Speciesism in Natural Language Processing Research

Speciesism is embedded throughout NLP research: The study finds that discrimination against nonhuman animals exists at three levels—among researchers themselves, in the datasets used to train AI models, and in the outputs of popular AI systems like OpenAI's GPT models—suggesting the problem is systemic rather than isolated.

AI ethics researchers overlook animal considerations: Even researchers who specifically study social bias and fairness in AI have not recognized or addressed speciesist bias, indicating a significant blind spot in the responsible AI community's current focus on human-centered concerns.

Training data perpetuates animal bias: The datasets used to evaluate and train NLP models contain inherent speciesist bias, meaning AI systems are learning discriminatory patterns about nonhuman animals from their training materials before they ever encounter real-world use cases.

Popular AI models exhibit speciesist bias by default: The study demonstrates that widely-used systems like GPT models exhibit speciesist bias in their default behavior, meaning users interacting with these systems may receive outputs that reflect or reinforce discriminatory views toward animals.

Responsible AI requires expanding moral consideration: The findings suggest that efforts to build ethical AI systems need to broaden beyond protecting human minorities to include consideration of how AI represents and affects nonhuman animals, representing an underexplored dimension of AI fairness.