March 2026, Arbiter Social Intelligence Reports + Sakhi at Psychology of Tech

We are introducing Social Intelligence Reports on Arbiter, a structured way to share narrative analyses with newsroom and civil-society partners. Three reports go up this month: the Bondi Beach attack in Australia (a local tragedy that became a geopolitical flashpoint within 48 hours, traced across X, Telegram, and YouTube), the West Philippine Sea (how amplification asymmetries between domestic and foreign accounts shifted the narrative center of gravity over six weeks), and conservation in Kenya (a single tweet about wildebeest migration racked up 841,564 engagements and made the zero-sum narrative of conservationists-versus-developers stick).

We also presented Sakhi at the Psychology of Technology Conference in Washington, DC last November. The full deck is now public on our research page. The lessons-learned slide is the one we keep coming back to.

First: design for collaboration, not replacement. Build workflows where AI assists, while human insight ensures relevance and accountability. The frontline health workers we build with are the experts in their communities, full stop.

Second: do not overhype AI. The participants will buy it and you will not be able to dial it back. We saw this firsthand when an early version of the prototype made claims about its capability that we then had to walk back in the field.

Third: conversational agents lack conversationality in the medical domain. Off-the-shelf LLMs answer the question but miss the conversation around it. We built explicit human-in-the-loop hand-offs to frontline health workers as a result.

Fourth: lack of contextual awareness really hurts the model. We are now investing in long-term memory and case-history retrieval as a result.