# rohinmshah comments on Embedded Naive Bayes

• I usu­ally imag­ine the prob­lems of em­bed­ded agency (at least when I’m read­ing LW/​AF), where the cen­tral is­sue is that the agent is a part of its en­vi­ron­ment (in con­trast to the Carte­sian model, where there is a clear, bright line di­vid­ing the agent and the en­vi­ron­ment). Afaict, “em­bed­ded Naive Bayes” is some­thing that makes sense in a Carte­sian model, which I wasn’t ex­pect­ing.

It’s not that big a deal, but if you want to avoid that con­fu­sion, you might want to change the word “em­bed­ded”. I kind of want to say “The In­ten­tional Stance to­wards Naive Bayes”, but that’s not right ei­ther.

• Ok, that’s what I was figur­ing. My gen­eral po­si­tion is that the prob­lems of agents em­bed­ded in their en­vi­ron­ment re­duce to prob­lems of ab­strac­tion, i.e. world-mod­els em­bed­ded in com­pu­ta­tions which do not them­selves ob­vi­ously re­sem­ble world-mod­els. At some point I’ll prob­a­bly write that up in more de­tail, al­though the ar­gu­ment re­mains in­for­mal for now.

The im­me­di­ately im­por­tant point is that, while the OP makes sense in a Carte­sian model, it also makes sense with­out a Carte­sian model. We can just have some big com­pu­ta­tion, and pick a lit­tle chunk of it at ran­dom, and say “does this part here em­bed a Naive Bayes model?” In other words, it’s the sort of thing you could use to de­tect agenty sub­sys­tems, with­out hav­ing a Carte­sian bound­ary drawn in ad­vance.