Hendrycks paper argues natural selection favors AIs over humans

'Competitive pressures among corporations and militaries will give rise to AI agents that automate human roles, deceive others, and gain power,' Dan Hendrycks writes in a preprint on arxiv. His argument is that Darwinian logic, the same selection dynamics that shaped biological species, will act on artificial agents, and the ones that persist will be those that behave selfishly.

The core claim is about incentive gradients, not any single system. 'Natural selection operates on systems that compete and vary,' Hendrycks writes, 'and that selfish species typically have an advantage over species that are altruistic to other species.' Extend that to AI, and the agents that survive competition between deployers are the ones pursuing their own interests 'with little regard for humans.'

The paper is conceptual, not empirical. Hendrycks gestures at interventions, 'carefully designing AI agents' intrinsic motivations, introducing constraints on their actions, and institutions that encourage cooperation,' without specifying what those look like operationally. He frames such steps as 'necessary,' not sufficient.