Can we simulate human evolution to create a somewhat aligned AGI?

Abstract

If AI alignment is intractable but human evolution reliably produces something close to human values, simulating or mimicking human evolution might offer a path, albeit imperfect, towards creating a superintelligent AI that is at least partially aligned with human values. This approach, however, presents several challenges, notably the difficulty of identifying and replicating the specific features of human evolution that are essential for the emergence of human-like values, and the risk that the simulated evolution process might not be competitive with other AI development approaches. Despite these hurdles, exploring this approach could be more worthwhile than solely focusing on achieving perfect AI alignment, which itself might be an intractable problem. – AI-generated abstract.