1. Introduction
Most continents, in particular those where the authors live, are exposed to the tangible efects
of the ageing of the population, with all the challenges for the society and for the individuals
(the ageing people, but also their relatives, caregivers, doctors) that this demographic change
raises.
In this paper, we present our vision on how to address some of these challenges; in particular,
we investigate how intelligent software agents could mitigate the risk of progressive loss of
physical, intellectual, and emotional activity of ageing people, and of their social exclusion,
in particular when a transition from diferent situations (moving from home to a protected
structure, or from a protected structure to an hospital) takes place.
We believe that one way to cope with the needs of old people, especially during these delicate
transitions where they are much more fragile than in stable situations, is to provide them with
highly-interactive culturally adaptive “Guides” that not only care for their health but also serve
as entertaining companions that interact through spoken dialogues and that stimulate their
cognitive abilities. Such Guides, designed and implemented as software agents compliant with
the strong notion of agency [
1
] and also characterised by emotional features à la Bates [
2
],
besides mentalistic ones à la Shoham [
3
], are meant to become a familiar and trustable point
of reference for their users. The Beliefs-Desires-Intentions (BDI) agent model, extended with
emotional features as suggested for example by Pereira et al. [
4
], Alfonso et al. [
5
], Su et al. [
6
],
would hence represent a suitable architecture for our Guides, and has been recently adopted
in healthcare applications [
7
]. Given the BDI-oriented nature of the Guides1, we will exploit
languages like DALI [
8
], SR-APL [
9
], IndiGolog [
10
], or Jason [
11
], meant as a standalone
infrastructure or, better, integrated in the environment- and organization-oriented JaCaMo
framework [
12
], for their implementation.
We use the word “Guide” to emphasise that such interactive assistants are always available
to their users when they need advice or company, wherever they are, be it the comfort of their
home, outdoors, but also at dificult times for example when moving to protected structures
or hospital is needed. The Guides may provide guidance — following a protocol pre-agreed
with doctors, psychologists and caregivers — to the elderly on all those situations which do not
require a doctor evaluation and assessment (entertainment, social activity, physical activity, diet,
medication to take according to the agreed protocol). By watching over the elderly, Guides can
collect precious information on their behaviour and can serve as a source of reliable information
for the doctors who care for the elderly. Hence, guidance is bidirectional: caregivers choices
and focus of attention can indeed be driven by the Guide, based on what it senses, observes,
hears, and deduces.
We expect that Guides should quickly learn the cultural profile of their users and flexibly adapt
to cross-cultural interaction. This means adapting the style of conversation to the conversation
party: old users may require the adoption of a limited, simple vocabulary, which includes words
familiar to them, while doctors may take advantage of a rich technical vocabulary.
By boosting cross-cultural interaction, our Guides will also facilitate people to get in touch
and to interact, hence achieving one of the main risks of ageing: social exclusion.
The targets of our investigation are indeed natural language processing, ontologies and
argumentation techniques to ensure cross-cultural interaction [
13, 14
]; agent-oriented approaches to
make such cross-cultural interaction intelligent and emotionally realistic and believable, which
requires explicit representations of the user’s state of mind [
15
]; and smart-* approaches to
make agent-oriented approaches eficient and well integrated with the existing environment
and infrastructures.
1By “BDI-oriented” here we mean, in a very broad sense, the conceptualization and implementation of the
Guides based on explicit knowledge/ beliefs, declarative and rule based reasoning/planning, explicit goals to achieve.
2. The “Guides” and Their Smart Environment
The architectural framework we have devised relies upon a “sensing layer” which is necessary
for creating smart environments where people are cared for. Improving the state of the art on the
sensing layer falls outside our research investigation: we plan to exploit as many out-of-the-box
techniques and tools as possible, among the many available ones [
16
], to allow the Guides
to sense what people are doing in an unobtrusive (or “acceptably obtrusive”) way. Among
these tools, we will consider cameras tracking people and their actions for detecting falls [
17
],
wearable and IoT devices [
18
], sleep sensors [
19
].
We are much more interested in exploring the potential of software agents as the building
blocks for analysing, designing, and developing the Guides. In particular, we are interested
in argumentation schemes that are specific for the elderly and for each culture, and in their
translation into properly formalised argumentation frameworks [
20
] where the Guides,
implemented as agents, can play a role; in the integration of ontologies and ontological reasoning
[
21
] inside such argumentation schemes; in the identification of a way to interact with the
user naturally also via voice interaction [
22
]; in the exploitation of NLP profiling techniques to
detect depression and anomalies in the emotional and cognitive status of the user [
23
].
The adoption of methods grounded on formal techniques throughout all the stages of the
Guide engineering will ensure that the actions of the Guide are trustworthy, explainable, and –
up to the extent ethics can be formalised and implemented – also ethical [
24, 25
].
A summary of the main research challenges are:
• developing an intelligent agent approach that supports natural language dialogues with
elder users that is suitable for elderly of a particular culture; the evaluation therefore
requires human subjects of diferent cultures to ensure that cultural adaptation works
well for diferent cultures as well as ensuring that the dialogues are suitable specifically
for elder users;
• connecting the agent technology with existing smart living environments and smart city
infrastructures, so that dialogues are appropriately situated;
• adding features for the intelligent agents to give emotional support for the users as well as
care for their overall health (which includes doing physical exercises, taking medication,
etc.);
• ensure that access to medical data used in dialogues with the users are ethical and that
data about living routines of the users provided to doctors are respectful of privacy.
• verifying formally that dialogues will never lead to unethical or culturally inappropriate
interactions.
To address these challenges, our vision builds on a number of technologies:
• Smart-environment and smart-city infrastructures. Since the Guides accompany
the elderly wherever they are, we need to access data from smart environment sensors
as well as inter-operate in “systems of systems" in the context for smart cities, and in
particular with hospital systems; we will rely on available standards for this, and connect
them to our multi-agent systems infrastructure to allow interconnection of Guides of
diferent people as well as between Guides and existing systems. To this aim we will exploit
the lessons we learned while engineering systems that integrate ambient intelligence/IoT
on the one hand, and agents/MAS on the other [
26, 27, 28
]. Besides existing work
emerged from the autonomous agents community, in order to immerse the Guides in a
smart environment we will also consider the potential ofered by open and configurable
frameworks like FIWARE2. Albeit not being an agent-oriented infrastructure, FIWARE
presents many features worth exploring: it has demonstrated its industrial strength
in many smart* application domains including healthcare, it implements distributed
smart components that interact asynchronously via message passing, and may be in
principle integrated with (or “under”) JaCaMo, to provide access to the surrounding smart
environment via a standard API.
• NLP profiling techniques and voice-based interaction . All the interactions between
Guides and humans will be through spoken dialogue in natural languages; although
existing tools will be used, in our vision we call for a seemless integration between
out-ofthe-box voice-based human-computer interaction tools and the sophisticated culturally
adaptive AI techniques described below, which address the communication level in a
broad sense.
• Big data and computer vision. We need to have summary information from the relevant
data for the various activities the Guides will accompany the user, for supporting medical
staf about the ongoing health state of the user, as well as symbolic representation of the
surroundings of the user; again we plan to use out-of-the-box techniques for this but
connected to our approach on representing environments for autonomous agents.
• Planning and reasoning. To provide useful information and support, and to engage
in complex dialogues, we rely on various formal techniques such as non-monotonic
reasoning, logic programming, and automated planning, adapted to the context of
transcultural smart-environment elder care. The literature on adoption of formal techniques
for modelling and implementing agent planning and reasoning mechanisms is vast and
many recent proposals may be taken under consideration for being integrated into the
Guides. They include, for example, extensions of goal-based plans used in BDI
programming languages to encapsulate both proactive and reactive behaviour, which supports
agent reasoning at runtime [
29
], contextual planning for multiple agents in ambient
intelligence environments, useful for making the plans developed by the Guides aware
of the surrounding smart environment [
30, 31, 32
], dynamic goal decomposition and
planning in scenarios characterized by a strong inter-dependency between action and
context, needed to cope with unexpected, or even catastrophic, events [
33
], automated
synthesis of a library of template-based process models that achieve goals in dynamic
and partially specified environments, which are exactly the kind of environments where
the Guides will operate [
34
].
• Ontologies. Ontologies will provide the necessary vocabulary (in various languages and
also in accordance with diferent cultures) to be used in the multi-agent dialogues that
the Guides will be able to engage. Normally the Guides only dialogue with their user, but
for example in medical consultations the Guides may need to participate in multi-agent
2https://www.fiware.org/, accessed on July 2020.
dialogues with the doctor and the user. The interplay between ontologies – and semantic
web in general – and BDI-style agents – and declarative agent approaches in general – has
been studied for a long time [
35
]. Our experience ranges from design and development of
ontologies in the health domain [
36, 37, 38
] to their integration into AgentSpeak [
13, 39
],
into MAS via CArtAgO artifacts [
40
], and into data-aware commitment-based MAS [
41
].
We will exploit this experience to provide the Guides with the semantic layer required
to boost their interaction with users. One further advantage of using ontologies is that
they could suitably cope with the dynamism that characterizes the application domain,
due not only to the dynamism of the environment, but also to the cultural specificity
of the elderly people. Many works on ontology evolution have been proposed in the
literature [
42
], including those connecting ontology evolution and belief revision that
seem extremely relevant for our research [
43
].
Finally, to make ontologies suitable to diferent cultures, profiles, ages, and genders, but
still interoperable, we plan to exploit our background on upper ontologies [
44, 45
] and
design ontologies in such a way that they have some upper layer shared among them,
and specialized sub-ontologies for diferent users and tasks.
• Argumentation. The core of the transcultural component of our vision are
argumentation protocols based on argumentation schemes (i.e., patterns of reasoning and dialogue
exchange); these are used to decide the utterances of the Guides when engaged in
dialogues. This is possible in practice given long-term work on the integration of
Argumentation Theory techniques into Jason and JaCaMo for both reasoning and communication
[
14, 46, 47
]. Also, the development of an argumentation-based inference mechanism for
BDI agents based on Toulmin’s model of argumentation [
48
] recently put forward [
49
]
can be used as an alternative basis for this part of our investigation.
• Theory of mind. “An individual has a theory of mind if he imputes mental states to himself
and others. A system of inferences of this kind is properly viewed as a theory because such
states are not directly observable, and the system can be used to make predictions about the
behavior of others” [
50
]. The theory of mind is the ability to attribute mental states – beliefs,
intents, desires, emotions, knowledge, etc. – to oneself and to others. Its philosophical
roots include the work by Dennet [
51
] that is very well known to researchers working
on BDI-oriented agent theories, languages and architectures. With aging, cognitive
abilities including theory of mind seem to decline [
52, 53
], and cultural factors impact its
development as well [
54
]. Based on these observations, another fundamental aspect of
our vision is that we are able to model the minds of users through formal representation of
their beliefs and intentions. Current dialogue systems have no such representation and the
literature in the area makes it clear that without such a representation, intelligent systems
cannot fully address the needs of their users. Our existing framework for theory of mind
relies on standard rationality assumptions. Based on our previous, recent investigations
[
55, 56, 15
] we aim at doing pioneering multi-disciplinary work on modelling the minds
of elderly and their culture, in particular what are the most appropriate ways to infer
beliefs and intentions of users given what they communicate. This is clearly specific to
the elderly public and diferent cultures, specially as the elder might have debilitating
diseases that compromise their cognitive processes or even for cultural reasons may want
to conceal certain states of mind.
• Organisations. Our approach also provides the ability to represent the various
organisations that the users are part of (for example elderly clubs, hospitals, former employees,
etc.) and this too needs to be adapted to support the diferent cultural systems where the
organisations are situated. By integrating MOISE [
57
], JaCaMo already supports the
specification and implementation of organizations [
58
], and the research on the exploitation
of organizations in MAS is still very active [
59, 60
].
• Runtime verification . Because the basis of our work is formal, this also allows us to
employ Runtime Verification (RV) techniques to ensure that Guides never make dialogue
utterances that are unethical or inappropriate for the elderly or for a particular culture.
RV can also be used at a lower, IoT, level, to check that what sensors transmit is in line
with some known pattern recognised as “normal behaviour”, to raise an alarm if sensory
inputs deviates from that pattern. Runtime verification engines based on computational
logic, like RML3 [
61
] and the trace expressions formalism it builds on [
62, 63, 64, 65
] are
a promising direction to address this challenge, and are integrated with Jason [
66
].
Clearly, to evaluate the results of this research, we will need a multidisciplinary team to
interact with elderly users, including geriatricians, as well as psychologists, sociologists interested
in population aging and philosophers interested in ethical AI systems.