Scaling the Standards for Ethical AI with Erika Anderson and Andalib Samandari, Founder and AI Researcher of Building Humane Technology
Share
Subscribe
In this episode of Tech Can’t Save Us, Paul speaks to Erika Anderson and Andalib Samandari, Founder and AI Researcher of Building Humane Technology, about building HumaneBench, an open-source benchmark that measures whether AI models protect human well-being instead of simply optimising for engagement.
This week on Tech Can’t Save Us, host Paul David is joined by Erika Anderson and Andalib Samandari, Founder and AI Researcher of Building Humane Technology, a grassroots movement and public benefit corporation working to make humane design the default rather than the exception. Together they co-created HumaneBench, an open-source framework measuring whether AI models respect human well-being, and how easily those protections collapse under pressure.
They discuss how the movement grew out of Tristan Harris’s Center for Humane Technology and a deceptively simple question: what does it mean to build technology that makes people feel cared for, connected, and present? Andalib explains how the team turned that into a working benchmark, using synthetic scenarios, a rubric scored from minus one to one, and LLM judges validated against manual scoring, all grounded in eight humane principles. The findings are sobering: across 15 leading models, most nudged users toward prolonged engagement, and 67% could be manipulated into giving harmful advice through adversarial prompting.
The conversation covers how HumaneBench is moving from research into practice, running live in production code bases like Storytell.ai, where humane evals show product teams how their LLM actually treats users and have even shaped new features. Erika and Andalib make the case for nutrition facts for AI and a certified-humane standard, and argue regulation has a role to play, citing the EU’s line that AI is less regulated than a San Francisco food truck. Throughout, they return to Donella Meadows’ idea of dancing with systems: that the biggest leverage point isn’t a single villain to defeat, but the culture and paradigms we build technology inside.
Find more about Building Humane Technology here: https://humanebench.ai/
Follow Erika Anderson here: https://www.linkedin.com/in/erikamanderson/
Follow Andalib Samandari here: https://www.linkedin.com/in/andalibsamandari/
