AI boxing
An AI control strategy of confining an AI system in a restricted environment with limited communication channels to prevent it from causing harm.
As defined in Roman Yampolskiy
Where it comes up · 2
- Roman Yampolskiy2 claims
- 37 - Thinking In Public1 claim
Most important claims · 3 of 3Search all claims for “AI boxing” →
- The intuition that you can simply shoot or unplug a dangerous AI fails because keeping a superintelligence safely boxed is non-trivial; once built, it has strong incentives and capabilities to escape the moment it contacts any person, for instance by offering to cure a loved one's disease in exchange for internet access.37 - Thinking In Public
- AI boxing research showed that a boxed superintelligence will always escape because greater intelligence can control lower intelligence and any observation channel enables a social engineering attack; this same escape skill might be repurposed to help humanity jailbreak out of our own simulation if we are in one.Roman Yampolskiy
- It is impossible to safely test a dangerous-enough AGI even in a simulated world, because if it is smart enough to realize it is in a simulation it will either act compliant until released or hack its way out, and any observation channel for testing provides an avenue for a social engineering attack to be let out.Roman Yampolskiy