S stu­pe­fies A when A ac­cepts S’s as­ser­tion with­out un­der­stand­ing it. I ar­gue stu­pe­fy­ing is an im­por­tant means both for good (co­op­er­a­tive, joint­ly ra­tio­nal in­quiry) and bad (ma­nip­u­la­tion), in ways that cur­rent mod­els of con­ver­sa­tion do not ac­count for.

I pro­pose and de­fend a norm that links won­der­ing, be­lief, and abil­i­ties to con­ceive: for any ques­tion one won­ders, one must be able to con­ceive po­ten­tial an­swers to it that one doesn’t re­ject. Some­times the best way to avoid vi­o­lat­ing this norm is to re­vise your be­liefs, but some­times it’s bet­ter to re­vise your con­cepts so that you are un­able to won­der the ques­tion, or so that you can con­ceive a new an­swer.

Ac­tu­al­ly 𝜑-ing doesn’t im­ply abil­i­ty to 𝜑 when the right kind of ran­dom­ness is in­volved. This kind of ran­dom­ness is in­volved in the sam­pling meth­ods that many cog­ni­tive sci­en­tists think we ac­tu­al­ly use in learn­ing. So, con­tra Fodor, we can learn con­cepts by hy­poth­e­sis test­ing with­out cir­cu­lar­i­ty.

We can take con­cepts to be in­con­sis­tent with­out go­ing in for in­fer­en­tial­ism. Bi­lat­er­al­ism is all that’s need­ed. But a puz­zle re­mains: what’s de­fec­tive about in­con­sis­tent con­cepts?

There is epis­temic val­ue to un­der­stand oth­ers “from the in­side”. This means that mere­ly pos­sess­ing a con­cept has an epis­temic cost, even ide­al­iz­ing away our com­pu­ta­tion­al lim­i­ta­tions.

A cer­tain kind of con­cep­tu­al frag­men­ta­tion is epis­tem­i­cal­ly ide­al.

The cat­e­go­ry of epis­temic emo­tions should be de­fined in nor­ma­tive rather than in­stru­men­tal terms.

We can hold both (i) that there is a ra­tio­nal re­quire­ment to pre­fer out­comes that are ob­jec­tive­ly best and (ii) that one can ra­tio­nal­ly pre­fer in ways that are par­tial to­wards one­self and one’s loved ones. But to do so, we can’t just ger­ry­man­der the ob­jec­tive bet­ter­ness or­der­ing over worlds. We need to re­think what it is an or­der­ing of.