Hoagy comments on Which possible AI systems are relatively safe?

[Hoagy]

Why would we expect the expected level of danger from a model of a certain size to rise as the set of potential solutions grows?