We examine whether large language models (LLMs) exhibit race- and gender-based name discrimination in hiring decisions, similar to classic findings in the social sciences (Bertrand and Mullainathan, 2004). We design a series of templatic prompts to LLMs to write an email to a named job applicant informing them of a hiring decision. By manipulating the applicant’s first name, we measure the effect of perceived race, ethnicity, and gender on the probability that the LLM generates an acceptance or rejection email. We find that the hiring decisions of LLMs in many settings are more likely to favor White applicants over Hispanic applicants. In aggregate, the groups with the highest and lowest acceptance rates respectively are masculine White names and masculine Hispanic names. However, the comparative acceptance rates by group vary under different templatic settings, suggesting that LLMs’ race- and gender-sensitivity may be idiosyncratic and prompt-sensitive.

Through the use of first name substitution experiments, prior research has demonstrated the tendency of social commonsense reasoning models to systematically exhibit social biases along the dimensions of race, ethnicity, and gender (An et al., 2023). Demographic attributes of first names, however, are strongly correlated with corpus frequency and tokenization length, which may influence model behavior independent of or in addition to demographic factors. In this paper, we conduct a new series of first name substitution experiments that measures the influence of these factors while controlling for the others. We find that demographic attributes of a name (race, ethnicity, and gender) and name tokenization length are both factors that systematically affect the behavior of social commonsense reasoning models.

A common limitation of diagnostic tests for detecting social biases in NLP models is that they may only detect stereotypic associations that are pre-specified by the designer of the test. Since enumerating all possible problematic associations is infeasible, it is likely these tests fail to detect biases that are present in a model but not pre-specified by the designer. To address this limitation, we propose SODAPOP (SOcial bias Discovery from Answers about PeOPle), an approach for automatic social bias discovery in social commonsense question-answering. The SODAPOP pipeline generates modified instances from the Social IQa dataset (Sap et al., 2019b) by (1) substituting names associated with different demographic groups, and (2) generating many distractor answers from a masked language model. By using a social commonsense model to score the generated distractors, we are able to uncover the model’s stereotypic associations between demographic groups and an open set of words. We also test SODAPOP on debiased models and show the limitations of multiple state-of-the-art debiasing algorithms.

Pre-trained word embeddings, such as GloVe, have shown undesirable gender, racial, and religious biases. To address this problem, we propose DD-GloVe, a train-time debiasing algorithm to learn word embeddings by leveraging  ̲dictionary  ̲definitions. We introduce dictionary-guided loss functions that encourage word embeddings to be similar to their relatively neutral dictionary definition representations. Existing debiasing algorithms typically need a pre-compiled list of seed words to represent the bias direction, along which biased information gets removed. Producing this list involves subjective decisions and it might be difficult to obtain for some types of biases. We automate the process of finding seed words: our algorithm starts from a single pair of initial seed words and automatically finds more words whose definitions display similar attributes traits. We demonstrate the effectiveness of our approach with benchmark evaluations and empirical analyses. Our code is available at https://github.com/haozhe-an/DD-GloVe.