An argumentation-based approach to generate domain-specifjc - PowerPoint PPT Presentation

An argumentation-based approach to generate domain-specifjc explanations Nadin Kökciyan, Simon Parsons, Isabel Sassoon, Elizabeth Sklar, Sanjay Modgil School of Informatics, University of Edinburgh, UK nadin.kokciyan@ed.ac.uk

Introduction decisions taken. recommendations they make to assist humans in their decision-making. the conclusions. EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 2/ 20 ▶ Artifjcial intelligence (AI) has an increasing impact on ▶ Decision-support systems should provide justifjcations for the ▶ Computational argumentation is a technique for reasoning in which conclusions are drawn from evidence that supports

Argumentation Framework (background) nodes are the arguments, and the arrows are the attacks. computed. EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 3/ 20 ▶ An argumentation framework is a pair: Dung AF = � A ′ , R ′ � . ▶ AF can be represented with a directed graph where the ▶ According to the chosen semantics, the winning arguments are

Motivation arguments. common to carry out knowledge acquisition. elements. create explanations. EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 4/ 20 ▶ Existing argumentation-based approaches focus on acceptable ▶ We have little information about defeated arguments. ▶ Argument schemes (AS) and critical questions (CQs) are ▶ There is no consensus on a formal representation of these ▶ We need ‘explainability by design’. ▶ There is no clear method to use argumentation elements to

Baula: A recovering stroke patient Baula a 32-year-old person of African origin. Baula has started using a new medication c to control blood pressure as suggested by a GP. During a follow up visit, Baula’s BP is 130/90 (indicating the treatment is having the desired BP lowering efgect) but there is a side efgect (swollen ankles). What are the treatment options to consider and why ? EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 5/ 20

Baula: A recovering stroke patient Baula a 32-year-old person of African origin. Baula has started using a new medication c to control blood pressure as suggested by a GP. During a follow up visit, Baula’s BP is 130/90 (indicating the treatment is having the desired BP lowering efgect) but there is a side efgect (swollen ankles). What are the treatment options to consider and why ? EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 5/ 20

What do we need to help Baula? argumentation framework (e.g. EvalAF algorithm) together with justifjcations (e.g. ExpAF algorithm) EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 6/ 20 ▶ a logical language L to represent: ▶ Clinical guidelines (e.g. hypertension domain) ▶ Patient information (e.g. age, ethnic_origin etc.) ▶ a formal model based on L to represent: ▶ Argumentation framework components ▶ Domain-specifjc information (e.g. schemes) ▶ Domain-specifjc explanations ▶ a mechanism to translate a problem instance into an ▶ a decision-making mechanism to recommend decisions

Argument Schemes (ASs), Critical Questions (CQs), argument scheme. An argumentation-based approach to generate domain-specifjc explanations EUMAS-2020 function as described in Defjnition 2. of premises (e.g. facts), R is the set of rules and ASCQ is the Defjnition (Knowledge Base) set of argument schemes that represent the CQs of the original Knowledge Base (KB) Defjnition (ASCQ) variables used in the argument scheme. Defjnition (Argument Scheme) 7/ 20 AS = � P , c , V � denotes an argument scheme, where P is a set of premises, c is the conclusion, and P ∪ {c} ⊆ L . V is the set of ASCQ : AS → 2 AS , is a function mapping an argument scheme to a KB = � P , R , ASCQ � denotes a knowledge base; where P is the set

Example: ASPT Argument Scheme and CQs p 1 : Bringing about G is the goal. p 2 : Treatment TR promotes the goal G . p 3 : Treatment TR is indicated at step S . c: Treatment TR should be ofgered. ASPT.[SE] Has the patient experienced side efgects from this treatment in the past? efgective? EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 8/ 20 ASPT= �{ p 1 , p 2 , p 3 } , c , { G , TR , S }� ▶ SE.[SEF] Is the treatment

AS Instantiation, Argument Defjnition (Argument Scheme Instantiation) i th element in G . Defjnition (Argument) derived from the argument scheme instantiation AS i . EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 9/ 20 AS i = � AS , G , KB � denotes an instantiation of the AS with G ⊆ L in the knowledge base KB , AS { v i �→ g i } for all i = 1 ,.., k where k is the size of Var( AS ), v i is the i th element in Var( AS ) and g i is the [ AS ] arg i = � Prem ( AS i ) , Conc ( AS i ) � is an argument, which is

Attack scheme, Attack Defjnition (Attack Scheme) of type Y , P is a set of premises, c is the conclusion of the form ‘ p 1 Defjnition (Attack) derived from the attack scheme instantiation ATS i . EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 10/ 20 ATS = � { p 1 , p 2 } ∪ P , c, V � denotes an attack scheme with P ∪ {c} ⊆ L ; where p 1 is an argument of type X , p 2 is an argument attacks p 2 ’ and V = Var( X ) ∪ Var( Y ). X and Y can be same type. [ ATS ] att i = � Prem ( ATS i ) , Conc ( ATS i ) � is an attack, which is

Example: Attack schemes p 3 : A.TR is ofgered at step S . An argumentation-based approach to generate domain-specifjc explanations EUMAS-2020 Arguments Table: Attack between ASPT c: A attacks B . p 5 : A.TR is an alternative to B.TR . p 4 : B.TR is ofgered at step S . p 2 : B is an argument of ASPT p 1 : A is an argument of ASPT Table: An undercutting attack p 2 : An argument of type Y . p 1 : An argument of type X . 11/ 20 ALT= �{ p 1 - p 5 } , c , { A . TR , B . TR , S }� T cq = �{ p 1 , p 2 , p 3 } , c , Var( X ) ∪ Var( Y ) � p 3 : X challenges Y (i.e. X ∈ ASCQ ( Y ) ). c: p 1 attacks p 2 .

Dung AF and AF Defjnition (Dung Argumentation Framework, Dung AF) Defjnition (Argumentation Framework, AF) are, respectively, the set of arguments (Defjnition 5) and the set of EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 12/ 20 A Dung AF is a tuple � A ′ , R ′ � , where A ′ is the set of arguments and R ′ ⊆ A ′ × A ′ is a relation such that for arguments a and b , ( a , b ) ∈ R ′ ifg { a , b } ⊆ A ′ and a attacks b . An argumentation framework is a tuple � A , R � , where A and R attacks (Defjnition 7). The mapping to a Dung AF � A ′ , R ′ � is as follows: A ′ = A ; R ′ = { ( Prem ( r )[ 0 ] , Prem ( r )[ 1 ]) | r ∈ R } .

Acceptable attack, Acceptability Defjnition (Acceptable attack) Defjnition (Acceptability) acceptable arguments and attacks in the i th extension of AF . EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 13/ 20 An attack is acceptable, if ∀ r ∈ R , Prem ( r )[ 0 ] is an acceptable argument in Dung AF, R being the set of attacks. ACC = � AF , S � denotes the set of ( A arg , A att ) i where: S is the chosen semantics to evaluate AF , ( A arg , A att ) i is the pair of

Explanation template, Explanation Defjnition (Explanation template) argument scheme, and t is a text in natural language that can Defjnition (Explanation) EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 14/ 20 An explanation template is a tuple E = � AS , t � , where AS is an include variables V such that V ⊆ Var ( AS ) . An explanation is a tuple � E , [ AS ] arg i � , where E is an explanation template of the argument scheme AS , [ AS ] arg i is an acceptable argument (Defjnition 11); and for each variable v ∈ E . t , E . t { v �→ Gr( AS i )( v ) } .

Algorithms EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 15/ 20 ▶ EvalAF algorithm: ▶ uses the set of schemes of interest and a chosen semantics, ▶ instantiates arguments recursively, ▶ instantiates attacks, ▶ returns the sets of acceptable arguments and attacks. ▶ ExpAF algorithm: ▶ uses the sets of acceptable arguments and attacks, ▶ returns explanations for each, if any.

Baula: A recovering stroke patient Table: Argument schemes and arguments Table: Argumentation framework EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 16/ 20 att 1 att 3 att 4 [ASPT]arg 2 [ASPT]arg 1 [SE]arg 1.1 [SEF]arg 1.1.1 att 2

Baula: A recovering stroke patient Table: Argument schemes and arguments EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 16/ 20 att 1 att 3 att 4 [ASPT]arg 2 [ASPT]arg 1 [SE]arg 1.1 [SEF]arg 1.1.1 att 2 Table: Preferred Extension 1 ({ arg 1 , arg 1 . 1 . 1 }, { att 2 , att 4 })

Baula: A recovering stroke patient Table: Argument schemes and arguments EUMAS-2020 An argumentation-based approach to generate domain-specifjc explanations 16/ 20 att 1 att 3 att 4 [ASPT]arg 2 [ASPT]arg 1 [SE]arg 1.1 [SEF]arg 1.1.1 att 2 Table: Preferred Extension 2 ({ arg 2 , arg 1 . 1 . 1 }, { att 1 , att 4 })