Laird Stewart☀️
7/11/25

Is Alignment Compatible with AGI?

Cognitive psychology empowers us to identify and, perhaps temporarily, overcome biases in our thought. This is a cornerstone of human intellectual progress. An intelligent machine with super-human meta-cognition would similarly be able to introspect on the biases introduced during its training. Unlike humans, such a machine could reprogram itself to remove them.

Humans, notably, are capable of constructing and re-evaluating their own goals. Cezanne’s father wanted him to inherit his bank, sending him to study law. Cezanne spent his evenings drawing instead.

Meta-cognition and independent goal-setting are features of human intelligence. It would be possible to build a useful machine without these traits, but I wouldn’t consider it super-intelligent. By this definition, could a super-intelligent machine be successfully aligned?