Cleo Nardo comments on Towards Hodge-podge Alignment

[Cleo Nardo]

The heuristic is “assemblage is safer than its primitives”.

Formally:

• For every primitive $p$ and assemblages $A_1$ and $A_2$ and wiring diagram $D$, the following is true:

• If $D\circ (A_1\otimes p)$ strongly dominates $A_1$ then $D\circ (A_2\otimes p)$ weakly dominates $A_2$.

• Recall that $D\circ (A\otimes p)$ is the wiring-together of $A$ and $p$ using the wiring diagram $D$.

• In English, this says that $p$ can’t be helpful in one assemblage and unhelpful in another.

I expect counterexamples to this heuristic to look like this:

• Many corrigibility primitives allow a human to influence certain properties of the internal state of the AI.

• Many interpretability primitives allow a human to learn certain properties of the internal state of the AI.

• These primitives might make an assemblage less safe because the AI could use these primitives itself, leading to self-modification.

[DragonGod]

Can you please take a look at this comment?