Summary of Tell Me Why: Training Preferences-based Rl with Human Preferences and Step-level Explanations, by Jakob Karalus
Tell me why: Training preferences-based RL with human preferences and step-level explanations
by Jakob Karalus
First submitted to arxiv on: 23 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Human-Computer Interaction (cs.HC); Machine Learning (cs.LG)
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 The proposed preference-based learning method allows humans to provide more expressive feedback by offering a factual explanation for their preferences over two trajectories. This approach enables humans to explain what parts of the trajectory are most relevant for their preferences, allowing for individual step-level annotations. The method is evaluated in various simulations using a simulated human oracle, with results showing that extended feedback can improve the speed of learning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to train agents allows people who aren’t experts to help through different interfaces. Recently, preference-based methods have become popular because they let us train in areas where direct feedback is hard to get. However, current methods have limitations and don’t give humans a good interface for giving feedback. This work proposes a new method that lets humans provide more detailed explanations for their preferences over two paths. These explanations help the human explain what parts of the path are most important for their preference. The approach is tested in various simulations using a fake human oracle, with results showing that the extra feedback can speed up learning. |
