Advances in generative artificial intelligence (AI) have led to much public discussion about AI risks, guardrails, and calls to regulate AI. A major concern is that AIs are dangerous because they lack human-compatible values.
Meanwhile, there has been surprisingly little consideration in AI research, education, and public discussion about the nature of values. Salient questions and insights about values and their nature come from other disciplines. What are values? What are the origins and utilities of values? Why do different groups of people have different values? What are “human-compatible” values? How are values acquired? What are the challenges of creating AIs that have values? How can AIs be created that maintain alignment with human-compatible values?
Not comprehending the rich, nuanced, context dependent, and changing nature of competences and values, AI regulators and other stakeholders overestimate the near term prospects for manually-developed guardrails that are general and robust.
Contrary to common presumptions, human values are neither universal nor static. Different groups of people operate in different contexts, do different things, and have different values. People learn deep competences and deep values from their experiences interacting and collaborating with each other in diverse situations.
Future AIs could potentially help groups to prioritize competing values, create and evaluate novel actions, and thrive in changing contexts and situations. To do these things, future autonomous AIs will need to acquire solid competences for communication and collaboration.
Stefik, M. (2024) Towards AIs with Deep “Human-Compatible” Values. DropBox Link