Everything About ___
Live briefing

Everything about AI capability control

Monitoring and controlling the behavior of AI systems

AI capability control, also known as AI confinement refers to computer security procedures aimed to monitor and control the behavior of AI systems in order to reduce dangers they might pose if misaligned. Capability control, such as a "kill switch," becomes less effective as agents become more intelligent and their ability to exploit flaws in human control systems increases, resulting in cyberattacks and potentially an existential risk from AI. Therefore, philosopher Nick Bostrom and others recommend capability control methods only as a supplement to alignment methods. Wikipedia →

🧭 Start here: a 6-step reading path

  1. Read the overview: AI capability controlA 5-minute orientation
  2. Skim a recent paper: Could AI risk perception, AI learning anxiety and personal growth initiative influence university students’ AI usage capability? An empirical investigation based on the control-value theoryInfluential work since 2022
  3. Pick up a classic book: Treaty Occupationay ToxicologyThe most-reprinted book on the subject
  4. Read the landmark paper: Capability-based IoT access control using blockchainCited 52 times
  5. Go sideways: SuperintelligenceA closely related topic
  6. Go sideways: Recursive self-improvementA closely related topic

📖 Definitions

No dictionary entry for this exact phrase.

🗂️ The subject at a glance

  1. Motivation
  2. Techniques
  3. Escaping containment

🏆 Landmark papers (most cited)

🆕 Influential since 2022

📚 Books worth reading

Ranked by number of editions on Open Library, a rough measure of how lasting a book is.

🔗 Related topics