Carles Sierra, Juan A. Rodríguez-Aguilar, Luís Manuel Artiles Martínez

Task allocation for human teams is of paramount importance in a plethora of real-
world settings. Teams bring together individuals with different competencies, interests
and perspectives, enabling them to tackle complex challenges that a single person cannot
handle due to lack of resources (e.g., knowledge and skills) or time. Effective teamwork
fosters a sense of belonging, shared purpose, and commitment among team members,
driving them to put in extra effort, remain focused on their goals, and ultimately reach
high-quality outcomes. From workplaces to educational settings and community activi-
ties, forming and allocating teams is crucial for achieving success. In this dissertation, we
tackle the problem of trustworthy task allocation for human teams. Specifically, we con-
tribute by putting forward tools to aid the process towards effective teamwork.
First, we review the literature regarding teams and team formation across several sci-
entific domains, including Computer Science, Organisational Psychology, Motivational
Psychology and Social Sciences. We study on which bases teams are formed in the differ-
ent scientific areas, and we explore which human characteristics influence teamwork and
team performance.
Second, we use the findings from the literature and we put together important hu-
man characteristics that benefit teamwork. We propose metrics that allow us to evaluate
a team across these characteristics. In particular, we discuss how to aggregate from an in-
dividual level to a team level several human characteristics such as competencies, person-
ality, gender, preferences and interpersonal relations. We propose four such aggregating
metrics, namely the competence affinity, congeniality, motivation and social cohesion. We
also introduce collegiality, a metric that considers the beneficial-to-teamwork individual
characteristics and can be used as a predictor for team performance.
Third, we study the problem of forming teams. In particular, we focus on settings
that involve multiple tasks and require teams that each team works on a different task,
while each individual can participate in at most one team. Hence, we define the Non Over-
lapping Many Teams to Many Tasks Allocation Problem (NOMTMT-AP). We show that
the NOMTMT-AP is NP-complete, and we put forward two algorithms for solving the
problem: an optimal solver and Edu2Com, an anytime heuristic solver. We conduct a
manifold empirical evaluation. Our evaluation allowed us to study (i) the quality, run-
time and anytime behaviour of Edu2Com when pitched against the optimal solver, (ii)
the solubility of Edu2Com along with the limitations of the optimal solver, and (iii) the
team performance when the teams are formed considering the individuals’ competencies,
personality, gender, preferences and interpersonal relations.
Fourth, towards trustworthiness, we address the problem of explaining why a team
formation algorithm formed the teams it outputs and not others. In this direction, we
identify a collection of questions that are intuitive and meaningful and cover the main
points of interest regarding team formation scenarios. Then, we introduce a general ex-
planatory algorithm that can wrap an existing team formation algorithm without modifying

it and build contrastive explanations. We conduct an empirical evaluation and show
that our algorithm builds contrastive explanations are easy to understand, requiring just
the reading level of a high-school student. Along with explaining team formation scenar-
ios, we turn our attention to a vital challenge regarding explanations. Specifically, we ad-
dress the problem of preserving privacy upon providing explanations. In this light, we put
forward a privacy breach detector that assesses whether an explanation is bound to reveal
private information. Finally, we propose a general framework that describes the interac-
tions between a team formation algorithm, an explanatory algorithm and a privacy breach
detector.