Belief Revision and Truth-Finding

Transcription

1 Belief Revision and Truth-Finding Kevin T. Kelly Department of Philosophy Carnegie Mellon University

2 Further Reading (with O. Schulte and V. Hendricks) Reliable Belief Revision, in Logic and Scientific Methods, Dordrecht: Kluwer,, The Learning Power of Iterated Belief Revision, in Proceedings of the Seventh TARK Conference,, Iterated Belief Revision, Reliability, and Inductive Amnesia, Erkenntnis,, 50: 1998

3 The Idea Belief revision theory... rational belief change Learning theory......reliable belief change Conflict? Truth

4 Part I Iterated Belief Revision

5 Baian (Vanilla) Updating Propositional epistemic state B

6 Baian (Vanilla) Updating New belief is intersection Perfect memory No inductive leaps new evidence E B

7 Baian (Vanilla) Updating New belief is intersection Perfect memory No inductive leaps E B B

8 Baian (Vanilla) Updating New belief is intersection Perfect memory No inductive leaps E B B

9 Epistemic Hell (a.k.a. Nirvana) B

10 Epistemic Hell (a.k.a. Nirvana) E Surprise! B

11 Epistemic Hell (a.k.a. Nirvana) Scientific revolutions Suppositional reasoning Conditional pragmatics Decision theory Game theory Data bases E B Epistemic hell

12 Ordinal Epistemic States Spohn 88 Ordinal-valued degrees of implausibility Belief state is bottom level ω + 1 ω 2 1 b (S) 0 S

13 Iterated Belief Revision epistemic state trajectory S 0 initial state input propositions E 0 E 1 E 2 *

14 Iterated Belief Revision epistemic state trajectory S 0 S 1 input propositions E 1 E 2 *

15 Iterated Belief Revision epistemic state trajectory S 0 S 1 input propositions E 1 E 2 *

16 Iterated Belief Revision epistemic state trajectory S 0 S 1 S 2 input proposition E 2 *

17 Iterated Belief Revision epistemic state trajectory S 0 S 1 S 2 input proposition E 2 *

18 Iterated Belief Revision epistemic state trajectory S 0 S 1 S 2 S 3 *

19 Iterated Belief Revision epistemic state trajectory S 0 S 1 S 2 S 3 b (S 0 ) b (S 1 ) b (S 2 ) b (S 3 ) belief state trajectory *

20 Generalized Conditioning *C* Spohn 88 S

21 Generalized Conditioning *C* Spohn 88 Condition entire epistemic state E S

22 Generalized Conditioning *C* Spohn 88 Condition entire epistemic state E *C S B S *C E

23 Generalized Conditioning *C* Spohn 88 Condition entire epistemic state Perfect memory Inductive leaps No epistemic hell if evidence sequence is consistent E *C S B S *C E

24 Lexicographic Updating *L* Spohn 88, Nayak 94 S

25 Lexicographic Updating *L* Spohn 88, Nayak 94 Lift refuted possibilities above non-refuted possibilities preserving order. E S

26 Lexicographic Updating *L* Spohn 88, Nayak 94 Lift refuted possibilities above non-refuted possibilities preserving order. E *L S B S *L E

27 Lexicographic Updating *L* Spohn 88, Nayak 94 Lift refuted possibilities above non-refuted possibilities preserving order. Perfect memory on consistent data sequences Inductive leaps No epistemic hell E *L B S S *L E

28 Minimal or Natural Updating *M* Spohn 88, Boutilier 93 S

29 Minimal or Natural Updating *M* Spohn 88, Boutilier 93 Drop the lowest possibilities consistent with the data to the bottom and raise everything else up one notch E S

30 Minimal or Natural Updating *M* Spohn 88, Boutilier 93 Drop the lowest possibilities consistent with the data to the bottom and raise everything else up one notch E *M S S *M E

31 Minimal or Natural Updating *M* Spohn 88, Boutilier 93 Drop the lowest possibilities consistent with the data to the bottom and raise everything else up one notch inductive leaps No epistemic hell E *M S S *M E

32 The Flush-to-α Method *F,* F,α Goldszmidt and Pearl 94 S

33 The Flush-to-α Method *F,* F,α Goldszmidt and Pearl 94 Send non-e worlds to α and drop E -worlds rigidly to the bottom boost parameter α E S

34 The Flush-to-α Method *F,* F,α Goldszmidt and Pearl 94 Send non-e worlds to α and drop E -worlds rigidly to the bottom α E *F,α S S *F,α E

35 The Flush-to-α Method *F,* F,α Goldszmidt and Pearl 94 Send non-e worlds to α and drop E -worlds rigidly to the bottom Perfect memory on sequentially consistent data if α is high enough Inductive leaps No epistemic hell E *F,α α S S *F,α E

36 Ordinal Jeffrey Conditioning *J,* J,α Spohn 88 S

37 Ordinal Jeffrey Conditioning *J,* J,α Spohn 88 E S

38 Ordinal Jeffrey Conditioning *J,* J,α Spohn 88 Drop E worlds to the bottom. Drop non-e worlds to the bottom and then jack them up to level α E α S

39 Ordinal Jeffrey Conditioning *J,* J,α Spohn 88 Drop E worlds to the bottom. Drop non-e worlds to the bottom and then jack them up to level α E *J,α α S S *J,α E

40 Ordinal Jeffrey Conditioning *J,* J,α Spohn 88 Drop E worlds to the bottom. Drop non-e worlds to the bottom and then jack them up to level α Perfect memory on consistent sequences if α is large enough No epistemic hell But... E S α *J,α S *J,α E

41 Empirical Backsliding Ordinal Jeffrey conditioning can increase the plausibility of a refuted possibility E α

42 The Ratchet Method *R,* R,α Darwiche and Pearl 97 S

43 The Ratchet Method *R,* R,α Darwiche and Pearl 97 Like ordinal Jeffrey conditioning except refuted possibilities move up by α from their current positions β β + α E S

44 The Ratchet Method *R,* R,α Darwiche and Pearl 97 Like ordinal Jeffrey conditioning except refuted possibilities move up by α from their current positions β β + α E *R,α B S B S *R,α E

45 The Ratchet Method *R,* R,α Darwiche and Pearl 97 Like ordinal Jeffrey conditioning except refuted possibilities move up by α from their current positions Perfect memory if α is large enough Inductive leaps No epistemic hell E B *R,α B β β + α S S *R,α E

46 Part II Properties of the Methods

47 Timidity and Stubbornness Timidity: no inductive leaps without refutation. Stubbornness: no retractions without refutation Examples: all the above Nutty! B B

48 Timidity and Stubbornness Timidity: no inductive leaps without refutation. Stubbornness: no retractions without refutation Examples: all the above Nutty! B B

49 Timidity and Stubbornness Timidity: no inductive leaps without refutation. Stubbornness: no retractions without refutation Examples: all the above Nutty! B B

50 Local Consistency Local consistency: new belief must be consistent with the current consistent datum Examples: all the above

51 Positive Order-invariance Positive order-invariance: preserve original ranking inside conjunction of data Examples: * C, * L, * R, α, * J, J, α.

52 Data-Precedence Data-precedence: Each world satisfying all the data is placed above each world failing to satisfy some datum. Examples: * C, * L * R, α, * J, α, if α is above S. S

53 Enumerate and Test Enumerate-and-test: locally consistent, positively invariant data-precedent Examples: * C, * L * R, α, * J, α, if α is above S. epistemic dump for refuted possibilities preserved implausibility structure

54 Part III Belief Revision as Learning

55 A Very Simple Learning Paradigm data trajectory mysterious system

56 A Very Simple Learning Paradigm data trajectory mysterious system

57 A Very Simple Learning Paradigm data trajectory mysterious system

58 A Very Simple Learning Paradigm data trajectory mysterious system

59 Possible Outcome Trajectories possible data trajectories e e n

60 Finding the Truth (*, S 0 ) identifies e for all but finitely many n, b(s 0 * ([0, e(0)],..., [n,[ e(n)])) = {e}{

61 Finding the Truth (*, S 0 ) identifies e for all but finitely many n, b(s 0 * ([0, e(0)],..., [n,[ e(n)]) = {e}{ truth

62 Finding the Truth (*, S 0 ) identifies e for all but finitely many n, b(s 0 * ([0, e(0)],..., [n,[ e(n)]) = {e}{ truth

63 Finding the Truth (*, S 0 ) identifies e for all but finitely many n, b(s 0 * ([0, e(0)],..., [n,[ e(n)]) = {e}{ truth

64 Finding the Truth (*, S 0 ) identifies e for all but finitely many n, b(s 0 * ([0, e(0)],..., [n,[ e(n)]) = {e}{ truth

65 Finding the Truth (*, S 0 ) identifies e for all but finitely many n, b(s 0 * ([0, e(0)],..., [n,[ e(n)]) = {e}{ completely true belief

66 Reliability is No Accident Let K be a range of possible outcome trajectories (*, S 0 ) identifies K (*, S 0 ) identifies each e in K. Fact: K is identifiable K is countable.

67 Completeness * is complete for each identifiable K there is an S 0 such that, K is identifiable by (*,( S 0 ). Else * is restrictive.

68 Completeness Proposition: If * enumerates and tests, * is complete.

69 Completeness Proposition: If * enumerates and tests, * is complete. Enumerate K Choose arbitrary e in K e

70 Completeness Proposition: If * enumerates and tests, * is complete.

71 Completeness Proposition: If * enumerates and tests, * is complete. data precedence positive invariance

72 Completeness Proposition: If * enumerates and tests, * is complete.

73 Completeness Proposition: If * enumerates and tests, * is complete. data precedence positive invariance

74 Completeness Proposition: If * enumerates and tests, * is complete.

75 Completeness Proposition: If * enumerates and tests, * is complete. data precedence convergence local consistency

76 Amnesia Without data precedence, memory can fail Same example, using * J,1.

77 Amnesia Without data precedence, memory can fail Same example, using * J,1.

78 Amnesia Without data precedence, memory can fail Same example, using * J,1.

79 Amnesia Without data precedence, memory can fail Same example, using * J,1.

80 Amnesia Without data precedence, memory can fail Same example, using * J,1.

81 Amnesia Without data precedence, memory can fail Same example, using * J,1.

82 Amnesia Without data precedence, memory can fail Same example, using * J,1.

83 Amnesia Without data precedence, memory can fail Same example, using * J,1. E E is forgotten

84 Duality conjectures and refutations tabula rasa... predicts may forget remembers doesn t predict

85 Rationally Imposed Tension compression for memory Can both be accommodated? rarefaction for inductive leaps

86 Inductive Amnesia compression for memory Bang! Restrictiveness: No possible initial state resolves the pressure rarefaction for inductive leaps

87 Question Which methods are guilty? Are some worse than others?

88 Part IV: The Goodman Hierarchy

89 The Grue Operation Nelson Goodman n e e n

90 Grue Complexity Hierarchy G ω (e) finite variants of e, e finite variants of e G ω even (e) G 4 (e) G 3 (e) G 2 (e) G 1 (e) G 0 (e) G 2 even (e) G 1 even (e) G 0 even (e)

91 Classification: even grues G ω even (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 1 α = 1 G n even (e) no α = n +1 α = 1 α = 1 G 2 even (e) G 1 even (e) G 0 even (e) no no α = 3 α = 2 α = 0 α = 1 α = 1 α = 1 α = 1 α = 0 α = 0

92 Classification: even grues G ω even (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 1 α = 1 G n even (e) no α = n +1 α = 1 α = 1 G 2 even (e) G 1 even (e) G 0 even (e) no no α = 3 α = 2 α = 0 α = 1 α = 1 α = 1 α = 1 α = 0 α = 0

93 Hamming Algebra a H b mod e a differs from e only where b does. Hamming

94 *R,1,*J,1 J,1 can identify G ω even even (e) Example a e a Learning as rigid hypercube rotation e

95 *R,1,*J,1 J,1 can identify G ω even even (e) a Learning as rigid hypercube rotation e

96 *R,1,*J,1 J,1 can identify G ω even even (e) e Learning as rigid hypercube rotation a

97 *R,1,*J,1 J,1 can identify G ω even even (e) e Learning as rigid hypercube rotation convergence a

98 Classification: even grues G ω even (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 1 α = 1 G n even (e) no α = n +1 α = 1 α = 1 G 2 even (e) G 1 even (e) G 0 even (e) no no α = 3 α = 2 α = 0 α = 1 α = 1 α = 1 α = 1 α = 0 α = 0

99 Classification: arbitrary grues G ω (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 2 α = 2 G 3 (e) no α = n +1 α = 2 α = 2 G 2 (e) no α = 3 α = 2 α = 2 G 1 (e) G 0 (e) no α = 2 α = 2 α = 1 α = 0 α = 0 α = 0

100 Classification: arbitrary grues G ω (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 2 α = 2 G 3 (e) no α = n +1 α = 2 α = 2 G 2 (e) no α = 3 α = 2 α = 2 G 1 (e) G 0 (e) no α = 2 α = 2 α = 1 α = 0 α = 0 α = 0

101 *R,2 is Complete Impose the Hamming distance ranking on each finite variant class Now raise the nth Hamming ranking by n S C 0 C 1 C 2 C 3 C 4

102 *R,2 is Complete Data streams in the same column just barely make it because they jump by 2 for each difference from the truth S 1 difference from truth 2 differences from truth C 0 C 1 C 2 C 3 C 4

103 Classification: arbitrary grues G ω (e) G 3 (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 2 α = 2 no Can t use Hamming rank α = n +1 α = 2 α = 2 G 2 (e) no α = 3 α = 2 α = 2 G 1 (e) G 0 (e) no α = 2 α = 2 α = 1 α = 0 α = 0 α = 0

104 Wrench In the Works Suppose * J,2 succeeds with Hamming rank. Feed e until it is uniquely at the bottom. k e By convergent success

105 Wrench In the Works So for some later n, a b k n Hamming rank and positive invariance. e If empty, things go even worse! Still alone since timid and stubborn

106 Wrench In the Works b moves up at most 1 step since e is still alone (rule) a k n b e Refuted worlds touch bottom and get lifted by at most two.

107 Wrench In the Works So b never rises above a when a is true (positive invariance) Now a and b agree forever, so can never be separated. So never converges in a or forgets refutation of b. k n a b e a

108 Hamming vs. Goodman Algebras a H b mod e a differs from e only where b does. a G b mod e a grues e only where b does. Hamming Goodman

109 Epistemic States as Boolean Ranks Hamming Goodman G ω odd (e) G ω even (e) G ω (e) e e

110 *J,2 can identify G ω (e) Proof: Use the Goodman ranking as initial state Then * J,2 always believes that the observed grues are the only ones that will ever occur. Note: Ockham with respect to reversal counting problem.

111 Classification: arbitrary grues G ω (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 2 α = 2 G 3 (e) no α = n +1 α = 2 α = 2 G 2 (e) no α = 3 α = 2 α = 2 G 1 (e) G 0 (e) no α = 2 α = 2 α = 1 α = 0 α = 0 α = 0

112 Methods *J,1* J,1; ; *M* Fail on G 1 (e) Proof: Suppose otherwise Feed e until e is uniquely at the bottom e data so far

113 Methods *J,1* J,1; ; *M* Fail on G 1 (e) By the well-ordering condition,...else infinite descending chain e data so far

114 Methods *J,1* J,1; ; *M* Fail on G 1 (e) Now feed e forever By stage n, the picture is the same e e positive order invariance e e n timidity and stubbornness

115 Methods *J,1* J,1; ; *M* Fail on G 1 (e) e e At stage n +1, e stays at the bottom (timid and stubborn). So e can t t travel down (rule) e doesn t t rise (rule) Now e makes it to the bottom at least as soon as e e e n

116 Classification: arbitrary grues G ω (e) Min Flush Jeffrey Ratch Lex Cond no α = ω α = 2 α = 2 G 3 (e) no α = n +1 α = 2 α = 2 G 2 (e) no α = 3 forced backsliding α = 2 α = 2 G 1 (e) G 0 (e) no α = 2 α = 2 α = 1 α = 0 α = 0 α = 0

117 Method *R,1* Fails on G 2 (e) with Oliver Schulte Proof: : Suppose otherwise Bring e uniquely to the bottom, say at stage k k e

118 Method *R,1* Fails on G 2 (e) Start feeding a = e k with Oliver Schulte k e a

119 Method *R,1* Fails on G 2 (e) with Oliver Schulte By some stage k, a is uniquely down So between k + 1 and k,, there is a first stage j when no finite variant of e is at the bottom k k e a

120 Method *R,1* Fails on G 2 (e) with Oliver Schulte Let c in G 2 (e ) be a finite variant of e that rises to level 1 at j k j k c a

121 Method *R,1* Fails on G 2 (e) with Oliver Schulte Let c in G 2 (e ) be a finite variant of e that rises to level 1 at j k j k c a

122 Method *R,1* Fails on G 2 (e) with Oliver Schulte k j k So c(j - 1) is not a(j - 1) c a

123 Method *R,1* Fails on G 2 (e) with Oliver Schulte k j k Let d be a up to j and e thereafter So is in G 2 (e) Since d differs from e, d is at least as high as level 1 at j 1 d c a

124 Method *R,1* Fails on G 2 (e) with Oliver Schulte Show: c agrees with e after j. k j k 1 d c a

125 Method *R,1* Fails on G 2 (e) with Oliver Schulte k j k Case: j = k+1 Then c could have been chosen as e since e is uniquely at the bottom at k 1 d c a

126 Method *R,1* Fails on G 2 (e) with Oliver Schulte Case: j > k+1 k j k Then c wouldn t t have been at the bottom if it hadn t t agreed with a (disagreed with e) 1 d c a

127 Method *R,1* Fails on G 2 (e) with Oliver Schulte k j k Case: j > k+1 So c has already used up its two grues against e 1 d c a

128 Method *R,1* Fails on G 2 (e) with Oliver Schulte Feed c forever after k j k By positive invariance, either never projects or forgets the refutation of c at j-1 1 d c d

129 Without Well-Ordering G ω (e) Min Flush Jeffrey Ratch Lex Cond no G 3 (e) no infinite descending chains can help! G 2 (e) no G 1 (e) G 0 (e)

130 Summary Belief revision constrains possible inductive strategies No induction without contradiction (?!!) Rationality weakens learning power of ideal agents. Prediction vs. memory Precise recommendations for rationalists: boosting by 2 vs.. 1 backslide vs.. ratchet well-ordering Hamming vs. Goodman rank