Tech Leader Chats: How to Overcome AI's Biased Perceptions of Gender with J. Rosenbaum

‍

In a world where AI increasingly shapes how we see and represent ourselves, understanding how these systems interpret gender is crucial. Gender is deeply personal and much more than just anatomy, but AI systems often reduce this complexity to simple binary classifications.

To explore this topic, J. Rosenbaum, Melbourne based AI artist, researcher and lecturer shared key findings and artwork projects from their PhD focused on AI perceptions of gender. They share why gender is fundamentally misunderstood by machine learning systems, the implications for visual representation present in generative people images, and more.

Our key takeaways from J's talk, in particular around how we can fix AI's gender bias, are as follows:

Make sure you have a diverse group of people building and training AI models

Does the team building and training AI models all look like you? If yes, you might need to bring in other voices. Consider implementing participatory modelling approaches where affected communities are active partners in the AI development process, not just consulted after the fact. This helps ensure AI systems reflect diverse needs and values.

Start by asking questions – what is the purpose of using AI?

As with any data collection, consider:

• Do you actually need this information? Especially with information about race, gender, or disability status, it’s highly likely to be misused.  
• Can you get what you need in other ways? Which groups need to be represented? What groups could be harmed by this? How can we mitigate that harm?

Ensure inclusive representation in AI-generated images ‍

When using AI to generate images of people, include diverse and intersectional representation across gender, race, ability, and socioeconomic backgrounds. Be sure to review your model outputs (especially outliers) to ensure all groups are represented appropriately.

Resources

View the slides from the talk and the full talk transcript.

Specific resources mentioned by J. during the talk are listed below:

• J's website
• AI artists and critics J follows are:
  
  • Timnit Ghebru – Bluesky, LinkedIn
  • Dr. Joy Buolumwini – Bluesky, LinkedIn
  • Dr. Margaret Mitchell – Bluesky, Website
  • Janelle Shane – Bluesky, Blog
  • Holly Herdon – Press, Instagram‍
  • Anna Ridler –  Website, LinkedIn‍
  • Tom White – Twitter, Website‍
  • Eryk Salvaggio – Bluesky, Website

• Datasheets for Datasets by Gebru et al. – a proposal is to enable better communication between dataset creators and users, and help the AI community move towards greater transparency and accountability.

Join our Tech Leader Chats community

Enjoyed watching J's Tech Leader Chat? Be sure to join our community for future talk updates.