A model for a motivational system
grounded on value based abstract
argumentation frameworks
Eugenio Di Tullio
University of Bari
Floriana Grasso
University of Liverpool
The problem
• Digital interventions to promote healthy lifestyles
• More successful if grounded in robust behavioural
– Variety of techniques to impact different stages of the
planning process
– Use of different styles of interaction
• New insights from recent research areas in AI:
– Persuasion technology
– Argumentation Theory
Argument & Computation
• Over the last decade a core and autonomous
discipline within AI, as opposed to permeating
reasoning research
– Emphasis on what persuades, not proofs
– Hypothesis that the extent of acceptance is subjective,
depending on the view of the audience
• Crucial aspect within practical reasoning, as
opposed to theoretical reasoning
– Presumptive schemata
– Embraced by recent research in MultiAgent Systems
Application areas
• Multi-agent dialogue interactions
– E.g. negotiation scenarios
• Legal reasoning
– E.g. tools to model and support case-based reasoning
• Healthcare advice
– E.g. changing attitudes to healthy eating
• Medical treatment
– E.g. automated argument over organ transplants
• Finance
– E.g. automated stock market exchanges/contract
Directions of research
• Abstract argument systems: formal exploitation
of simple notions of “attack” and “defense”.
• Logics for argumentation: formal models for
automated reasoning.
• Dialogue protocols: emphasis on the rules that
should govern the argumentation process and the
participants to it.
• Argument schemes: stereotypical patterns for
presumptive reasoning, fallacies.
Abstract Argumentation
An argumentation
framework (AF) is a pair
F = (A, R) where
A is a set of arguments
R ⊆ A × A is a relation
representing “attacks” or
Different semantics to
establish which arguments
are “in” and which are
Transtheoretical model of
Motivational dialogues
• A subclass of argumentative dialogues
• Main feature: a discussion around a behaviour,
and the considerations of pros and cons of
such behaviour
– Can’t be based on “facts” only
– Highly entrenched in the “value system” of the
parties engaging in the discussion
– There is no right or wrong answer – things change
when perspectives change
A1:You should exercise twice a week because it improves
your health.
B2: Why is it good for my health?
A3: Because exercise improves your stamina.
B4: But then I might as well go to work by bike.
A5: No, exercising is better for your health.
B6: But exercise is boring.
A7: What is more important: your health or having fun?
B8: I find my health is more important. I guess I should
(from van der Weide et al 2010 “Practical Reasoning Using Values”)
Practical Reasoning on
Values and Perspectives
• Values: “desirable trans-situational goals, varying in
importance, that serve as guiding principles in the life
of a person or other social entity”
Schwartz,S, Advances in experimental social psychology 25 (1992)
• Value Based Argumentation Framework (VAF)
(Bench-Capon et al, Argumentation Research Group, University of Liverpool)
– Tuple <AR, attack,V, val, P>
– AR set of arguments; attacks relation on AR,V set of
values, val function mapping from AR to V P set of
possible audiences.
– An argument relates to value v if accepting it
promotes or defends v.
Practical Reasoning on
Values and Perspectives -ctd
• Values are discussed in terms of what condition
promotes or demotes them
– E.g. treating people the same promotes the value of
equality, exercising promotes the value of being
• Conditions and values can be promoted in
– Not exercising is not healthy, exercising once a month
is healthier, once a week is healthier still
• Perspective: preorder on states representing a
“topic” for discussion (e.g. health, fun)
Practical Reasoning on
Values and Perspectives -ctd
• Perspectives can positively or negatively
influence other perspectives forming chains
– (E.g. health positively influence wellbeing)
• When an agent has a “preference” over a
perspective, this perspective becomes a “value”
– used as a “guiding principle” by the agent
– agent will try to reach as state maximally preferred
from that perspective
Health, Fun and Conformity
perspectives are also values
Exercise perspective
is not a value, but
influences values
(from van der Weide et al 2010 “Practical Reasoning Using Values”)
Motivational System
• Ultimate aim:
– Build an environment for digital interventions,
based on motivational dialogues
– Authoring tools for creating user profiling,
communication plans/strategies, styles of
– Mobile technology to acquire user’s preferences
and lifestyles
– Centred around the notion of “Value System”
Prototype implementation
• A prototype built on top of Aspic
(Argumentation Service Platform with
Integrated Components - 6FP - Opensource)
– A platform to manage argumentation dialogues, which also
provides services like reasoning, decision-making, learning.
• Concentrates on the managing of the value
system and the practical reasoning on
– Left aside: user’s profiling and behavioural strategy
System architecture
Value System Ontology
Managing values
• Two value systems are maintained
– For the user and for “the system”
– Typical for dialogue systems representing “mutual”
• Interaction driven by a plan or “strategy” based
on the user’s behavioural profiling (e.g.
Transtheoretical model)
– E.g. topic of discussion based on the stage of change
• System attempts to “utilise” user’s own beliefs
and value system to maximise motivation impact
• The system contains the states:
1. Eating junk food less than 4 times a month;
2. Eating junk food between 4 and 8 times a month;
3. Eating junk food more than 8 times a month
• And the perspectives to evaluate the states
are Healthy eating; Fitness; Health; Social life
• Initial value systems are:
Evaluation of Transition A
• System attempts to make user “aware” of
influences which could impact the evaluation of
transition A
New evaluation of A
New connections
• Preliminary work towards an argumentation
based motivational system
• Attempt to combine insights from argumentation
theory and behavioural theories
• Still prototypical: need to include more reliable
user model and full blown communication
• Adherence to standards (e.g. Aspic) looks
beneficial in terms of cross-discipline evaluation

similar documents