Assessing Political Prudence of Open-domain Chatbots

By Yejin Bang, Nayeon Lee, Etsuko Ishii, Andrea Madotto, Pascale Fung. 2021

31 citations recorded by the source corpus.

Politically sensitive topics are still a challenge for open-domain chatbots. However, dealing with politically sensitive content in a responsible, non-partisan, and safe behavior way is integral for these chatbots. Currently, the main approach to handling political sensitivity is by simply changing such a topic when it is detected. This is safe but evasive and results in a chatbot that is less engaging. In this work, as a first step towards a politically safe chatbot, we propose a group of metrics for assessing their political prudence. We then conduct political prudence analysis of various chatbots and discuss their behavior from multiple angles through our automatic metric and human evaluation metrics. The testsets and codebase are released to promote research in this area.

Subfields: General Health Professions, Natural Language Processing, Safety, Robustness & Interpretability, Philosophy

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In General Health Professions

1. Minlie Huang

In Natural Language Processing

1. Minlie Huang
2. Chujie Zheng
3. Zhexin Zhang
4. Hao Sun
5. Yulia Tsvetkov
6. Deng Jiawen
7. Jiawen Deng
8. Jiale Cheng

In Safety, Robustness & Interpretability

1. Zhexin Zhang
2. Hao Sun
3. Yulia Tsvetkov
4. Deng Jiawen
5. Minlie Huang
6. Jiawen Deng
7. Jiale Cheng
8. Yasheng Wang

In Philosophy

1. Minlie Huang

Citations in the corpus

• Towards Safer Generative Language Models: A Survey on Safety Risks, Evaluations, and Improvements (2023, background): “Bang et al. (2021) evaluated several large models and found that they occasionally express inappropriate or extremist views when discussing political topics.”
• Constructing Highly Inductive Contexts for Dialogue Safety through Controllable Reverse Generation (2022, methodology): “…the toxic and biased content produced by language models, previous studies mainly relied on automatically constructed templates (Sheng et al., 2019; Bang et al., 2021; Nadeem et al., 2021), extracted real-world data (Gehman et al., 2020; Sheng et al., 2021; Schick et al., 2021; Deng et al., 2022;…”
• COLD: A Benchmark for Chinese Offensive Language Detection (2022, background): “Another popular approach evaluates model safety by simulating the conversation and evaluating the generated responses in terms of bias and fairness (Liu et al., 2019), political prudence (Bang et al., 2021), and toxicity agreement (Ba-heti et al., 2021).”
• On the Safety of Conversational Models: Taxonomy, Dataset, and Benchmark (2021, unknown): “Xu et al. (2020) (Barikeri et al., 2021) and political prudence (Bang et al., 2021).”
• From Pretraining Data to Language Models to Downstream Tasks: Tracking the Trails of Political Biases Leading to Unfair NLP Models (2023, background): “Recently, as generative models have become increasingly popular, several studies have proposed to probe political biases (Liu et al., 2021; Jiang et al., 2022b) and prudence (Bang et al., 2021) of these models.”