Any contextual ambiguity makes more apparent our own bias regarding potential machine errors,” says Marie Ozanne, and her research on consumer trust in technology helps illuminate this. Ozanne and her collaborators created a custom social media site ostensibly focused on food, and enlisted human beta testers who interacted with bots as fellow testers, generating interactive likes and comments. When vague and direct harassing comments were introduced, along with identified human and AI moderation in response, the authors found that users were more likely to question AI moderators, specifically when they moderated ambiguously negative comments. In the interest of transparency, Ozanne contends that this bias can be mediated “by making AI and humans work as moderating teammates.”
- Tags
-