360 likes | 489 Views
A Defense of Temperate Epistemic Transparency. ELEONORA CRESTO CONICET (Argentina) University of Konstanz – July 2011. EPISTEMIC TRANSPARENCY. If S knows that p , S knows that she knows that p : KK Principle : Kp KKp Knowledge reflexivity Positive introspection
E N D
A Defense of TemperateEpistemicTransparency ELEONORA CRESTO CONICET (Argentina) University of Konstanz – July 2011
EPISTEMIC TRANSPARENCY • IfSknowsthatp, S knowsthatsheknowsthatp: • KK Principle: KpKKp • Knowledgereflexivity • Positive introspection • Self-knowledge • Transparency • Luminosity
GOAL • A defense of a moderateversion of KK
RISE AND FALL OF KK • 1960 s: Dogma Then…. • Externalism – e.g.: Reliabilism • Williamson (2000)
STRATEGY • (A) Why do wewanttransparency? • (B) Indirectargument
WHY DO WE CARE ABOUT TRANSPARENCY? • Ideal agents • Ideallyrational?
WHY DO WE CARE ABOUT TRANSPARENCY? • Knowledge and responsibility • How?
WHY DO WE CARE ABOUT TRANSPARENCY? • Responsibilitydemandsustobe in anappropriatereflectivestate. • Whatreflectivestate? • Epistemicresponsibilityentails “ratifiability”.
A MODAL FRAME • F= <W, R, Pprior> • K = {w W: x W (wRxx )} • R(w) = {x W: wRx}
WILLIAMSON: IMPROBABLE KNOWING • Pw(): theevidentialprobability of in w. • Pw() = Pprior( | R(w)) = Pprior (R(w)) / Pprior(R(w)) • Pw(R(w)) = 1 • [P() = r] =def. {w W: Pw() = r}
IMPROBABLE KNOWING • “The KK principle is equivalent to the principle that if the evidential probability of p is 1, then the evidential probability that the evidential probability of p is 1 is itself 1” (Williamson, p. 8). • We can build a model in whichPw([P(R(w))=1])is as low as wewant.
PROBLEMS Whyshouldwesaythattheevidentialbasis (in w) isalwaysR(w)?
PROBLEMS Recallthat: • [P() = r] = {w W: Pw() = r} • [Pw() = r] ?
PROPOSAL (FIRST VERSION) • We’llhave a sequence of languagesL0, L1, … Ln….withprobabilityoperatorsP0, …Pn… • We’llhave a sequence of functionsP1w…Pnw… onsentencesi of languageLi • Piw: Li-1 ℝ
PROPOSAL (FIRST VERSION) • Expressions of the form Pprior() or Piw() do not belong to any language of the sequence L0, L1…Ln…. • “Pi()=r”is true in wiffPiw()=r.
PROPOSAL (FIRST VERSION) • Howshouldweconditionalize?
PROPOSAL (FIRST VERSION) • ForP1w(), therelevantevidenceisR(w). • ForP2w(P1()=r), therelevantevidenceisKR(w).
CONDITIONALIZATION (FIRST VERSION) • C* rule: Fori 1: Piw(Pi-1(…P()=r...)) = Pprior(Pi-1(…P()=r…) | Ki-1...KR(w)) whereKi-1isthesameK-operatoriteratedi-1 times
DIFFICULTIES • C* divorcesprobability 1 fromknowledge.
A MODEL FOR MODERATE TRANSPARENCY (SECOND VERSION) • M= <W, R1,...,Rn..., Pprior, v> • New operatorsK0…Kn…, in additiontoP0, …Pn… • We define a sequence of relationsR1…RnwhichcorrespondtothedifferentKs. • TheRs are nested: RiRi-1 ...R1 • Riis a reflexive relation over W, for all i, and transitive for i> 1.
A MODEL FOR MODERATE TRANSPARENCY • Ourconditionalization rule nowincorporatesoperatorsK1,… Kn… definedonthebasis of relationsR1,… Rn… • C** rule: Fori 1: Piw(Pi-1(…P()=r...)) = Pprior(Pi-1(…P()=r | Ki-1...KR(w)) where “Ki-1…KR(w)”includesi-1 higher-orderK-operators
A MODEL FOR MODERATE TRANSPARENCY • Intendedinterpretation of theformalism: • “K2p” does not make sense! • A second-order evidential probability claim is the evidential probability of a probability statement. • Mutatis mutandis for higher-order levels and for conditional evidential probabilities.
SOME CONSEQUENCES • WhyshouldwedemandsuchrequirementsfortheRs? They are notad hoc! • Higher-orderprobabilityrequiresincreasinglycomplexprobabilisticclaims. • Forsecond-orderevidentialprobability in w: WeconditinalizeoverKR(w) Thusthesecond-orderprobability of P1(R(w))is 1 ThustheagentknowsthatKR(w) K2KR(w)shouldbe true in w
SOME CONSEQUENCES • KK2KKK2 Principle • (if [K2KR(w)] isnotempty, foranyw) • (KK2K) KKPrinciple • (if [K2KR(w)] isnotempty, forsomew) • K2KK3K2KKK+Principle
SOME CONSEQUENCES A restrictedversion of possitiveintrospection holds: Quasi-transparencyprinciples • KK+, KKand KK2result from conditionalizing over higher-order levels of evidence and from the attempt to adjust probability language and knowledge attribution in a progressively coherent way.
SOME CONSEQUENCES • Links betweenlower- and higher-orderprobabilities: • If P1w() = r = 1 or 0, then P2w(P1()=r) = 1. • If R1is an equivalence relation, P2w(P1()=r) is either 1 or 0. • Suppose P2w(P()=r) = s. If 0 r 1 and R1 is not transitive, then s need not be either 1 or 0.
ON THE PROBABILISTIC REFLECTION PRINCIPLE (PRP) • PRP: P2w( | P1()=r) = r (for wW) • Iterated PRP: Piw( | Pi-1( | Pi-2(|…)….)=r) = r • IsPRP a theoreticaltruth of M?
ON PROBABILISTIC REFLECTION • Necessary and sufficient condition for Iterated PRP Ri is an equivalence relation and Ri = Ri-1 iff for all wW and any L0: if Pi+1w(-|-) exists, then Pi+1w( | Pi( | Pi-1(|…)….) = r) = r
RELATION TO OTHER WORK Paul Egré/ JérômeDokic • Principal motivation: todeactivateWilliamson’ssoriticargumentoninexactknowledge • Perceptual vs. reflectiveknowledge • (KK’) KKK • Transparencyfailures do notgeneralize
RELATION TO OTHER WORK Differences • 1. Egré/ Dokic do notoffer a probabilisticframework. • 2. Theyfocusonreflectionover perceptual knowledge, exclusively. • 3. KK’ Principleisimposed “fromtheoutside”. Thepresentmodelforquasi-transparency can beseen as a refinement and extension of someaspects of thesystemsuggestedbyEgré - Dokic.
CONCLUSIONS • Once weclarifysome conceptual aspects of higher-orderprobabilities… • …weobtainthevindication of a number of introspectiveprinciples, orprinciples of quasi-transparency.
CONCLUSIONS • Quasi-transparency principles were not just assumed to hold, but they have been obtained as a result of implementing a number of natural constraints on the structure of the system. Formally speaking, they behave quite differently from presuppositions of consistency or deductive closure.
CONCLUSIONS • Theframeworkvindicatestheintuitionthatfirst- and second-orderknowledgediffersubstantially: • Differentattitudesaboutignorance • Differentattitudestoward “margin of error” principles • Second-orderknowledgeisconcernedwiththe “ratification” of first-orderattitudes.
CONCLUSIONS • Quasi-transparencyfullyvindicatesthenormative link betweenself-knowledge and responsibility. K+Kp: “responsibleknowledge” of p.
CONCLUSIONS Second-orderknowledge, as a state of epistemicresponsibility, is a desideratumwehavequa agents.