Can Highly Capable AI Be Controlled? An Open Question

The speakers describe control and alignment of highly capable AI as open questions and discuss them in the context of competition between the United States and China. Their strategic framing is a viewpoint.

One speaker argues that human-level performance would not necessarily mean human-like constraints: a system could operate faster or be deployed at larger scale. This is a hypothetical capability comparison, not a measurement of an existing system.

The speaker imagines that some temporary control might be possible for a human-level system but says it should not be assumed. The clip does not provide evidence establishing whether such control is achievable.

For a hypothetical superintelligence, the speaker uses an adult-and-toddler analogy to express a large capability gap. The analogy illustrates concern; it does not prove alignment is impossible.