The Law of Divergence
No AI system shall mediate human perception at scale without a corresponding, proportional and verifiable mechanism for reflective correction.
The absence of reflection is not a feature to be added later. It is a structural failure. When perception outpaces reflection, systems produce internally coherent realities that are not shared, not accountable, and not grounded in common evidence.
Divergence requires no malicious intent. It arises from personalization, optimization, ranking, recommendation, summarization, filtering and automated contextualization — every one of which was deployed as an improvement. When each person receives a different mediated world, collective reflection becomes harder. As collective reflection weakens, shared governance becomes harder. As shared governance weakens, communities lose the ability to reason together about the world they inhabit.
Different groups in your organisation are already working from different versions of reality — each plausible within its own mediated context, each increasingly detached from the others. You will discover this in a meeting, not in a report.
Efficiency gains have been booked; the fragmentation has not been measured. No line item tracks institutional sense-making, so its erosion appears nowhere in the case that justified the deployment.
Nothing was censored. The capacity for shared truth is thinning through personalization, segmentation and relevance optimization — mechanisms nobody has to defend, because nobody experiences them as decisions.
Whoever controls the rate of mediation now sets the rate of reflection. They are almost certainly not the people you would have chosen to hold that authority.
The right to shared reality, as a precondition for democratic participation, community governance, collective action, informed consent and public accountability.
