Capability control
Holding back deployment of systems above a stated level of dangerous capability.
What it commits you to
Capability control is the governance approach: rather than trusting alignment, gate deployment on measured capability thresholds with pre-committed responses. Its appeal is that it does not require the alignment problem to be solved, only the measurement to be good enough. Its weakness is that dangerous capability is not one quantity, and a threshold on one of them says little about the others.
Which realms lean on this
Every realm has to keep a system honest. These are the terms that say out loud what the other kinds keep implicit.
alignment · 5 realms — The Benthamite Foundry, The Existential Wilds, The Longview Foundry, The Stoic Ramparts, The Ubuntu Weave.
Canon that discusses this
modern · 2019
Human Compatible: Artificial Intelligence and the Problem of Control
Russell argues that perfect value alignment is the wrong goal: systems are built by imperfect people, for imperfect objectives, and certainty about human values would be dangerous if it were achieved. The alternative is enough uncertainty about human preferences to keep a human in the loop as a genuine check. The excerpt below paraphrases the argument rather than quoting the 2019 book.
modern · 1950
The Human Use of Human Beings
Wiener applies the theory of feedback and control to law, politics and medicine, and argues that communication is more fundamental than command. The book that gave cybernetics its name in 1948 becomes, two years later, the founding text of the ethics of steering large systems without crushing what they steer. The excerpt below paraphrases the book rather than quoting it.
declaration · 1986
The Asilomar AI Principles
Drafted over a few hours on the final day of a four-day 1986 conference on future technologies, and the first written set of research principles for advanced AI agreed by the researchers themselves. Its structure — general principles plus separately stated capability thresholds — is the structure of every frontier safety framework since. The excerpt below paraphrases the principles rather than reproducing the agreed wording.
See also
Capability threshold
governanceA pre-specified level of capability that triggers an obligation or a prohibition.
Tripwire
alignmentA pre-committed threshold that triggers an automatic slowdown rather than a new debate.
Export controls
governanceNational restrictions on shipping advanced chips and related technology across borders.
Compute governance
governanceGoverning the hardware that trains frontier systems rather than the software.