Summary of Rel-a.i.: An Interaction-centered Approach to Measuring Human-lm Reliance, by Kaitlyn Zhou et al.
Rel-A.I.: An Interaction-Centered Approach To Measuring Human-LM Reliance
by Kaitlyn Zhou, Jena D. Hwang, Xiang Ren, Nouha Dziri, Dan Jurafsky, Maarten Sap
First submitted to arxiv on: 10 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel evaluation framework called Rel-A.I. is introduced to assess large language models’ (LLMs) ability to communicate uncertainty, risk, and limitation by measuring how much humans rely on their generated text. The approach focuses on behavioral responses from human interlocutors, rather than simple calibration methods. The study finds that contextual features of the interaction significantly impact human reliance behavior, with factors such as knowledge domain and greeting tone influencing the extent to which people rely on LLMs. For instance, humans rely 10% more on LMs when responding to calculation-based questions and 30% more on LMs perceived as competent. The results highlight the importance of considering interactional context in evaluating the risks of human-LM interactions, suggesting that calibration and language quality alone are insufficient. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can generate text that might not be entirely accurate or reliable. This paper talks about a new way to evaluate these models’ ability to communicate uncertainty, risk, and limitation. Instead of just looking at how well the model’s answers match what’s expected, this approach looks at how humans respond to the model’s output. It turns out that different situations can affect how much people trust what the model says. For example, if you’re asking a question about math, people are more likely to rely on the model’s answer than if they were discussing something else. This study shows that just knowing how good or bad the model is at answering questions isn’t enough – we also need to think about the context in which those answers are being given. |
