Do Not Make Human–AI Agreement the Goal
Human–AI agreement can hide anchoring or stubbornness. A better workflow preserves both provisional views, examines divergence and records why judgement changed.
Human-centred technology—from everyday life to deep engineering.
Human–AI agreement can hide anchoring or stubbornness. A better workflow preserves both provisional views, examines divergence and records why judgement changed.
A siren or phone alert is only one link in a public-warning chain. Real safety depends on authority, reach, understanding and a feasible action people can take in time.
Rooftop solar can stop in a power cut by design. Understanding islanding, inverter control and critical loads turns a vague resilience promise into a system you can test.
A backup can preserve every byte and still fail the person who needs it. A small recovery rehearsal reveals missing permissions, unreadable formats and knowledge trapped in one person's head.
Alkemata proposes portable capsules that combine ideas, evidence and practical workflows, helping readers use their own LLMs to build projects and return experience to shared knowledge.
A public decision is effective only when its recipient can receive, understand and act on it. Good notification systems join delivery evidence to accessible content and accountable recovery.
Digital tools can prepare and document asylum interviews, but trained humans must hear consequential accounts, manage ambiguity and remain answerable for credibility judgments.
A truthful case tracker shows what happened, who holds responsibility, what the person must do next, the deadline and how to challenge an incorrect state.
Risk scores can organise evidence and support system-level research, but they cannot turn group statistics into an individual fact capable of justifying guilt, sentence or liberty.
Alkemata’s new project dossiers take a modest idea, push it to its most consequential form, stress-test its effects, and work backwards towards a practical experiment.