The best approaches for mitigating "the intelligence curse" (or gradual disempowerment)

Abstract

Addressing the risk of gradual human disempowerment as artificial intelligence capabilities outpace the value of human labor requires targeted regulatory and technical interventions rather than general economic diffusion. Mandatory interoperability for alignment and fine-tuning constitutes a primary mechanism to decentralize control. By requiring AI developers to provide third parties with deep access to model weights and fine-tuning interfaces, a competitive ecosystem of alignment providers can emerge. This structural unbundling allows for the customization of models to diverse user interests without the security risks associated with full open-sourcing. Complementing this, the development of personal AI representatives aligned to individual users enables humans to maintain agency by assisting with wealth management, political strategy, and the protection of shareholder or voting rights. Accelerating societal awareness through frequent, transparent model deployments further facilitates early negotiation and reduces the capability gap between private and public systems. These interventions prioritize maintaining individual leverage and democratic agency over broad human-AI capability enhancement, which may inadvertently accelerate disempowerment by shortening the transition period. – AI-generated abstract.