Brief

The concept of Agent Foundations (AF) is critiqued for containing a significant amount of "bunk," or unsubstantiated claims. However, this "bunk" could potentially be valuable if it were more empirically grounded. Two opposing viewpoints, held by hypothetical individuals Alice and Bob, illustrate how agreement on these premises can stem from fundamentally different, even incompatible, reasons, leading to a shared disagreement with the conclusion that AF needs more empirical grounding. AI

The concept of Agent Foundations (AF) is debated, with arguments suggesting it's either a well-defined, self-contained theoretical task or an ill-defined endeavor. The author argues against the middle ground that AF requires empirical grounding, stating that if AF is a valid field, its empirical aspects exist outside of AF itself, similar to how computer implementation serves computability theory. The core idea is that AF should either stand on its own conceptual merits or its components should be reframed within other, more coherent fields. AI