt. Users define their priorities and evaluate each cat according to these criteria through a series of micro-choices. As responses are given, the IER constructs a ranking of candidates, highlighting the necessary trade-offs, such as accepting a noisier cat for a more affectionate temperament.

Analysis of Values and Choices

The tool asks users to describe their values by weighting various ranking criteria. It then identifies contradictions between these values and the rankings obtained. Users can specify the criteria that truly matter in their dilemma, whether it be salary, free time, ecological impact, safety, or personal fulfillment. They also have the option to rephrase, add, or remove criteria.

Based on these evaluations, the IER establishes a ranking that reflects the preferences revealed over the course of the responses, rather than the initial intentions. The integrated artificial intelligence acts as an assistant, suggesting criteria or proposing an initial sorting, but never replacing human decision-making. Most of the tool's features remain usable without the intervention of the AI.

Validation and Accessibility

To test the effectiveness of their system, researchers asked volunteers to rank short films. Another experiment was conducted with four teaching assistants tasked with sorting student projects. In both cases, the rankings produced with the tool proved to be consistent and easy to justify.

Although the tool is available for free online, it is only in English and requires some adjustment time for the user. The researchers emphasize that the IER is not a gimmick, but a useful tool for high-stakes decisions, such as recruitment, candidate evaluation, project prioritization, or even adopting a pet.