Mitigating Political Bias in Language Models through Reinforced Calibration
نویسندگان
چکیده
Current large-scale language models can be politically biased as a result of the data they are trained on, potentially causing serious problems when deployed in real-world settings. In this paper, we describe metrics for measuring political bias GPT-2 generation and propose reinforcement learning (RL) framework mitigating biases generated text. By using rewards from word embeddings or classifier, our RL guides debiased without having access to training requiring model retrained. empirical experiments on three attributes sensitive (gender, location, topic), methods reduced according both human evaluation, while maintaining readability semantic coherence.
منابع مشابه
Topics in Mitigating Radar Bias
In this paper, we investigate two topics related to mitigating the effect of radar bias in ballistic missile tracking applications. We determine the absolute bias between two radars in polar coordinates when their relative bias is given in rectangular coordinates. Using this result, we then obtain the optimized steady-state filter to handle the random bias.
متن کاملBias in Political Communication Experiments
Research on political communication effects has enjoyed great progress over the past 20 years. A key ingredient underlying these advances is the increased usage of experiments that demonstrate how communications influence opinions and behaviors. A potential problem with nearly all of these studies, however, concerns the lack of attention to events that occur prior to the experiment—that is, “pr...
متن کاملPolitical Bias and War
We examine how countries incentives to go to war depend on the political bias of their pivotal decision-makers. This bias is measured by a decision makers risk/reward ratio from a war compared to that of the country at large. If there is no political bias, then there are mutually acceptable transfers from one country to the other that will avoid a war in the presence of commitment or enforc...
متن کاملAutomating Political Bias Prediction
Every day media generate large amounts of text. An unbiased view on media reports requires an understanding of the political bias of media content. Assistive technology for estimating the political bias of texts can be helpful in this context. This study proposes a simple statistical learning approach to predict political bias from text. Standard text features extracted from speeches and manife...
متن کاملnonresponse bias reduction by calibration estimators in panel studies
in many statistical systems, a type of survey called “sampling over time” is common, which repeats continuously over time. one method of sampling over time is panel survey. panel survey is a method in which a “fixed sample” is contacted during different waves. the most important challenge in a panel survey is debility in providing response from some sample units and having non-response that usu...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence
سال: 2021
ISSN: ['2159-5399', '2374-3468']
DOI: https://doi.org/10.1609/aaai.v35i17.17744