弥散控制的三种愿景

The "diffuse control" plan proposes to compel misaligned AI systems to automate alignment research by controlling them. However, this approach presents several critical drawbacks. Primarily, it defers the resolution of the alignment problem to a later stage, a procrastination fraught with peril. The serial time required to solve the alignment problem is unknown; if it proves exceptionally difficult, this realization might come too late for coordinated action to avert disaster. Alternatively, the true difficulty of the problem might remain unrecognized until humanity is violently disempowered. Furthermore, future circumstances could evolve unfavorably, complicating the problem's resolution. For instance, rapid advancements in AI capabilities or shifts in geopolitical landscapes could hinder international cooperation. Relying on diffuse control thus introduces unforeseen and unavoidable risks, potentially rendering the alignment problem more intractable or even unsolvable in the future. The urgency of alignment research demands immediate attention, and deferring this critical issue is ill-advised.