from Hacker News

Values in the wild: Discovering values in real-world language model interactions

by mprast on 4/22/25, 6:15 AM with 1 comments

• by totetsu on 4/22/25, 6:52 AM

  There's a widespread shift toward treating complex moral and social dynamics as problems that can be solved with empirical tests and algorithmic scoring. In the context of AI, this often means trying to extract or measure "values" from text, projecting moral agency onto systems that are fundamentally statistical and synthetic. But this is not really about understanding values, it's about trying to automate judgment.

  This tendency is rooted in the logic of platform capitalism: tech companies want to scale social systems to the global level while avoiding responsibility for the moral and cultural environments they create. They act as if it's possible to moderate, govern, and define fairness algorithmically, without admitting that human morality, community norms, and deliberation don't scale cleanly. In reality, they're shaping the very terms of discourse, subtly steering culture, politics, and public reasoning through infrastructures designed for engagement and growth, not ethical depth.

  The result is an epistemic and moral outsourcing, where we increasingly defer to technical systems to tell us what is fair, acceptable, or valuable. This is not just a methodological misstep, it's a cultural displacement. We risk losing the habit of judgment, the space for ambiguity, and the responsibility to deliberate collectively as moral agents.