Comment Empirical Testing for Social Theories (Score 1) 70
Anthropic is investigating classic political problems of social organization, but with new tools. It seems we now have empirical ways to test our philosophical theories of moral or ethical behaviour.
"Over many millennia, mechanisms like norms, reputation, costly signaling, and recourse have been refined to make human coordination go well," says the study. "While language models have inherited the content of that history, they don't necessarily carry the disposition produced by it." That's Anthropic's theory of how good behaviour develops in humans, and their vague idea of why the same behaviour isn't automatically exhibited by LLMs. But how does one come by the missing disposition? Is it just a matter of instilling the right ethical maxims, such as the Ten Commandments or Asimov's Rules of Robotics? Or is it perhaps a matter of social training and behavioural alignment, a sort of evolved "virtue ethics" -- and if so, whose society, whose values? Or do we need to consider utilitarian theories of the greatest good for the greatest number -- and how would AI's settle that?
It's easy to be suspicious of Big Tech's motives, but from a social studies perspective this seems like incredibly interesting work. We can now implement theories of social co-operation and test their results.
"Over many millennia, mechanisms like norms, reputation, costly signaling, and recourse have been refined to make human coordination go well," says the study. "While language models have inherited the content of that history, they don't necessarily carry the disposition produced by it." That's Anthropic's theory of how good behaviour develops in humans, and their vague idea of why the same behaviour isn't automatically exhibited by LLMs. But how does one come by the missing disposition? Is it just a matter of instilling the right ethical maxims, such as the Ten Commandments or Asimov's Rules of Robotics? Or is it perhaps a matter of social training and behavioural alignment, a sort of evolved "virtue ethics" -- and if so, whose society, whose values? Or do we need to consider utilitarian theories of the greatest good for the greatest number -- and how would AI's settle that?
It's easy to be suspicious of Big Tech's motives, but from a social studies perspective this seems like incredibly interesting work. We can now implement theories of social co-operation and test their results.
