Exploring How Large Language Models Understand Sexual Communication in Hypothetical Sexual Situations.

Area of Science:

• Artificial Intelligence
• Sexual Health
• Human-Computer Interaction

Background:

• Large language models (LLMs) are increasingly popular for information retrieval.
• Their application in sexual health, particularly regarding consent and refusal, is emerging.
• Understanding LLM capabilities in interpreting sexual communication is crucial.

Purpose of the Study:

• To evaluate how LLMs identify sexual consent and refusal in hypothetical scenarios.
• To compare LLM interpretations of consent with expert judgment.
• To assess LLM performance in understanding nuanced sexual communication.

Main Methods:

• Four LLMs were presented with seven manipulated sexual scenarios.
• Models assessed consent based on communication cues and context, providing explanations.
• LLM responses were compared against evaluations from eight gender-based violence experts.

Main Results:

• LLMs accurately identified consent and refusal in scenarios with clear verbal communication, aligning with expert consensus.
• Discrepancies arose when scenarios involved intoxication, impacting LLM understanding of consent.
• LLMs showed limitations in interpreting complex, context-dependent sexual communication.

Conclusions:

• LLMs demonstrate proficiency in recognizing explicit consent cues but falter with ambiguity.
• Expert disagreement with LLMs on intoxication-related consent underscores AI limitations.
• Enhancing LLM training on subtle human behavior is vital for reliable sexual health applications.