Modulation Scheme for Biasing the Emotional Process of Autonomous Agents: A Component-Based Approach

Introduction

Autonomous Agents (AAs) are software entities that carry out tasks on behalf of users or other programs with a certain degree of independence and autonomy. In doing so, AAs make use of knowledge about the environment and representations of desires and goals (Chiriacescu, Soh, & Shell, 2013). This type of intelligent system has been crucial for the advance of fields such as software engineering, human-computer interaction, and artificial intelligence. In these fields, AAs have been designed to carry out tasks that require the imitation of human cognitive functions, including decision making, planning, and reasoning, to name a few (Ligeza, 1995; Maes, 1995; Sun, 2009).

Develop components inspired by human cognitive functions allows AAs to perform more complex tasks minimizing human intervention. That is why researchers in artificial intelligence (AI), human computer interaction (HCI) and software engineering focus on improving the problem solving, reasoning and communication skills of AAs. On the one hand, the research community in the AI field has devoted efforts to create human-like systems for communication and reasoning as well as to reproduce in computer environments the brain processes that perform them (Gubbi, Buyya, Marusic, & Palaniswami, 2013). On the other hand, in the HCI field some interfaces and mechanisms that improve the interaction of these systems with other agents (computational or human agents) have been developed (Martínez-Miranda & Aldea, 2005; Perlovsky & Kuvich, 2013).

According to literature (Ben Ammar, Neji, Alimi, & Gouardères, 2010; Pérez, Cerezo, Serón, & Rodríguez, 2016), it is notable the importance of AAs having the abilities for agent-to-agent and human-to agent communication, coordination, negotiation in order to achieve its objectives. Using components inspired by cognitive aspects of the human brain has allowed creating systems whose behavior is similar to humans. Nowadays, the use of the effect as another cognitive component has become popular in these types of systems, because emotions are immersed in all expressions of human behavior and intelligence. (Brackett & Salovey, 2006).

Researchers in psychology argues, that human emotions can be seen as a process that involves a subjective appraisal of significant events as well as the preparation of the organism for dealing with such events (LeDoux & Hofmann, 2018; Rukavina et al., 2016a). Evidence shows that emotions influence cognitive functions (Ayesh, Arevalillo-Herráez, & Ferri, 2016; Phelps, 2006) modifying the normal operation of processes such as attention, perception, and learning. Psychologically, emotions alter attention, trigger certain behaviors, and activate relevant associative networks in memory. According to Phelps (2006), emotions are necessary to establish long-term memories. In addition, they provide opportunities for learning, from simple reinforcement learning to complex and conscious planning (LeDoux, 2000; Novak & Gowin, 1985).

As mentioned earlier a key objective of artificial intelligence is the development of software systems capable of doing complex tasks that produce intelligent responses (Perlovsky & Kuvich, 2013), systems that act and reason like humans. In this context, the literature reports an increasing interest in the development of AAs with abilities to evaluate and respond to emotional stimuli, (Brown et al., 2015; Dias, Mascarenhas, & Paiva, 2014; Rukavina et al., 2016b; Wang et al., 2012). Recent works have proposed the incorporation of affective processing in AAs by designing computational models of emotions (CMEs), which are software systems designed to imitate the mechanisms of the human emotional process (emotional evaluation of perceived stimuli, elicitation of emotions, and generation of emotion-driven behaviors) in computing environments (L.-F. Rodríguez, Ramos, Rodriguez, & Ramos, 2014; Luis-Felipe Rodríguez & Ramos, 2015).