Governing Reality

Information architectures become governance-relevant when they allocate salience, representation, access, friction, authority, memory, feedback, and practical options in ways that materially shape later judgment and action. This is not a claim that algorithms determine people or that one system explains every outcome. The research separates mechanisms and tests them against rival explanations.

Men’s loneliness, status insecurity, economic pressure, sexual and relational frustration, and need for belonging are not invented by recommendation systems. The research question is whether engagement incentives select and amplify explanations that convert those conditions into recurring grievance, distorted social norms, antagonism, and dependence on the systems or creators offering the explanation.

That framing does not make harms symmetrical. Women and gender-marginalized people can face disproportionate violence, coercion, exclusion, and material loss. Men can also be disempowered by architectures that narrow their options, monetize insecurity, and train nonreciprocal models of relationship. Explanation does not remove accountability.

Current adjacent evidence includes experimental work on engagement ranking and social-norm misperception, research on online masculinity content and young men, and work testing platform designs for more productive dialogue.

The program studies model behavior, product incentives, and system boundaries together. Recent research found that sycophantic AI can increase users’ conviction that they are right while reducing willingness to repair interpersonal conflict. Agent research also shows that a nominal human-in-the-loop is insufficient when people cannot observe, understand, or interrupt consequential actions in time.

• Human-AI relation: sycophancy, simulated reciprocity, dependence, memory, refusal, and long-use spillover.
• Agent governance: delegation, authority attenuation, escalation, tool use, provenance, and consequential state transfer.
• Evaluation: declared versus operative behavior, power distribution, refusal tests, longitudinal repair, and counterfactual design.
• Cybernetics: feedback, observability, control, adaptation, requisite variety, path dependence, and the location of the observer.

Adjacent work includes experimental research on AI sycophancy and field research on agent oversight.

• Crimes Against Consciousness: case-based public research and podcasting. Evidence, counter-reading, unknowns, counter-architecture, follow-up.
• The Table: small structured conversations where participants state their current model, uncertainty, and update conditions. Clarified disagreement is a valid output.
• Counter-architecture: translate a documented failure mechanism into a different rule, interface, incentive, authority boundary, or institution and test the tradeoffs.
• Polycentria: the technical and institutional counter-design program for consequential boundaries between humans, models, agents, tools, and institutions.

Cases distinguish direct records, first-person accounts, third-party accounts, and coder inference. Rival explanations are required. Unknown is a valid result. A later interaction is used to test whether claimed change actually appeared in behavior. The Point of Relation codebook is currently v0.1 and has not yet been reliability-tested.

Read the research codex, working lexicon, and research orientation.

I am looking for researchers, engineers, policy practitioners, designers, funders, affected communities, critics, and people with direct experience of the systems being studied. Agreement with the vocabulary is not a requirement. Useful contributions include cases, counterexamples, methods criticism, technical review, funding for bounded research phases, pilot environments, and arguments that would force the model to change.