Today, our host, Carter Considine, explores one of the toughest hurdles in the AI space: the reality that today’s algorithms are not only reinforcing, but even amplifying, age-old biases.

Carter unpacks cases such as that of Google's Gemini AI, which sparked outrage after generating controversial outputs echoing real-world racial and gender stereotypes. He dissects the implications of these biases on companies leading AI innovation and why we need transparency in AI model development, as well as more diverse datasets and revamped testing methodologies.

Our host also discusses potential solutions proposed by AI researchers, such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization. As AI continues to develop at a rapid pace day by day, we’re hoping that future breakthroughs in the space involve the building of more inclusive algorithms.

Key Topics:

How Cutting-Edge Generative AI Models Have Generated Biases (0:27)
Developments and Continued Limitations in Generative AI Models (2:12)
Technical: Under the Hood (3:37)
Why RLHF is Not Enough (5:14)
Moving Forward (6:50)


More info, transcripts, and references can be found at ethical.fm

Ver todos los episodios