Tutorial: Nonmonotonic Logic

Similar documents
Tutorial: Nonmonotonic Logic (Day 1)

Reasoning by Cases in Structured Argumentation.

Tackling Defeasible Reasoning in Bochum:

Tutorial: Nonmonotonic Logic (Day 2)

Outline. 1 Plausible Reasoning. 2 Preferential / Selection Semantics (KLM, Shoham) 3 Bibliography

Outline. Golden Rule

Handout Lecture 8: Non-monotonic logics

Nonmonotonic Logic. Daniel Bonevac. April 16, Accident victims who are cool to the touch may be in shock.

Argumentation and rules with exceptions

A Brief Introduction to Nonmonotonic Reasoning

Argumentation-Based Models of Agent Reasoning and Communication

ESSENCE 2014: Argumentation-Based Models of Agent Reasoning and Communication

On the Semantics of Simple Contrapositive Assumption-Based Argumentation Frameworks

A Theorem Prover for Prioritized Circumscription

Introduction to Structured Argumentation

Belief revision: A vade-mecum

Argumentative Characterisations of Non-monotonic Inference in Preferred Subtheories: Stable Equals Preferred

Introduction to Structural Argumentation

ESSLLI 2007 COURSE READER. ESSLLI is the Annual Summer School of FoLLI, The Association for Logic, Language and Information

Commonsense Reasoning and Argumentation

The Generation and Evaluation of Generic Sentences

IS HUMAN REASONING REALLY NONMONOTONIC?

Simulating Human Inferences in the Light of New Information: A Formal Analysis

On the Relationship of Defeasible Argumentation and Answer Set Programming

4 ENTER Adaptive Logics

SEMANTICAL CONSIDERATIONS ON NONMONOTONIC LOGIC. Robert C. Moore Artificial Intelligence Center SRI International, Menlo Park, CA 94025

A modal perspective on defeasible reasoning

DEFAULTS WITH PRIORITIES

Outline. Adaptive Logics. Introductory Remarks (2) Introductory Remarks (1) Incomplete Survey. Introductory Remarks (3)

Non-monotonic Logic I

INTRODUCTION TO NONMONOTONIC REASONING

On the Semantics of Simple Contrapositive Assumption-Based Argumentation Frameworks

(A 3 ) (A 1 ) (1) COMPUTING CIRCUMSCRIPTION. Vladimir Lifschitz. Department of Computer Science Stanford University Stanford, CA

A Propositional Typicality Logic for Extending Rational Consequence

General Patterns for Nonmonotonic Reasoning: From Basic Entailments to Plausible Relations

Adaptive Logics. p p. Manuel Bremer Centre for Logic, Language and Information

A Tableaux Calculus for KLM Preferential and Cumulative Logics

Revising Nonmonotonic Theories: The Case of Defeasible Logic

Weak Completion Semantics

ABDUCTIVE LOGICS IN A BELIEF REVISION FRAMEWORK

Inductive, Abductive and Pragmatic Reasoning

Resolutions in Structured Argumentation

CogSysI Lecture 9: Non-Monotonic and Human Reasoning

The logical meaning of Expansion

Reasoning with Inconsistent and Uncertain Ontologies

The Modal Logic S4F, the Default Logic, and the Logic Here-and-There

Developing Default Logic as a Theory of Reasons. John Horty. University of Maryland

Deciding Consistency of Databases Containing Defeasible and Strict Information*

The Cause and Treatments of Floating Conclusions and Zombie Paths *

Default Logic Autoepistemic Logic

Combining Inductive Generalization and Factual Abduction

Adaptively Applying Modus Ponens in Conditional Logics of Normality

A Review of Argumentation Based on Deductive Arguments

arxiv: v1 [cs.ai] 16 Aug 2011

Propositional Logics and their Algebraic Equivalents

Artificial Intelligence 52 (1991) Elsevier

Contamination in Formal Argumentation Systems

A Unifying Semantics for Belief Change

Defeasible logic versus Logic Programming without Negation as Failure

Argumentation among Agents

logic is everywhere Logik ist überall Hikmat har Jaga Hai Mantık her yerde la logica è dappertutto lógica está em toda parte

2. The Logic of Compound Statements Summary. Aaron Tan August 2017

MAKING ARGUMENT SYSTEMS COMPUTATIONALLY ATTRACTIVE Argument Construction and Maintenance 1

KRR. Example. Example (formalization) Example (cont.) Knowledge representation & reasoning

Revisiting Unrestricted Rebut and Preferences in Structured Argumentation.

Abstract Rule-Based Argumentation

Nested Epistemic Logic Programs

Which Style of Reasoning to Choose in the Face of Conflicting Information?

Argumentation with Abduction

Artificial Intelligence. Propositional logic

Reasoning under inconsistency: the forgotten connective

Section 5 Circumscription. Subsection 5.1 Introducing Circumscription. TU Dresden, WS 2017/18 Introduction to Nonmonotonic Reasoning Slide 128

Nonmonotonic Tools for Argumentation

Break the Weakest Rules

Kybernetika. Niki Pfeifer; Gernot D. Kleiter Inference in conditional probability logic. Terms of use: Persistent URL:

Knowledge Representation (Overview)

Lecture 2. Logic Compound Statements Conditional Statements Valid & Invalid Arguments Digital Logic Circuits. Reading (Epp s textbook)

Dialectical Frameworks: Argumentation Beyond Dung

Pei Wang( 王培 ) Temple University, Philadelphia, USA

Maximal ideal recursive semantics for defeasible argumentation

Combining Modes of Reasoning: an Application of Abstract Argumentation

Chapter 13 Uncertainty

Reasoning: From Basic Entailments. to Plausible Relations. Department of Computer Science. School of Mathematical Sciences. Tel-Aviv University

Propositional Logic: Part II - Syntax & Proofs 0-0

Chapter 3: Propositional Calculus: Deductive Systems. September 19, 2008

WUCT121. Discrete Mathematics. Logic. Tutorial Exercises

Argumentation for Propositional Logic and Nonmonotonic Reasoning

First-Degree Entailment

Preference-based Argumentation

Defeasible Inheritance-Based Description Logics

Partial Meet Revision and Contraction in Logic Programs

3. The Logic of Quantified Statements Summary. Aaron Tan August 2017

The Conditional in Mental Probability Logic

Přednáška 12. Důkazové kalkuly Kalkul Hilbertova typu. 11/29/2006 Hilbertův kalkul 1

REMARKS ON INHERITANCE SYSTEMS

Propositional logic (revision) & semantic entailment. p. 1/34

An assumption-based framework for. Programming Systems Institute, Russian Academy of Sciences

EQUIVALENCE OF THE INFORMATION STRUCTURE WITH UNAWARENESS TO THE LOGIC OF AWARENESS. 1. Introduction

A Preference Logic With Four Kinds of Preferences

Reasoning in Uncertain Situations

Transcription:

Tutorial: Nonmonotonic Logic PhDs in Logic (2017) Christian Straßer May 2, 2017

Outline Defeasible Reasoning Scratching the Surface of Nonmonotonic Logic 1/52

Defeasible Reasoning

What is defeasible reasoning? 1/52

2/52

abductive inference infer a (good!?) explanation α β and β thus α 2/52

3/52

closed world assumption reasoning on the assumption that the given information is complete 3/52

default inferences Tweety is a bird. Thus,...? 4/52

5/52

inductive generalisations 5/52

Domains of defeasible reasoning everyday reasoning expert reasoning (e.g. medical diagnosis) scientific reasoning 6/52

Commonalities tentative conclusions jumping to conclusions retraction possible if problems arise 7/52

Commonalities tentative conclusions jumping to conclusions retraction possible if problems arise Two tiers of defeasible reasoning 1. illative tier (support, concluding) 2. dialectic tier (retraction) 7/52

Pessimism in the 60ies Toulmin in The Uses of Argument, 1958 8/52

Toulmin Scheme (Toulmin, 1958) Premises Conclusion Backing Warrant Defeat 9/52

Nonmonotonic Logic to the Rescue Artificial Intelligence, Volume 13, Issues 1 2, Pages 1-174,(April 1980), Special Issue on Non-Monotonic Logic 10/52

Nonmonotonic Logic to the Rescue Artificial Intelligence, Volume 13, Issues 1 2, Pages 1-174,(April 1980), Special Issue on Non-Monotonic Logic capture defeasible reasoning in a mathematically precise way reproduce the success of CL in the domain of mathematical reasoning in the less sterile/idealized domain of defeasible reasoning where incompleteness 10/52

Nowadays cooperations between formal and informal logicians e.g., Douglas Walton and formal argumentation (e.g., Gordon, Prakken and Walton, Artificial Intelligence, 2007) shift of normative standards in cognitive science, Stenning and Van Lambalgen, 2008, MIT Press Pfeifer, Studia Logica, 2014 11/52

12/52

13/52

Monotonicity If A 1,..., A n B then A 1,..., A n, A n+1 B. 14/52

Monotonicity If A 1,..., A n B then A 1,..., A n, A n+1 B. Premises/Input A 1,..., A n, Logic Conclusions/Output B 1,..., B m, B 14/52

Monotonicity If A 1,..., A n B then A 1,..., A n, A n+1 B. Premises/Input A 1,..., A n, A Logic Conclusions/Output B 1,..., B m, B 14/52

Cautious Monotonicity If A 1,..., A n B and A 1,..., A n C, then A 1,..., A n, B C. 15/52

Cautious Monotonicity If A 1,..., A n B and A 1,..., A n C, then A 1,..., A n, B C. Premises/Input A 1,..., A n Logic Conclusions/Output B, C 15/52

Cautious Monotonicity If A 1,..., A n B and A 1,..., A n C, then A 1,..., A n, B C. Premises/Input A 1,..., A n Logic Conclusions/Output B, C 15/52

Rational Monotonicity If it is not the case that A 1,..., A n B, and moreover A 1,..., A n C, then A 1,..., A n, B C. 16/52

Rational Monotonicity If it is not the case that A 1,..., A n B, and moreover A 1,..., A n C, then A 1,..., A n, B C. Premises/Input A 1,..., A n, Logic Conclusions/Output C B My beliefs are robust/cumulative under adding consistent information. 16/52

Rational Monotonicity If it is not the case that A 1,..., A n B, and moreover A 1,..., A n C, then A 1,..., A n, B C. Premises/Input A 1,..., A n, B Logic Conclusions/Output C B My beliefs are robust/cumulative under adding consistent information. 16/52

The nonmononotonic Zoo See: Straßer, C., & Antonelli, G. A. (2014). Non-monotonic logic. In E. N. Zalta (Eds.), The Stanford Encyclopedia of Philosophy inheritance networks default logic logic programming autoepistemic logic circumscription preferential semantics maximal consistent subset approaches adaptive logics formal argumentation etc. 17/52

Scratching the Surface of Nonmonotonic Logic

In the following we will work with defeasible theories consisting of 1. non-defeasible information F including strict rules 2. defeasible information D including defeasible rules How exactly these defeasible theories are expressed in terms of a formal language depends on the underlying formalism. 18/52

Default logic A B: B follows defeasibly from A. 19/52

Default logic A B: B follows defeasibly from A. Example: (slightly enhanced) Nixon Diamond facts: {Nixon Republican, Nixon Quaker} defeasible information: {Republican pacifist, Quaker pacifist, Quaker pro-syndicate, Republican pro-gun} pro-gun Nixon is a is a Republican Quaker Pacifist Pacifist pro-syndicate 19/52

pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate Idea: Apply iteratively modus ponens to defaults while preserving consistency. This way build step-wise an extension (sets of beliefs that are obtained in this way). This may lead to several choices. 20/52

Basic Idea: Extension 1 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 21/52

Basic Idea: Extension 1 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 21/52

Basic Idea: Extension 1 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 21/52

Basic Idea: Extension 1 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 21/52

Basic Idea: Extension 1 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 21/52

Basic Idea: Extension 1 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 21/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 22/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 22/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 22/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 22/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 22/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate 22/52

Basic Idea: Extension 2 pro-gun is a Republican Pacifist Nixon is a Quaker Pacifist pro-syndicate There are two extensions: Cn({pacifist, Nixon, Republican, Quaker, pro-syndicate, pro-gun}) Cn({ pacifist, Nixon, Republican, Quaker, pro-syndicate, pro-gun}). 22/52

Default consequences Question What to derive from a defeasible theory? Answer(s) build all extensions of the theory Sceptical reasoner: derive what is in the intersection of all extensions e.g., do not derive pacifist from the Nixon diamond Credulous reasoner: derive what is in the union of all extensions e.g., derive both pacifist and pacifist from the Nixon diamond mapping out rational choices 23/52

Floating Conclusions Nixon Republican Quaker Hawk Dove political where Hawk Dove and Dove Hawk What are the extensions? Does political follow? 24/52

Floating Conclusions: Extension 1 Nixon Republican Quaker Hawk Dove political {Nixon, Republican, Hawk, Quaker, political, Dove} 25/52

Floating Conclusions: Extension 2 Nixon Republican Quaker Hawk Dove political {Nixon, Republican, Hawk, Quaker, political, Dove} 26/52

Floating Conclusions: Extension 3 Nixon Republican Quaker Hawk Dove political {Nixon, Republican, Hawk, Quaker, political, Dove} 27/52

Reasoning by Cases no handling of disjunctive facts out-of-the-box for instance: Σ = {Republican Democrat, Republican political, Democrat political}.? Republican Republican Democrat? political Democrat since the default is not triggered by the fact, MP cannot be applied 28/52

Extension-based Approaches: Default Logic (Reiter) idea: split the factual part of the knowledge base (Gelfond, Lifschitz, Przymusinska, 1991) Republican Republican Base 1 Republican Democrat political Base 2 Democrat Democrat 29/52

Extension-based Approaches: Default Logic (Reiter) idea: split the factual part of the knowledge base (Gelfond, Lifschitz, Przymusinska, 1991) Republican Republican Base 1 Republican Democrat political Base 2 Democrat Democrat two extensions: 1. Republican, political 2. Democrat, political 29/52

Problematic Example for Disjunctive Defaults Consider the following example: 1. Either his left hand or his right hand is broken. lhb rhb 30/52

Problematic Example for Disjunctive Defaults Consider the following example: 1. Either his left hand or his right hand is broken. lhb rhb 2. If somebody writes legibly then usually the right hand is not broken. wl rhb 30/52

Problematic Example for Disjunctive Defaults Consider the following example: 1. Either his left hand or his right hand is broken. lhb rhb 2. If somebody writes legibly then usually the right hand is not broken. wl rhb 3. He writes legibly. wl 30/52

Problematic Example for Disjunctive Defaults Consider the following example: 1. Either his left hand or his right hand is broken. lhb rhb 2. If somebody writes legibly then usually the right hand is not broken. wl rhb 3. He writes legibly. wl With disjunctive default logic we get two extensions: 1. wl, rhb, lhb 2. wl, rhb 30/52

Different Idea Meta-rule for defaults: If A B and C B then A C B. OR 31/52

A Problematic Example for OR Suppose we have Σ = {p q r, q s, s v, r u, u v, p}. q s v p q r v r u v 32/52

A Problematic Example for OR q s v p q r v r u v 33/52

A Problematic Example for OR q s v p q r s u v r u v by (OR): from s v and u v 33/52

A Problematic Example for OR q s v p q r s u v r u v by (OR): from s v and u v by (Right-Weakening), from q s and r u 33/52

A Problematic Example for OR q s v p q r s u v r u v by (OR): from s v and u v by (Right-Weakening), from q s and r u by (OR): from q s u and r s u 33/52

A Problematic Example for OR t s! q s v p q r s u v r u v Suppose now we also have t and t s. the possible defeater has no effect on the generalized path 34/52

A Problematic Example for OR t s! q s v p q r s u v t! r r u v Suppose now we also have t and t r. the additional possible defeater has no effect on the generalized path 35/52

Greedy Reasoning (Conclusion maximizing) Best-candidate job new-car new-car Can we derive job? 36/52

Problem with Cautious Monotonicity Birthday party: we know Will comes and that (Will Diane Eli Peter): Will Diane Eli Peter 37/52

Problem with Cautious Monotonicity Birthday party: we know Will comes and that (Will Diane Eli Peter): Will Diane Eli Peter If we add Eli to the facts: we also have the following extension: Will Diane Eli Peter 37/52

Another approach: adaptive logics / default assumptions Model defeasible rules as strict rules with defeasible assumptions: Will Diane becomes Will π 1 Diane Diane Eli becomes Diane π 2 Eli Eli Peter becomes Eli π 3 Peter. where π 1, π 2, π 3 are normality assumptions which are assumed to be true as much as possible. π 1 π 2 π 3 Will Diane Eli Peter 38/52

Another approach: adaptive logics / default assumptions π 1 π 2 π 3 Will Diane Eli Peter since we know that not all of them are at the party, it cannot be that all π 1, π 2 and π 3 are true. 39/52

Another approach: adaptive logics / default assumptions π 1 π 2 π 3 Will Diane Eli Peter since we know that not all of them are at the party, it cannot be that all π 1, π 2 and π 3 are true. We have the following options: {π 1, π 2 } {π 1, π 3 } {π 2, π 3 } {π 1 } {π 2 } {π 3 } 39/52

Another approach: adaptive logics / default assumptions π 1 π 2 π 3 Will Diane Eli Peter since we know that not all of them are at the party, it cannot be that all π 1, π 2 and π 3 are true. We have the following options: {π 1, π 2 } {π 1, π 3 } {π 2, π 3 } {π 1 } {π 2 } {π 3 } the interpretations in the lowest row are not maximally normal 39/52

Another approach: adaptive logics / default assumptions π 1 π 2 π 3 Will Diane Eli Peter since we know that not all of them are at the party, it cannot be that all π 1, π 2 and π 3 are true. We have the following options: {π 1, π 2 } {π 1, π 3 } {π 2, π 3 } {π 1 } {π 2 } {π 3 } the interpretations in the lowest row are not maximally normal 39/52

Another approach: adaptive logics / default assumptions π 1 π 2 π 3 Will Diane Eli Peter since we know that not all of them are at the party, it cannot be that all π 1, π 2 and π 3 are true. We have the following options: {π 1, π 2 } {π 1, π 3 } {π 2, π 3 } {π 1 } {π 2 } {π 3 } the interpretations in the lowest row are not maximally normal 39/52

Contraposition π 1 π 2 π 3 Will Diane Eli Peter Note that in this approach defeasible rules are contrapositable! 40/52

Contraposition π 1 π 2 π 3 Will Diane Eli Peter Note that in this approach defeasible rules are contrapositable! π 1 π 2 π 3 Will Diane Eli Peter 40/52

Consequence Relation Given a set of premises Σ and a set of normality assumptions Π we define: Definition Ξ MCS(Σ, Π) iff 1. Ξ Π 2. Ξ Σ is consistent 3. for every Ξ Π, if Ξ Ξ then Ξ Σ is inconsistent. Definition Σ Π A iff A Ξ MCS(Σ,Π) Cn(Ξ Σ). 41/52

Preferential Semantics Let M(Σ) be the set of all models of Σ. We order them as follows: M Π M iff Π(M ) Π(M). Define: Σ Π A iff M = A for all M min Π (Σ) 42/52

Preferential Semantics Let M(Σ) be the set of all models of Σ. We order them as follows: M Π M iff Π(M ) Π(M). Define: Σ Π A iff M = A for all M min Π (Σ) Sometimes the perspective is inverted: instead of assuming normality assumptions to be true, we assume abnormality assumptions to be false. So, now Π contains abnormalities! 42/52

Preferential Semantics Let M(Σ) be the set of all models of Σ. We order them as follows: M Π M iff Π(M ) Π(M). Define: Σ Π A iff M = A for all M min Π (Σ) Sometimes the perspective is inverted: instead of assuming normality assumptions to be true, we assume abnormality assumptions to be false. So, now Π contains abnormalities! Let M(Σ) be the set of all models of Σ. We order them as follows: M Π M iff Π(M) Π(M ). Define: Σ Π A iff M = A for all M min Π (Σ) 42/52

An Application: Inconsistency-adaptive logics in CLuN all connectives have the same truth-tables as in classical logic, just the table for negation is indeterministic: A A 0 1 1 [0/1] 43/52

An Application: Inconsistency-adaptive logics in CLuN all connectives have the same truth-tables as in classical logic, just the table for negation is indeterministic: A A 0 1 1 [0/1] note that A, A CLuN B. The logic is paraconsistent. 43/52

An Application: Inconsistency-adaptive logics in CLuN all connectives have the same truth-tables as in classical logic, just the table for negation is indeterministic: A A 0 1 1 [0/1] note that A, A CLuN B. The logic is paraconsistent. but also p, p q CLuN q 43/52

An Application: Inconsistency-adaptive logics in CLuN all connectives have the same truth-tables as in classical logic, just the table for negation is indeterministic: A A 0 1 1 [0/1] note that A, A CLuN B. The logic is paraconsistent. but also p, p q CLuN q abnormalities are contradictions: Π = {A A A L}. 43/52

Example Let Σ = { p q, p, r, r, r s}. We have the following model types: model p p q q r r s s M 1 Yes M 2 Yes Yes M 3 Yes Yes M 4 Yes Yes M 5 Yes Yes Yes M 6 Yes Yes Yes M 7 Yes Yes Yes M 8 Yes Yes Yes Yes Note that M 1 is minimally abnormal! (i.e., M 1 M i where 1 i 8). 44/52

Other examples Σ = {p q, p r, q, r, q r}. Does p follow? What are the minimally abnormal models? 45/52

Shoham / Kraus-Lehmann-Magidor General perspective: impose a partial order on a given class of all models S read: M M as M is more normal than M (S, ) is called a preferential structure (given smoothness) define Σ S A iff for all -minimal models M in S of Σ, M = A. 46/52

Shoham / Kraus-Lehmann-Magidor Each such S fulfills the following properties: Reflexivity: A A Cut: If A B C and A B then A C Cautious monotonicity: If A B and A C then A B C. Left Logical Equivalence: If A B then A C iff B C. Right Weakening: If A B and C A then C B. OR: If A C and B C then A B C. is called a preferential consequence relation if it fulfills these properties. 47/52

Shoham / Kraus-Lehmann-Magidor Theorem is a preferential consequence relation iff = S for some preferential structure S. 48/52

Cautious Monotonicity Blackboard. 49/52

No Rational Monotonicity Blackboard. 50/52

Other properties Where S, is a preferential structure. Do we get: Relevance: If bird S fly, penguin S fly then penguin lives-in-alaska S fly? 51/52

Other properties Where S, is a preferential structure. Do we get: Relevance: If bird S fly, penguin S fly then penguin lives-in-alaska S fly? Drowning: If bird S fly, bird S wings, penguin S fly then penguin S wings. 51/52

Thank you! 52/52