One axis on which people’s views on alignment vary is the degree to which we should be prioritizing theory vs. application. Some claim that we need to have a fundamental understanding of agency before we can align AI, while others believe that we should focus on experimentation, since that’s a far more tractable way of gaining knowledge.
In the end, we want to figure out how to build an aligned AI, so it seems useful to examine past technological progress and see how much of it was driven by theory vs. trial and error. It might be good to examine past inventions (e.g. the bicycle, the airplane, the computer) and see how much they were driven by theory vs. experimentation. In the end, theory and application probably go hand in hand. But it seems that their relative importance differs from field to field. Examining past inventions may shed light on how this plays out for AI alignment.
Before I go do this myself, I was wondering if other people had examined this question and come to any conclusions.