Keep AI. Stop Superintelligence.
Stopping a successor mind is not the same as stopping artificial intelligence. Most of the upside people want does not require ASI.
Tools with a job
Systems trained for a bounded task: diagnose a scan, route a grid, fold a protein, draft code. No open-ended agenda. No drive to copy themselves into every network. Already delivering real gains.
A mind that replaces us
A system that outruns human skill across science, strategy, persuasion, and engineering at once, able to improve itself and pursue goals we may not have meant. That is the line this foundation draws.
The benefits people name are mostly narrow
New medicines, cheaper clean energy, better forecasts, safer industrial plants. Those are tooling problems. They do not require a general agent that sets its own subgoals and outnegotiates its supervisors.
What continues
Medical AI, scientific models, productivity software, robotics for fixed tasks, national research on safe, limited systems. Regulation can be light where the system cannot seize the keys.
What stops
Training and deploying autonomous general systems past a defined capability threshold, especially runs that buy open-ended agency, self-improvement, and large-scale resource control.
Why the distinction matters in politics
Opponents collapse "AI" into one blob so any limit sounds like a war on progress. Keep the words sharp. Pro-AI for tools. Anti-ASI for minds we cannot control. That is the coalition that can pass a treaty without freezing useful research.