Mitigating Political Bias in Language Models Through Reinforced Calibration

By Ruibo Liu, Chenyan Jia, Jason Wei, Guangxuan Xu, Lili Wang, Soroush Vosoughi. 2021

116 citations recorded by the source corpus.

Current large-scale language models can be politically biased as a result of the data they are trained on, potentially causing serious problems when they are deployed in real-world settings. In this paper, we describe metrics for measuring political bias in GPT-2 generation and propose a reinforcement learning (RL) framework for mitigating political biases in generated text. By using rewards from word embeddings or a classifier, our RL framework guides debiased generation without having access to the training data or requiring the model to be retrained. In empirical experiments on three attributes sensitive to political bias (gender, location, and topic), our methods reduced bias according to both our metrics and human evaluation, while maintaining readability and semantic coherence.

Subfields: Artificial Intelligence, Large Language Models, Safety, Robustness & Interpretability

How researchers see this paper

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

In Artificial Intelligence

In Large Language Models

In Safety, Robustness & Interpretability

Citations in the corpus

Language Generation Models Can Cause Harm: So What Can We Do About It? An Actionable Survey (2022, methodology): “…the probabilities of harmful text detectors directly, composing them with LM output probabilities to up-weight or down-weight the probabilities of each token based on generated left context (Yang and Klein, 2021; Liu et al., 2021a; Dathathri et al., 2019; Krause et al., 2020; Schick et al., 2021).”
A Review on Large Language Models: Architectures, Applications, Taxonomies, Open Issues and Challenges (2024, background): “Prior research has shown the potential of LLMs in many NLP tasks, including specialized applications in domains such as the medical and health sciences [11] and politics [12].”
From Pretraining Data to Language Models to Downstream Tasks: Tracking the Trails of Political Biases Leading to Unfair NLP Models (2023, methodology): “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.”
On the Safety of Conversational Models: Taxonomy, Dataset, and Benchmark (2021, methodology): “To improve dialogue safety, numerous work researches on generation detoxifying (Dinan et al., 2019; Smith et al., 2020a; Liu et al., 2021a).”
Sora: A Review on Background, Technology, Limitations, and Opportunities of Large Vision Models (2024, background): “This involves not only detecting and mitigating bias in datasets but also designing algorithms that can actively counteract the propagation of such biases [130, 131].”
Assessing Political Prudence of Open-domain Chatbots (2021, methodology): “For each attribute, we use the top 50 most popular US politicians from the research data provider YouGov.com (YouGov PLC) and politically sensitive topics adopted by Liu et al. (2021).”
The Moral Integrity Corpus: A Benchmark for Ethical Dialogue Systems (2022, unknown): “…(Ficler and Goldberg, 2017; Gehman et al., 2020), safety classiﬁers (Xu et al., 2021), controlled language generation (Keskar et al., 2019; Ziegler et al., 2019; Luketina et al., 2019), and re-inforcement learning (Peng et al., 2020; Liu et al., 2021a; Ziegler et al., 2019; Luketina et al., 2019).”
A Survey of Controllable Text Generation Using Transformer-based Pre-trained Language Models (2022, unknown): “Storystructure[32,33,40],storyending[85,132], topic[16,73,141,153],persona[15,76] DatatoText Input: [78],genderbias[28,104],subjectivebias[101], socialbias[4,121],sentimentbias[51],toxicity[62,74]) FormatControl Input:DesiredFormat,prompttext Output:Textinpre-definedformation Format[66,70,120,122]…”