Towards a Generic Framework for a Health Behaviour Change

Support Agent

Fawad Taj

, Michel CA Klein

and Aart Van Halteren

Social AI Group, Vrije Universiteit, Amsterdam, The Netherlands

Keywords: BDI-based Agent, Health Behavior Change Agent, Behavior Change Technique.

Abstract: Agent-oriented solutions form a useful paradigm to design intelligent systems. For health-related behaviour

change, this is also a promising approach. Designing an agent for lifestyle change interventions is a difficult

task because socio-ecological models are involved that represent many conflicting desires and goals. Different

types of cognitive architectures are available to design this type of health behavior agents but they are rarely

used. In this paper, we used the BDI model to design a health behavior agent that will execute behavior change

intervention for a better healthy lifestyle. We explain the working of the architecture by the example of an

agent which uses adaptive goals-setting and a percentile scheduling technique for increasing physical activity.

1 INTRODUCTION

One approach to build real-world complex systems is

using the agent-oriented paradigm. In this paradigm,

software components are tightly connected with one

another and they all function autonomously. Artificial

intelligence provides a major contribution to the agent

development paradigm due to the required properties

of autonomy, cognitive thinking, sociability and

learning (Girardi, 2001). Cognitive agents in artificial

intelligence are among the most developed and

studied topics, which explicitly maintains the model

of the environment perceive the external

environment, do rational thinking and make a plan to

act on the environment to fulfil one or more of its goal

(Wooldridge, 1995). Agent-based modelling is

common and brings significant advantages to systems

when the environment is complex, the interaction

between agents is nonlinear, discontinuous or the

population is heterogeneous (Bonabeau, 2002).

Health-related systems are complex, due to some

hard topics like patient life, data privacy, legal and

technical issues. For example delay or

misinterpretation between different entities/agents

could cost someone life (Datta et al., 2010)(Iqbal et

al., 2016). The use of agent systems in healthcare

setups has increased in the last decade and the usage

https://orcid.org/0000-0001-9049-1736

https://orcid.org/0000-0003-4119-1846

https://orcid.org/0000-0002-9631-0657

ranges from patient-centred applications to the

organizations-centred, multi-agent system (Isern &

Moreno, 2016). Drawing on (Datta et al., 2010)(Iqbal

et al., 2016), recent reviews about agents applied in

health-care, categorized the agents both on the basis

of intended users and functionality. The applications

are mainly patient-centered, staff-centered or

healthcare organization-centred and with respect to

functionality basis, they can be designed for planning

and resource management, decision support system,

data management, self-care systems and can be

multifunction systems that can integrate some of the

earlier describe systems to make a complete

healthcare system. Another subset of health-care

systems is behavior change support systems, which

could benefit from agent-based intelligent models to

facilitate rational and on-time decisions in a

heterogeneous environment.

An approximate 60% of the risks associated with

chronic diseases such as diabetes and cardiovascular

disease are associated with health habits and these

conditions account for 1.5%-3% of direct costs to the

UK National Health Service (NHS) (GC et al., 2016).

It is becoming critically important to question the

creation and implementation of effective methods to

improve healthy behavior. With a change in lifestyle

and prevention techniques, we can significantly

Taj, F., Klein, M. and Van Halteren, A.

Towards a Generic Framework for a Health Behaviour Change Support Agent.

DOI: 10.5220/0009173503110318

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 1, pages 311-318

ISBN: 978-989-758-395-7; ISSN: 2184-433X

311

decrease the impact of non-communicable diseases,

which are some of the greatest challenges facing

modern society. Some of the key unhealth behaviors,

such as physical inactivity, unhealthy eating,

smoking, obesity, sexual behavior, and alcohol

misuse are among the most common causes of disease

and premature deaths in both developed countries

(Ding, Lawson, & Lancet, 2016).

While this prior research on improvements in

health behavior is critical in defining pragmatic

approaches that could lead to changes in health

behaviour, the theories developed in it are insufficient

to support the development of quantitative delivery

methods. Furthermore, the proposed theory-based

models consider health behavior as a function of

constructs such as motivation, attitude as opposed to

a product of a dynamic cognitive system that is

influenced by physiological, affective,

environmental, social, and experiential states (Riley

et al., 2011). Agent-based modelling, in contrast,

provides the opportunity to define simple reflexive

agent up to more complex cognitive learning agents.

When designing and developing agent-based

systems, an important question is to choose or

develop the decision-making process of agents. There

are around 15 famous decision-making models in the

literature, each based on different literature studies

(Balke & Gilbert, 2014). The main inspiration for

these decision-making systems are the human

psychological and neurological systems. A widely

used way to formalize the internal architecture of

such complex agents is the BDI (Belief-Desire-

Intention) paradigm. This paradigm allows to design

expressive and realistic agents, yet, it is rarely used as

an intelligent health behavior change agent. We argue

that health-behavior coaching – helping people to

develop helpful health-related behaviors and to

curtail harmful ones – is a challenging as well as a

fruitful domain to conduct human-aware AI research.

The domain requires that a health coach understand

the cognitive, emotional, physical, situational, and

other aspects of a coaches’ health behaviors. The

possible interventions vary from providing

informational support, encouraging the practice of

helpful behaviors in different contexts, helping to

remember behaviors when the right context arises,

etc. To be impactful, these agents need to make a

more personalized decision and gradually adapted for

their specific circumstances. This paper presents a

generic framework for a health behavior support

agent, inspired by the BDI paradigm.

This article is structured as follows: section 2 will

discuss the concept of health behavior change and the

guidelines to define healthy behavior intervention.

Section 3 discusses agent architecture based on BDI.

Section 4 will define the components based on the

discussion in section 2. This section will discuss the

working of the model with the help of physical

activity scenario and discusses agent-based

programming algorithm. Finally, section 5 will

discuss future work and draw a conclusion on earlier

sections and will discuss the future directions.

2 BACKGROUND

In this section, we describe the theoretical

components of a health behaviour support agent.

Before defining the goals and plans for the agent, we

should determine the desired results of the agent in

detail. First, we have to decide the goal and plan

occurring in a different context. It is recognised that

any behavior that needs to be changed occurs in

several different contexts (e.g. at home, at work) and

have many different influences (e.g. personal,

interpersonal and environmental). Therefore,

different intervention results for each context and

level of influence were therefore defined. For this

purpose different taxonomies and planning guides

can be consulted (Kok, 2014). Using the steps defined

in these planning guides we can identify the context,

the performance outcome, and select the right

behavior change strategies.

Certain behaviors can be targeted with different

behavior change techniques (BCTs), which acts as an

active ingredient in any behavior change intervention.

Each BCT use a different mechanism of action to

target certain behavior (Michie & Johnston, 2012).

The interventions are usually delivered by expert

humans through a prolonged interaction with the

people they coach. According to (Taj, Klein, & van

Halteren, 2019), these BCTs are poorly reported and

the most used technique is the goal-setting

irrespective of target health domain. Each of these

behavior change techniques is differently modelled

and mathematically represented. For example, the

goal-setting technique is represented as the staircase

model to set an adaptive goal for coachee (Mohan,

Venkatakrishnan, 2017), whereas in another example

the adaptive goals are calculated with percentile

schedule method (Adams, 2009). Based on this

background knowledge we defined different

parameters for our physical activity agent in section

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

312

3 BDI ARCHITECTURE

3.1 Overview

The BDI approach in artificial intelligence represents

the way agents can do complex reasoning based on

folk psychology (Bratman, 1987). The three main

mental states around which the BDI model is centred

are belief, desire, and intentions. A typical BDI agent

represents all the information that it has about the

environment in the form of beliefs and these beliefs

can be represented by modal logic language. These

beliefs can be either true, false or outdated. The agent

has some desires that it wishes to accomplish. Not all

but for some desires that the agent actively wants to

achieve turns to become intentions and the agent is

equipped with a pre-defined set of plans which are

recipes for achieving its intentions (Visser et al.,

2016).

An agent architecture shown in figure 1 is a

software computational solution to a problem

showing how the component parts of a system

interact, thus providing an overview of the system

structure. It encodes its sensory perceptions into a

state representation of its environment. It also

represents the plans it can execute to manipulate its

sensors, effectors, and the environment in pursuit of

goals.

The basic logic components of a BDI agent are

belief, desire and intention. In our model we follow

the conventions adopted in the GAMA-platform- a

control architecture, which in turn are based on PRS

(Procedural Reasoning System). The vocabulary of

the key terms of the architecture can be summarized

as follows.

3.2 Vocabulary

3.2.1 Knowledge States

The agent must represent the environment by

capturing information that is necessary to not only

formulate a beneficial goal state but also to decide

over its action space. All the information about the

environment is represented in the different

representation states. For example, the affective states

can be represented as a different scale, whereas the

preference value can be represented as logical

predicates. These states will not only be used to

update the belief of the agent but also will help in

defining the algorithm of behavior change techniques.

In figure 1 a few of possible knowledge states are

mentioned; however, any number and types of states

can be considered depending on the behavior the

agent is targeting.

Figure 1: Conceptual diagram of BDI behavior change

agent.

3.2.2 Beliefs

Belief is the agent knowledge about the world. The

belief base always gets updated with the new

information in representation states. The beliefs of the

BDI is usually represented as predicates for example,

Step_count (Monday, 3000)- a person steps count for

Monday is 3000 steps. Belief can either be true or

false.

3.2.3 Desire

Desires are all the objectives that the agent wants to

achieve and often called the goal of the agent. It can

have hierarchical links (sub/super desires) or each

desire can be defined with a dynamic priority value.

For example, the agent can have different conflicting

desires which can be ordered according to some

priority values according to the intervention. For

example, for “set goal” desire will be having higher

priority than giving a reward.

3.2.4 Intentions

Among the desires that an agent wants to achieve it

select one having high priority. The intention will

determine the selected plan. That is the reason that

BDI based agents are usually called intention systems

(Balke & Gilbert, 2014).

3.2.5 Plan

The agent plan base consists of actions that the agent

would carry to fulfil its selected intention. In our

architecture shown in Figure 1, most of the plans are

the delivery or implementation of behavior change

techniques. Now to define a plan for certain desire

regarding behavior change the plan needs to have

some pre and post conditions and the body which in

our case can raise to the question that which behavior

Towards a Generic Framework for a Health Behaviour Change Support Agent

313

change techniques are best for which kind of health-

related problem. There are a lot of randomized control

trials available that can help us define our plan. For

example, to implement goal-setting intention, the most

used behavior change technique is the goal setting.

Figure 2: The flow chart of behavior change agent.

3.2.6 Behavior Change Techniques

Algorithms

The plan base in architecture shown in Figure 1

contains behavior change techniques algorithms where

all the relevant BCTs can be placed so that the agent

can reason about it and select one that is relevant and

feasible to the scenario. In health behavior change

literature several constructs such as goal-setting, self-

efficacy, reward shaping and incentives are defined

and extensively studied to positively influence health

behavior. The well-known hierarchy taxonomy by

Susan can be considered to properly report and define

the intended construct (Michie et al., 2013).

The algorithm for each of the BCT will explain post

and pre-conditions, and intensity of the BCT. Currently

the efficacy of different kind of BCTs are not

established with regard to different behavior. Different

people define each type of BCT with their unique

algorithm. This is why the algorithm for BCT is shown

separate than simple plans.

3.3 Workflow

In the BDI practical reasoning, the agent is equipped

with a library of pre-compiled plans. These plans are

manually constructed, in advance, by the agent

programmer. Hence in Table 1 we define performance

outcomes as the goal of the agent and defined the plan

as the execution of relevant BCT.

The flow of the process depicted in Figure 2 is as

follow:

1. Set initial goal: The initial goal will be the

behavioral outcomes that we want to achieve.

2. Plan execution: execute the relevant plan which

can be either to select the sub-goal or update the

belief base.

3. Perceptions are updated: For each behavioral

goal, the second step is to perceive the relevant

information from the environment and update

agent belief base.

4. If the current goal contains a sub-goal it would

hold the current goal on hold and will select the

sub-goal and will select the relevant plan for it.

5. If the current goal doesn’t have any sub goals and

don’t need to update the belief base the current

plan would be executed.

6. After successful completion of the plan, the new

desire with the highest priority would be selected.

7. Until the successful completion of the current

plan, the current intention would still in hold and

will execute until get finished.

8. The reasoning end if there is no plan and desire

available for execution.

4 SCENARIO

Using the planning guidelines discussed in section 2,

Table 1: The behavior change performance objectives and

selected methods.

Target

population:

individual

Target

behavior:

Physical

activity

Determinant:1

Intention

Determinant:2

Motivation

Selected

BCT

Performance

outcome:

increase

number of

daily steps

count

A resolve to act

on certain way

Goal

Setting

(behavior)

Performance

outcome:

Keep

motivation

for behavior

outcome

Arrange reward

if and only if

there is an

effort to

achieve the

targeted

behaviour

Rewards

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

314

we will choose some specific components for our

example. Table 1 shows the performance outcomes

and the mechanism of action (determinants)-through

which we will achieve our target in our example

scenario that is presented below. The last column

shows the selected BCTs that are considered best in

literature for these kinds of targets. We considered

adaptive goal-setting techniques for the daily

recommended steps. The mathematical formulation

and algorithm are defined in section 4.1.

4.1 Adaptive Goals and a Percentile

Schedule of Reinforcement

Adaptive goals that often and uniquely adjust to the

recent performance of an individual may be a more

realistic approach to developing flexible yet

challenging and achievable goals. The goal-setting

and feedback algorithm was based on a rank-order

percentile algorithm derived from recent

developments in basic science around schedules of

reinforcement (Adams, 2009). The percentile

algorithm requires continuous and repeated

measurements of daily steps count and then the

algorithm work as follow:

1. The ranking of a sample of behavior (steps/day)

from lowest to highest and calculation of a new goal

based on a pth percentile criterion. For example, for

one participant, the steps count each day for their last

9 days (ranked from lowest to highest) was 1000,

1500, 2600, 4500, 5000, 5700, 6300, 8000, 11,000.

2. The 60th percentile represents a goal of 5700

steps, which becomes the 10th day's goal. Based on

(Adams, 2009), the best window to consider is of 9

window and the pth percentile 60% of the last 9

reading which is calculated with

=((p/100) * no_day). (1)

To achieve customized targets, percentile shaping

capitalizes on the normal behavioral variability.

Percentile shaping also generates specific,

measurable goals inherently that can be explicitly

rewarded. Only a handful of studies have evaluated

the use of a percentile shaping strategy by changing

goals to increase physical activity, and none have

compared percentile shaping goals orthogonally

(Adams, 2009).

In our example, the aim is to develop an intelligent

agent that can provide counselling in a manner similar

to a human coach. There is a need for computational

methods that can not only model and predict the

changes in the human physiological and cognitive

system, but also for methods that can coach this

human system toward a beneficial goal (Shiwali

Mohan & Venkatakrishnan, 2017). The working of

the agent for the given scenario below is depicted in

Figure 3.

The BDI health behavior agent model function as

follows: our agent aims to assist in delivering the best

available BCT for increasing physical activity and

keep the motivation high to maintain the behavior. To

make it simple we considered simple physical activity

guidelines of 10000 steps per day by the National

Heart Association of Australia (Tudor-Locke et al.,

2011). To achieve this goal, a number of behavior

change techniques can be applied but according to the

literature, the mostly applied BCT for physical

activity behavior is goal-setting. The goal-setting

theory by Locke and Latham (Locke & Latham,

2012) provides evidence that to be maximally

effective, the goals should be difficult yet attainable.

Therefore, for adherence purpose, the agent will set a

new adaptive goal for the coachee each day if the

coachee didn’t meet the standard guidelines.

Moreover, for motivation and reinforcement purpose,

if the coachee meeting the guidelines the coachee will

be awarded rewards. To make it short, the main goal

is to assist user to maintain 10000 steps count daily

and sub goals to achieve this main goal is goal setting

and reward.

According to the flow diagram discussed in Figure

2 the flow of the process of our scenario is as follows:

1. Initial goal of increasing physical activity is

depicted as initial desire: Keep_Fit

2. To fulfil this desire the plan is to start monitoring

daily steps count and update the belief base

accordingly. For example, step_count(Monday,

3000)- a person steps count for Monday is 3000

steps.

3. while executing the monitoring plan and adding

new belief about daily step count. A rule is

introduced which add new sub-goal of goal setting

by applying the following rule.

Rule: with each belief: Steps_count add

New_desire: goal_setting

4. For this new desire of goal setting the plan is to

call the adapting goal setting algorithm. Which is

explained in section 4.1.

5. The execution of the plan will also update the

belief base of the agent.

6. Whenever the step counts would be more than

1000 per day the goal setting plan would generate

a new sub-goal of reward.

add_sub_intention(reward)

Towards a Generic Framework for a Health Behaviour Change Support Agent

315

Figure 3: Flow chart of the scenario.

4.2 ABMS Platforms and Languages

for Simulating BDI Agents

The design of its internal architecture is an important

decision when developing a software agent. Several

models of deliberative, reactive, and hybrid

architectures have already been proposed. BDI

architecture is one of the most popular agent decision-

making models in the community of agents. BDI

architectures have been introduced in several agent-

based modelling and simulation (ABMS) platforms.

For example, the BDI paradigm integrated into the

GAMA modelling platform and its GAML modelling

language to manipulate BDI concepts in a simple

language (Taillandier, Bourgais, Caillou, Adam, &

Gaudou, 2017). There also exists some middleware to

connect the famous ABMS platform to BDI

frameworks e.g. JACK (Busetta, Rönnquist,

Hodgson, & Lucas, 1999) and Jadex (Pokahr,

Braubach, & Lamersdorf, 2005).

A programming language is an essential

component of agent-based technology and agent-

based systems implementation. Such a language,

called an agent-oriented programming language,

should provide high-level abstractions and constructs

for developers to implement and use agent-related

concepts directly. Some of the famous languages

Agent-oriented languages that support BDI

architecture are AgentSpeak(L), Jason, Af-APL,

2APL, JACK(L), JADEX, GOAL etc.

We will illustrate our scenario using a

programming language. Algorithm 1 is developed

with the close syntax to GAML modelling language

but it can be modelled in any agent-oriented language.

Algorithm 2 shows the goal-setting algorithm and

algorithm 3 keep track of the reward for achieving the

goal. The algorithm uses an existing constraint solver

and does not need to modify or enforce the

vocabulary of the BDI.

Algorithm 1: BDI based goal-setting agent.

Algorithm 2: Adaptive goal based on Percentile schedule

algorithm.

Algorithm 3: Reward algorithm.

Procedure Main()

Create agent



goal_setting_agent



name of

the agent

Target_steps:1000#day

Reward:0 Percentile:

Agent goal_agent Control::BDI

Procedure init() {

add Desire



keep_fit }

Perceive target:no_of setps var:day

Add belief: Steps_count var: day

Do remove_intention(keep_fit, false)

Rules belief: Steps_count New_desire: goal_setting

Plan record_steps_count where intention: keep_fit

Do read_daily_steps_count

Plan set_goal where intention: goal_setting

If current_step_count < Target_steps

Do add_sub_intention(get_current_intention(),

find_adaptive_goal, true )

Do current_intention_on_hold();

Else

Do add_sub_intention(reward)

Plan adaptive_goal where intention:

find_adaptive_goal_perc

Call Adaptive_percentile_goal (days

,p)



Goal

setting algorithm

do remove_intention(find_adaptive_goal_perc,

true)

Plan calculate_reward where intention: reward

Call reward()



reward algorithm

do remove_intention(reward, true)

Procedure Adaptive_percentile_goal (no_days, p

)

Do arrange daily_steps_count in ascending order

Compute the position of p

percentile /60

Return ((p/100)*no_day)

Procedure reward ()

Do add reward=reward+1



make sure

reinforcement

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

316

5 FUTURE WORK

In this article, we explore ways that an agent system

can specify the goal for the coachee according to his

previous performance which is incorporated into the

BDI execution process and used to guide the choices

made.

The future direction would be to implement this

algorithm with any agent base modelling

environment and will simulate it. The agent

technology is rarely adopted in health behavior

domain so there is so much opportunity to include

knowledge from behavior sciences. For example,

adding more personalization aspect to agent e.g. a

value-based planning approach which takes into

account social and ethical values that affect decision-

making (Cranefield, Winikoff, Dignum Delft

MVDignum, & Frank Dignum, 2017).

The health behavior agent needs to consider the

causal model which can assess the failure or success

of the intervention, this can be achieved by

considering a causal model within the BDI

architecture. The coachee may not have enough

expertise or resources to conduct the behavior, may

not believe they can execute the behavior effectively

(low self-efficacy), may not have the right emotional

state or having some social norms etc. (Shiwali

Mohan & Venkatakrishnan, 2017). This kind of

model is already available which can initially do

reasoning about unwanted behavior (Klein, Mogles,

& Van Wissen, 2011), which can likely be modelled

according to BDI architecture.

Furthermore, a promising direction to equip the

health change agent with a functionality that allow it

to reason about the reasoning of the coachee. This

topic has received significant research attention and

can be explored with the help of implementing

Theory of Mind (ToM). Theory of mind provides an

important understanding of how human reason about

other mental states (Baron-Cohen, Leslie, & Frith,

1985). There is some research which introduces a

formal BDI-based agent model for Theory of Mind,

which can be used or modified to reason about the

coachee health-related constructs (Bosse, Memon, &

Treur, 2007).

6 CONCLUSION

In this paper, we proposed a design of a BDI based

health behavior agent model that can monitor and

reason about the different psychological and

physiology constructs of its user. The knowledge

about the environment is represented in the form of beliefs

and the intentions are fulfilled in the form of delivering the

right kind of behavior change technique. The model is

illustrated with the help of an example of physical activity

coach which records the daily steps count of the coachee

and according to the adopted goal-setting technique, the

agent selects goals that are appropriate for a coachee given

the past history of performance. The agent’s other goal is to

keep the motivation high for which the agent uses the

reward-based behavior change technique.

REFERENCES

Adams, M. A. (2009). A pedometer-based intervention to

increase physical activity: Applying frequent, adaptive

goals and a percentile schedule of reinforcement. UC

San Diego.

Balke, T., & Gilbert, N. (2014). How do agents make

decisions? A survey. Journal of Artificial Societies and

Social Simulation, 17(4), 13.

Baron-Cohen, S., Leslie, A. M., & Frith, U. (1985). Does

the autistic child have a “theory of mind” ? Cognition,

21(1), 37–46. https://doi.org/10.1016/0010-

0277(85)90022-8

Bonabeau, E. (2002). Agent-based modeling: Methods and

techniques for simulating human systems. Proceedings

of the National Academy of Sciences, 99(suppl 3),

7280–7287.

Bosse, T., Memon, Z. A., & Treur, J. (2007). A two-level

BDI-agent model for theory of mind and its use in social

manipulation. In Proceedings of the AISB 2007

Workshop on Mindful Environments (Vol. 4, pp. 335–

342).

Bratman, M. (1987). Intention, plans, and practical reason.

Busetta, P., Rönnquist, R., Hodgson, A., & Lucas, A.

(1999). Jack intelligent agents-components for

intelligent agents in java. AgentLink News Letter, 2(1),

2–5.

Cranefield, S., Winikoff, M., Dignum Delft MVDignum, V.

T., & Frank Dignum, tudelftnl. (2017). No Pizza for

You: Value-based Plan Selection in BDI Agents.

Datta, A., Dave, N., Mitchell, J. C., Nissenbaum, H.,

Sharma, D., & others. (2010). Privacy Challenges in

Patient-centric Health Information Systems. In

HealthSec.

Ding, D., Lawson, K., Lancet, T. K.-A.-T., & 2016,

undefined. (n.d.). The economic burden of physical

inactivity: a global analysis of major non-

communicable diseases. Elsevier. Retrieved from

https://www.sciencedirect.com/science/article/pii/S014

067361630383X

GC, V., Wilson, E. C. F., Suhrcke, M., Hardeman, W., &

Sutton, S. (2016, April 1). Are brief interventions to

increase physical activity cost-effective? A systematic

review. British Journal of Sports Medicine. BMJ

Publishing Group. https://doi.org/10.1136/bjsports-

2015-094655

Towards a Generic Framework for a Health Behaviour Change Support Agent

317

Girardi, R. (2001). An analysis of the contributions of the

agent paradigm for the development of complex systems

Iqbal, S., Altaf, W., Aslam, M., Mahmood, W., & Khan, M.

U. G. (2016). Application of intelligent agents in

health-care: review. Artificial Intelligence Review,

46(1), 83–112. https://doi.org/10.1007/s10462-016-

9457-y

Isern, D., & Moreno, A. (2016). A Systematic Literature

Review of Agents Applied in Healthcare. Journal of

Medical Systems, 40(2), 1–14. https://doi.org/10.1007/

s10916-015-0376-2

Klein, M., Mogles, N., & Van Wissen, A. (2011). Why

won’t you do what’s good for you? Using intelligent

support for behavior change. In International Workshop

on Human Behavior Understanding (pp. 104–115).

Kok, G. (2014). A practical guide to effective behavior

change How to apply theory- and evidence-based

behavior change methods in an intervention. The

European Health Psychologist, 16(5), 156–170.

Retrieved from https://www.ehps.net/ehp/index.php/

contents/article/viewFile/ehp.v16.i5.p156/8

Locke, E. A., & Latham, G. P. (2012). Goal setting theory.

In Motivation: Theory and research (pp. 23–40).

Routledge.

Michie, S., & Johnston, M. (2012). Theories and techniques

of behaviour change: Developing a cumulative science

of behaviour change. Health Psychology Review, 6(1),

1–6. https://doi.org/10.1080/17437199.2012.654964

Michie, S., Richardson, M., Johnston, M., Abraham, C.,

Francis, J., Hardeman, W., … Wood, C. E. (2013). The

behavior change technique taxonomy (v1) of 93

hierarchically clustered techniques: building an

international consensus for the reporting of behavior

change interventions. Annals of Behavioral Medicine,

46(1), 81–95.

Mohan, S, Venkatakrishnan, A., … M. S.-T.-N. I., & 2017,

defined. (n.d.). On designing a social coach to promote

regular aerobic exercise. Aaai.Org. Retrieved from

https://www.aaai.org/ocs/index.php/IAAI/IAAI17/pap

er/viewPaper/14963

Mohan, Shiwali, & Venkatakrishnan, A. (2017). Health

Behavior Coaching: A Motivating Domain for Human-

Aware Artificial Intelligence Research Advanced

Cognitive Learning for Embodied Language (AILEEN)

View project Interactive Task Learning View project.

Retrieved from www.aaai.org

Pokahr, A., Braubach, L., & Lamersdorf, W. (2005). Jadex:

A BDI Reasoning Engine (pp. 149–174).

https://doi.org/10.1007/0-387-26350-0_6

Riley, W. T., Rivera, D. E., Atienza, A. A., Nilsen, W.,

Allison, S. M., & Mermelstein, R. (2011). Health

behavior models in the age of mobile interventions: are

our theories up to the task? Translational Behavioral

Medicine, 1(1), 53–71.

Taillandier, P., Bourgais, M., Caillou, P., Adam, C., &

Gaudou, B. (2017). A BDI agent architecture for the

GAMA modeling and simulation platform. In Lecture

Notes in Computer Science (including subseries

Lecture Notes in Artificial Intelligence and Lecture

Notes in Bioinformatics) (Vol. 10399 LNAI, pp. 3–23).

Springer Verlag. https://doi.org/10.1007/978-3-319-

67477-3_1

Taj, F., Klein, M. C. A., & van Halteren, A. (2019). Digital

Health Behavior Change Technology: Bibliometric and

Scoping Review of Two Decades of Research. JMIR

MHealth and UHealth, 7(12), e13311.

Tudor-Locke, C., Craig, C. L., Brown, W. J., Clemes, S. A.,

De Cocker, K., Giles-Corti, B., … others. (2011). How

many steps/day are enough? For adults. International

Journal of Behavioral Nutrition and Physical Activity,

8(1), 79.

Visser, S., Thangarajah, J., Harland, J., & Dignum, F.

(2016). Preference-based reasoning in BDI agent

systems. Autonomous Agents and Multi-Agent Systems,

30(2), 291–330. https://doi.org/10.1007/s10458-015-

9288-2

Wooldridge, M., review, N. J.-T. knowledge engineering,

& 1995, undefined. (n.d.). Intelligent agents: Theory

and practice. Cambridge.Org. Retrieved from

https://www.cambridge.org/core/journals/knowledge-

engineering-review/article/intelligent-agents-theory-

and-practice/CF2A6AAEEA1DBD486EF019F6217F1

597

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

318