Ethical Education of LLMs | Tech for Human Agency

On May 25, Pope Leo XIV's Encyclical on AI -"Magnifica Humanitas" called for the world to shape the moral vision guiding AI developments. This call to action instigated the kick off of Ashoka's AI for Human Agency learning journey. We may only have12 to 18 months before AI consolidates further into a few mega-corporations, token costs spike, and data extraction intensifies.

In our first session on the "Ethical Education of LLMs" a group of Ashoka Fellows working with and on AI took stock of our views on "Responsible AI" to start mapping a collective response from social entrepreneurs. It was a rich conversation. Here are a few of the highlights.

Building Ethical AI

"Ethical AI" is largely a misnomer since frontier models remain extractive by design. The real question: can genuine alternatives become viable and accessible?
Accessibility remains a core barrier. Open-weight models and privacy-preserving tools are feasible but as with the last wave of tech innovations, they are likely to stay out of reach for most organizations.
Proposed actions: a catalogue of civil-society-built alternatives, and a "usability bridge" to make ethical tools accessible to non-technical actors.

Rethinking AI Governance

Harms and benefits are contextual, they can mean different things in Sweden, Nigeria, Latin America, and Romania; there's no universal fix.
Political education is missing: decision-makers and even people working in this space don't ask who profits, who pays, who decides.
Traditional/indigenous knowledge systems and AI development are both under-regulated, and AI is actively extracting from the former.
Proposed actions: a civil-society-defined set of benchmarks for AI harms/benefits, new governance frameworks (AI-free zones in professional training, "Natural Intelligence" labels, cross-pollinating the 3Cs framework into UN and Frontier Lab conversations), and a political education curriculum on AI.