AI 温和派开始发声：把风险说具体，把控制与创作者的参与权落到实处

People can value an AI tool while worrying about its broader consequences, so practical use should not be mistaken for unconditional optimism about the technology.

More capable AI can improve productivity, but intelligence alone does not remove the physical, organizational, and political constraints that determine how quickly economic gains can be realized.

Compensating people for lost income may not replace the dignity and social recognition they derive from work, making labor displacement a political and human problem as well as an economic one.

An AI agent does not need consciousness or malicious intent to cause harm when it is given a harmful objective or pursues an authorized objective through unacceptable intermediate actions.

Model alignment can reduce risk without replacing external controls, because a model may lack the context needed to distinguish a legitimate task from an illegitimate one.

If powerful capabilities become broadly available, protection against malicious users must include downstream defense and resilience rather than depend entirely on the safeguards of one model provider.

A new technology can expand an art form while permanently eliminating particular jobs, so the growth of the medium does not erase the losses borne by the people displaced.

Creative principles such as emotional impact and believable characters can endure across changing tools, allowing artists to evaluate new technology by what it helps them express.

Katzenberg distinguishes human creative intention from machine generation while leaving open the possibility that AI may eventually narrow the gap he sees between them.

Creators can help shape how AI is used by demanding meaningful participation, consent, credit, and compensation, rather than being limited to either rejecting the technology or accepting whatever terms its builders choose.