Goal-Oriented Agents Lab
The Goal-Oriented Agents Lab (GOAL) is an interdisciplinary group that carry out research on finalistic behavior in intelligent agents. Key areas of activity are Cognitive Systems, Social Cognition, Action Control, Decision Making, and Emotions. Since the 70s, members of the group developed a novel approach to cognition, known as goal theory.
- anticipation and goal-directed behavior;
- belief dynamics;
- intentional action;
- intertemporal choice;
- motor control;
Via San Martino della Battaglia 44 00185 - Roma - Italy
Research at GOAL is based on the assumption that human intelligent behavior is essentially goal-oriented, and that the mind evolved to serve (and thus is constrained by) the finalistic nature of behavioral control.
A goal is defined as an anticipatory internal representation that
- can be used as a target state to guide one's conduct,
- grounds value judgments on reality, marking a state of affair as positive, negative, or neutral with respect to the agent,
- acts as a filter/constraint on cognitive processes that are not typically goal-oriented but nevertheless evolved to support goal-oriented reasoning and action (memory, belief formation, etc.).
As such, a goal is a goal already before and without being selected to actually guide behavior, and the process by which a goal becomes active and might end up orienting the agent's action (goal processing) is a key element in the study of goal-oriented behavior.
This implies that GOAL research tackles a variety of domains, from action control to decision making, from belief formation to emotion regulation, from social interaction to normative behavior, and more (for the full list, see our Research page).
Albeit crucial, goals are not the only finalistic mechanism orienting intelligent behavior: other important candidates include adaptive and social functions (e.g. reproduction is the ultimate driving force behind courtship, regardless of whether two potential partners have that goal in mind or not) and simpler anticipatory mechanisms involved in action control (e.g. the expectations embedded in so called anticipatory classifiers, where an anticipation of a future state of affair triggers a rigid response to it). At GOAL we also study how internal goals relate to external adaptive functions, and how complex goal-oriented cognition might have evolved from (and continuously interact with) simpler anticipatory mechanisms for action regulation.
Finally, the notion of goal that we endorse is operational and derived from cybernetics, and it refers to a use of an internal representation, rather than to a kind of representation (in this perspective, belief p and goal p are different uses of the same representation p). As such, it has both similarities and differences with other finalistic notions commonly employed in the study of purposive behavior, such as desires, intentions, preferences, utility, etc. Disentangling the complex relationship between these conceptual entities is also part of our job description at GOAL.
Regarding methodology, GOAL research is model-driven: we all share the aim of producing operational models of whatever phenomenon or aspects of goal-oriented behavior we focus on, instead of confining ourselves to local predictions. On a grander scale, we have the long-term ambition of connecting one day all the dots in a unified theory of cognition as goal-oriented – but there are many dots in that picture, admittedly, so we prefer to scale-up our models one step at a time.
In order to do that, we employ a combination of conceptual analysis, computational modeling, and empirical research (using both experiments, field data, and computer-based simulations). Here again we believe in the importance of "closing the loop": empirical research should be always driven by theoretical concerns and possibly paired with a clear computational model of the underlying process being studied, and in turn empirical data should feedback on both theory and model.
The following are the main lines of research currently pursued by GOAL members (click on the names associated to each area for more details):
Interrelations among cognition, motivation, and emotion (Maria Miceli, Cristiano Castelfranchi): our group contributes to, and in certain areas pioneered, the growing body of research on the relationship between cognition, motivation, and emotion. Whereas until recently emotions were considered contrary to utility and disruptive, we champion the opposite view, emphasizing the functional and adaptive role of emotions, by virtue of their anticipatory, interpretative and evaluative features. In particular, we work on:
- Appraisal, evaluation, and self-evaluation
- Cognitive functioning and motivation, especially motivated reasoning, defence mechanisms, and loss of motivation
- Cognitive anatomy of emotions, in terms of beliefs and goals compounds
Intertemporal decision-making (Fabio Paglieri): people tend to devaluate outcomes that are far away in the future, and the rate and shape of such temporal discounting has important implications for self-control and long-term planning. However, not much is known on the cognitive mechanisms and evolutionary pressures responsible for this behavior, and our research addresses this gap. In particular, we focus on:
- Comparative and developmental studies on intertemporal choice
- Cultural and context-based variations in temporal preferences
- Psychologically plusible criteria for diachronic rationality
Argumentation and goals (Fabio Paglieri, Cristiano Castelfranchi): people typically argue to pursue goals that are beyond the scope of the dialogue itself, and yet such extra-dialogical goals exert a crucial infuence on the argumentative process. This fact, and more generally the importance of decision-making for arguments, has been somehow neglected in argumentation theories. In contrast, we focus on the following:
- Resource-bounded rationality in argument production and interpretation
- Argumentative decisions and their motives
- Integration of argumentation and belief change