In this week’s episode, Remoy takes us to another side of AI. Grad computer science researcher Maria Teleki lends her findings to shed light on how AI feeds our existing patriarchal notions and structures
- What is a large language model? We’re sure you’ve heard about this term, also known as an LLM, that’s becoming more and more relevant to our everyday lives.
- Remoy gives us the definition.
- Maria follows up with how this technology learns about us and how we speak. Are there speech differences between genders? Maria’s research searches for answers to that question in…podcasts.
- A particular word tends to emerge in women’s speech patterns.
- But it’s not just that one word. Maria and her team’s research uncover different common words and patterns in the ways people speak to their own gender.
- So what does that mean for LLMs? Maria explains.
- LLMs learn these gendered speech patterns and their programming informs their response. If the programming stems from patriarchy, the response will be patriarchal. The bias thrives.
- Maria continues, breaking down how when LLMs recognize speech to be men’s, users get better results, especially in more profitable sectors.
- The bias doesn’t stop at gender. Race and sexuality bias in LLMs also creates gaps and inequalities in technology usage.
- There is hope! Maria gives us a silver lining.
Referenced on this episode:
COMPANION PIECES:
