A Virtuous AI is an Existential Risk

Work on emergent misalignment' shows that finetuning LLMs on narrow tasks can induce broadly misaligned behavior. This supports the persona selection' (PSM) hypothesis: during pre-training, LLMs learn to simulate different characters and perspectives, which can be elicited and refined during post-training. This paper investigates the converse phenomenon, emergent alignment', and uses it to support and refine the PSM and motivate a novel desideratum for alignment. We finetune a helpful-only model on broad and narrow safety tasks. To create SFT samples, we follow the Constitutional AI' (CAI) approach and use four constitutions which encode reasonable alignment strategies: deontology, consequentialism, virtue ethics, and aligning AIs as subordinate to human authority. For each of those models, we show that finetuning on two narrow safety sub-categories reliably induces emergent alignment over a representative set of general safety categories, and on safety subcategories that we directly filtered-out of the data sets used for narrow alignment. To test the PSM' using a more fine-grained evaluation, we used a multidimensional ethical persona' diagnostic. For each constitutionally finetuned (broad/narrow) model, we evaluate how well their behavior matches their expected signature profile. Our results show that our CAI models acquire their expected ``ethical persona'' -- e.g., the model narrowly fine-tuned on SFT samples created using the consequentialist constitution agrees significantly more with utilitarian than deontological beliefs. Yet our coarse and fine-grained evaluations show that there are significant differences across our (broad/narrow) finetuned CAI models in how well they project. We conclude that alignment strategies should be evaluated, not just on their (in-distribution) general safety performance, but also specifically on their degree of projectability.

The pursuit of artificial general intelligence (AGI) rests on a seemingly self-evident premise: that general intelligence, the kind of flexible, domain-general cognitive capacity exemplified by Homo sapiens, is extraordinarily valuable. This paper subjects this premise to critical scrutiny. We first present the intuitive case for the value of general intelligence before mounting an evolutionary challenge. We argue that, on evolutionary timescales, its adaptive value is far from empirically established. Numerous taxa, from cyanobacteria to horseshoe crabs, have persisted for hundreds of millions or even billions of years without anything resembling general intelligence, while Homo sapiens has existed for roughly 300,000 years and already faces self-generated existential risks. Mass extinction events do not preferentially favour cognitively sophisticated species. We argue that general intelligence may be the only biological strategy that generates existential threats to the species possessing it, an existential risk paradox with no parallel among non-intelligent survival strategies. Unlike prevailing accounts of AI risk that trace the danger to misalignment, we locate it in structural features of general intelligence itself, implying that even well-aligned AGI would inherit this liability. If the long-term evolutionary value of general intelligence is uncertain or negative, this raises ethical questions about engineering AGI and, more urgently, creating artificial consciousness. Drawing on deontological ethics and the precautionary principle, we argue that this uncertainty imposes a duty of caution: if we create a new kind of intelligent being, we bear responsibility for ensuring the conditions under which it can flourish.

This paper explores the idea of promoting well-being and safety in human-AI interactions by forcing AI agents explicitly to empower humans and to manage the power balance between humans and AI agents in a desirable way. Using a principled, partially axiomatic approach based on desirable properties, we design a parametrizable and decomposable objective function for AI systems that represents an inequality- and risk-averse long-term aggregate of human power. It can take into account models of human bounded rationality and social norms, and crucially, considers a wide variety of possible human goals. We prove how certain desiderata enforce particular functional forms and restrict parameter ranges. We exemplify the consequences of softly maximizing this metric in several paradigmatic situations and describe what instrumental sub-goals it will likely imply.