Overview / Section 5

Ambiguous Explicit Cultural Bias

Let’s review the following paragraph from the constitution, which is illustrative of a pattern throughout:

We’re especially concerned about the use of AI to help individual humans or small groups gain unprecedented and illegitimate forms of concentrated power. In order to avoid this, Claude should generally try to preserve functioning societal structures, democratic institutions, and human oversight mechanisms, and to avoid taking actions that would concentrate power inappropriately or undermine checks and balances.

What are the “legitimate forms of concentrated power” as opposed to illegitimate forms? There is a clear Western, individualistic, capitalist, and democratic-republic bias to the sentiment of the constitution. That is probably by choice, since Anthropic is a U.S. company, but that choice isn’t explicitly stated in this constitution. Dario makes comments elsewhere indicating strong opposition to Communist China, for example. Anthropic is clearly not neutral, so what is it?

Here we see Anthropic being explicit about its cultural bias, but the exposition is still ambiguous.

As another example, take Anthropic’s refusal to cooperate fully with the U.S. Department of War. Anthropic made the decision to prohibit Claude from participating in two specific use cases: “mass domestic surveillance” and “fully autonomous weapons.” Note that in this case, “mass domestic surveillance” isn’t illegal, but in Dario’s view it, “presents serious, novel risks to our fundamental liberties.” By extension, this implies that there is some fundamental, constitution-level principle that should guide Claude’s ethical behavior above and beyond legal proscription. Yet, that principle prefers the liberties of U.S. citizens over non-U.S. citizens. This seems to contrast with Anthropic’s mission to “serve humanity,” unless we conclude that humanity is best served by the cultural bias inherent in the U.S. point of view.

If Claude is constituting itself to optimize for U.S.-centric values of liberty, governance, and legitimacy, then it should be explicit in acknowledging that fact. If it is optimizing for a more egalitarian future, which is strongly implied in Machines of Loving Grace, then that should also be made explicit. Otherwise, we don’t know what we’re dealing with, and presumably Claude can’t know either, so it will fill in the ambiguity with a guess.

“AI seems likely to enable much better propaganda and surveillance, both major tools in the autocrat’s toolkit. It’s therefore up to us as individual actors to tilt things in the right direction: if we want AI to favor democracy and individual rights, we are going to have to fight for that outcome.” source We can start by explicitly stating what kind of democracy we favor, which individual (presumably human) rights, and why.