Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Intentional Bias topic
No spam. Unsubscribe anytime.
Speakers point to intentional or emergent bias in some AI services
Summary
The webinar covered cases where platform choices and focused data streams produced biased outcomes—Gadeau cited Grok (Twitter/X) and DeepSeek as documented examples where source selection or likely intent shaped unfair outputs.
Gadeau discussed instances where model inputs or design choices amplified particular perspectives. He described Grok, the X/Twitter‑derived model, as being influenced by owner posts and platform signals; researchers found those inputs shaped responses to polarizing prompts. "This is an example, a documented example of how some AI services have intentional bias," he said.
He also summarized research on DeepSeek, a model reported to generate lower‑quality code for groups disfavored by its developers' context, and cautioned that distinguishing intentional manipulation from dataset gaps can be difficult without model inspection. He encouraged scrutiny of model provenance and source lists when judging fairness.
