The project
Overview
The World Health Organization (WHO) defines health literacy as “the ability to access, comprehend, evaluate and communicate information to promote, maintain and improve health in a variety of settings across the life course". Since 2000, when a low level of health literacy was denounced by the WHO itself, the topic has received high attention. Low health literacy levels are, in fact, potential causes for public health endangerment and inequalities in accessing medical facilities. Low health literacy relates to a lower engagement with health services and lack of understanding of medication instructions.
Studies concerning the recent COVID-19 pandemic highlighted that misinformation exposure was associated with misinformation belief, and that misinformation belief was linked to fewer preventive behaviours. For the case of neurological disorders, low health literacy has consequences beyond the health of the patient: prejudice and stigma plague societies as fear caused by lack of understanding further weighs the affected persons. Epilepsy disorders have a long history, in this sense, and the efforts to support patients suffering from it have multiplied, in recent years.
Although the incidence of active epilepsy is high throughout the world, long-standing ignorance and stigma about the disorder is still standing. Disseminating knowledge about epilepsy even among people who do not suffer from epilepsy has, therefore, the potential to have a substantial impact both on the public health level, by informing people about how to deal with seizures, and on the societal level, reducing prejudice and negative stereotypes. For the specific case of epilepsy, a recent study reported that “adults need more information about epilepsy, appropriate seizure first aid training, recognition of seizure symptoms, and familiarity with resources for epilepsy information”. Low literacy about epilepsy has also started to be tackled by implementing specific interventions targeted at pre-school children.
As communication changes, digital approaches to knowledge dissemination must keep the pace to counter the rapid spread of misinformation on digital media, as they appear to have originated mostly on social media. The rapid advancement of virtual assistants, given their attractiveness and the technological push towards synthetic humans, both virtual and physical, makes it important to anticipate the next wave of interactive technologies based on natural language and embodied conversational agents, either in the form of Virtual Humans or in the form of Social Robots. Given the attractiveness and engaging power of these rapidly growing technologies, it is important to investigate how to a) develop theoretical models of dialogue management that support argumentative dialogues and b) implement these models in technological frameworks that stay up to date with the continuously changing interaction paradigms set by the entertainment industry.
Specifically, argumentative dialogue capabilities would allow a virtual agent to counter users’ statements by presenting counterfactual knowledge to limit the spreading of false information.
Topics
Details
Classic approaches to ABD adopt the same setting that has been successfully used for argumentation based inference: that is, inference rules are derived to establish a course of action that is deterministic given a system configuration. Structural relationships among claims and various kinds of replies are established in a formal protocol dedicated to establishing whether a speech act is legal or not. This allows to provide a formal description of situations when a dialogue terminates or, in the case of competitive settings, is won. Persuasion is the most studied situation in ABDs: in this type of setting, a claim provided by an agent A is supported by data, constituting an argument that can be explicitly put forward as a reply to a why move made by an agent B, which explicitly requests the speaker to explain the reasons why a statement should be accepted. Claims can be attacked by counter-arguments, which are other claims aimed at proving previous statements as false. Conceding and retracting moves respectively declare the acceptance of a statement or a change of attitude towards it, from commitment to non-commitment. Note that this does not imply a change of belief, as it is usually specified that the publicly declared position of an agent may not reflect what the agent believes.
​
The problem that characterises Argumentation Based Dialogue with respect to Argumentation Based Inference is the presence of different agents in the setting. This introduces multiple, not necessarily aligned, knowledge bases and, possibly, different/conflicting goals in the pursuit of a solution to a problem. While there are attempts to deal with the partial knowledge each agent has concerning the others' goals and knowledge using rule-based systems, these approaches have been recently surpassed by more flexible, probabilistic approaches, modelling opponents in terms of probability distributions over their possible beliefs and goals and using these to compute the utility of each legal dialogue move depending on their own goals and beliefs. Moreover, recent work put forward the need to model the degree or strength of an agent's belief towards a statement, modelled as the probability of the statement being true, rather that assuming it to either be or not be true.
From the technological point of view, specialised frameworks to manage virtual humans are presently available. In these frameworks, game engines have been sometimes used but only as rendering modules. Modern game engines, however, are serious candidates to host most of the behavioural logic and the realisation modules in an integrated solution. The number of possibilities offered by these advanced software tools poses a question with respect to the externalisation of services that the game engine would be able to provide itself. In particular, the fast-paced advances in the design of interactive experiences on an industrial level must be matched by the academy to avoid newly developed interfaces being damaged by comparison with the extremely competitive design of digital entertainment applications.
In the Human Computer Interaction field, where interface design is crucial, dealing with users' expectations may become a serious challenge as the gaming industry continues to grow, since ``Entertainment gaming experiences color players' interactions with other digital media, setting expectations for user experience and interactivity''. Since the industry itself provides the technology needed to support this design process by making it easily accessible, connecting the development of applications designed for academic purposes to the advances of industrial solutions helps researchers not to fall behind when the perceived quality of an interactive system depends also on comparison with the standards set by the gaming industry.
​
The Framework for Advanced Natural Tools and Applications with Social Interactive Agents (FANTASIA) was developed to integrate a modern game engine (Unreal Engine) and its development toolkit with ease of access to AI capabilities. The Unreal Engine has recently been applied to several different applications beyond the digital games industry and it is now considered one of the main environments in which to develop Real Time Interactive 3D (RTI3D) applications. FANTASIA allows to use AI services from high-profile providers from inside the Unreal Engine, it integrates the AGruM library to implement Bayesian Networks and it provides access to a graph database, Neo4j. This is an innovative technology that has been gaining popularity in the last years for its representation power with respect to traditional databases and for the integration with Graph Data Science algorithms.
​
FANTASIA enables developers working with Embodied Conversational Agents to leverage on the powerful tools for interactive experiences design provided by the industry while using tools that allow the development of advanced AIs. It supports the integration of graph-based reasoning and Bayesian decisions with the AI framework that is built in the Unreal Engine, which is based on Behaviour Trees. The combination of all these different approaches to AI implementation in a single development experience provides the foundations for an harmonic formalisation of the different processes involved in ABD management.
Concluding
Summary
The SPECIAL project is highly multidisciplinary and covers three axes of investigation concerning public health, technology, and linguistics. The general goal is to develop theories and technologies for dialogue management allowing virtual agents to increase digital health literacy. This is an important goal, in general, as a better understanding of health issues from the public has a positive impact both from the economic and from the societal point of view. It is of critical importance to increase digital health literacy towards neurological disorders, as people who suffer from these are often subject to prejudice and stigma. The SPECIAL project aims at exploring how the next generation of virtual assistants, Embodied Conversational Agents, can provide support in this.