# Contested concept: control (of ai)

_**What you are reading**: one contested-concept page in Deliberus's deliberation graph. Deliberus maps human reasoning — including claims that are contested, mistaken, or extreme — so that disagreement can be examined precisely. A claim's presence here is not an endorsement: the graph records that someone argued it, how it connects to other reasoning, and how far scrutiny has gotten. The metadata on this page describes the state of that scrutiny, not the truth of the claim._

---

- **Contestedness**: semantic
- **Diagnosis**: Different notions of control make 'we can remain in control' look easy or impossible, and they mask the disagreement over centralization versus distribution of power.
- **Canonical**: `/concept/control (of ai)` (data: `/concept/control (of ai).json`)

## Senses in use
- **Centralized containment at the highest level (nonproliferation mindset) versus control exercised at the lowest possible level (resilience mindset)** — used in 3 claim(s)
- **Human-in-the-loop: review and approval of every AI action** — used in 3 claim(s)
- **Layered, intermediate oversight such as auditing, monitoring, fail-safes, and least privilege** — used in 3 claim(s)
- **Controlling a boxed superintelligent system, such as alignment for a 'galaxy brain'** — used in 2 claim(s)

## Claims using this term
- [Because human-in-the-loop control diminishes the benefits of automation, it either devolves into the human acting as a rubber stamp or is o…](/claim/claim_128dff06a828.md) _(sense: Human-in-the-loop: review and approval of every AI action)_
- [In cybersecurity, the principle of 'least privilege' ensures that actors only have access to the minimum resources needed for their tasks, …](/claim/claim_2a4a0f7695e0.md) _(sense: Layered, intermediate oversight such as auditing, monitoring, fail-safes, and least privilege)_
- [Defense against superintelligence requires concentrating power and exercising central control over AI technology.](/claim/claim_4fa0bd9d8e28.md) _(sense: Centralized containment at the highest level (nonproliferation mindset) versus control exercised at the lowest possible level (resilience mindset))_
- [Fortunately, there are many other flavors of control, such as auditing and monitoring, that fall between the two extremes [of delegating sa…](/claim/claim_6fc91c4287f5.md) _(sense: Layered, intermediate oversight such as auditing, monitoring, fail-safes, and least privilege)_
- [Human-in-the-loop control means a system in which every AI decision or action requires review and approval by a human.](/claim/claim_72cc9834122f.md) _(sense: Human-in-the-loop: review and approval of every AI action)_
- [AI systems with more and more power will in turn lead to a loss of control.](/claim/claim_8270e6c8e2f0.md) _(sense: Controlling a boxed superintelligent system, such as alignment for a 'galaxy brain')_
- [In most scenarios, human-in-the-loop control (a system in which every AI decision or action requires review and approval by a human) greatl…](/claim/claim_84c67670c87d.md) _(sense: Human-in-the-loop: review and approval of every AI action)_
- [If we presume superintelligence, the AI control problem evokes the metaphor of building a galaxy brain and then keeping it in a box, which …](/claim/claim_9c5d8b553fb9.md) _(sense: Controlling a boxed superintelligent system, such as alignment for a 'galaxy brain')_
- [In the resilience mindset, control over AI systems is usually exercised at the lowest possible level.](/claim/claim_c33e8d8a1c15.md) _(sense: Centralized containment at the highest level (nonproliferation mindset) versus control exercised at the lowest possible level (resilience mindset))_
- [In the nonproliferation mindset, control over AI systems is exercised at the highest (most centralized) level possible.](/claim/claim_c632cbc9db65.md) _(sense: Centralized containment at the highest level (nonproliferation mindset) versus control exercised at the lowest possible level (resilience mindset))_
- [There are many flavors of control that fall between model alignment and human-in-the-loop control, such as auditing and monitoring, fail-sa…](/claim/claim_fff1e5346997.md) _(sense: Layered, intermediate oversight such as auditing, monitoring, fail-safes, and least privilege)_