What would happen if society suddenly stopped advancing? As artificial intelligence continues to constitute a larger part of our lives this is an issue we face – AI scholars call it “value lock-in.”

Modern large language models (LLMs) are trained on existing data from the internet, produced by humans from the past or present, and regurgitate it for our use. Since the LLM is trained on material from the past or present, the societal values which underpin its responses will not change in time.

So, how do we overcome this? One solution is to periodically re-train existing AI models on new, human fabricated datasets that shift the values of the LLM to better align with modern humanity. This does, however, take lots of effort and would likely not be profitable for the private enterprise that currently leads AI development.

Perhaps this issue will be solved by someone in the far future, when societal values have significantly changed and the artificial intelligence value lock-in problem magnifies.