9:00 Welcome
9:30 JointAction4HRI project presentation
9:45 On a Multidisciplinary Analysis of Joint Activity
Kathleen Belhassein - LAAS-CNRS & CLLE LTC
The particularity of the JointAction4HRI project is to bring together different fields of social and engineering sciences around the same issue: gaining a better understanding of the mechanisms involved in joint action, in particular in the context of human-robot interactions. Our goal is to question the use of the terms joint action, joint attention, collaboration, cooperation or coordination by gathering the contributions of these different disciplines. Indeed, an identical term can be defined differently depending on the discipline or the authors (Parker, 2008; Vesper et al., 2017) while cooperation (Colombi et al., 2009), collaboration (Mutlu, Terrell & Huang, 2013) or joint action (Fiebich & Gallagher, 2012; Sebanz, Bekkering & Knoblich, 2006) are sometimes used to designate the same phenomenon. Besides this diversity in terminology, we will consider the different mechanisms associated to these terms. To what extent does joint attention differ from shared attention? Is having a common goal mandatory to speak of joint action or collaboration, or can we define as such walking side by side on a footpath? Must we necessarily reason on the other’s knowledge and intentional states to establish a common ground and be able to act jointly? Beyond making available clear and explicit terminology, comparing the variation of uses of these terms and their associated concepts from different disciplinary perspectives is an important milestone to better understand how we manage to act together and be good partners. This last point in particular is crucial to improve the interaction between humans and assistive robots.
10:15 Discussion
Moderator : Michele Guidetti
10:45 Coffee break
11:15 Natural Interactions: Theoretical and Motor Aspects of JointAction4HRI
Stephen Butterfill - University of Warwick
The JointAction4HRI project aims to characterise ‘ingredients required for the successful realization of a collaborative task’. In the first part of this talk, I argue that doing so requires separating the thing to be explained (acting together cooperatively and with a purpose) from the things which explain it (such as intentions and knowledge states, perhaps). This is a problem because the leading, best developed attempts to characterise ingredients of joint action do so by specifying mental states and processes which might explain how those ingredients occur. But the problem can be solved by taking a theoretically modest approach. In the second part of this talk, I review some discoveries about the role of motor representation in joint action (especially della Gatta et al, 2017; Sacheli et al, 2018). These suggest a theory of how humans collaborate in very small-scale joint actions, actions such as playing a chord together or clinking glasses. The theory leads to a challenge given JointAction4HRI’s commitment to the idea that ‘a human agent should be able to engage in an interaction with a robot in a natural way’ together with the assumption that some artificial agents will be attractive partners because they have abilities humans lack (and conversely). The challenge is to understand how, despite the role of motor representation in joint action, individuals with different action abilities might interact in a natural way when performing very small-scale joint actions.
11:45 Discussion
Moderator : Hélène Cochet
12:00 Narrative structuring of experience for extended social interactions in humans and robots
Peter Ford Dominey - INSERM & CNRS
One of the key results in modern cognitive neuroscience is that the human ability to understand and interact in the world is based on experience, via the body, in the physical and social world. A second, older and less appreciated result from developmental psychology demonstrates the crucial role of narrative in organizing this experience and allowing the developing child to make meaning from experience. This implies that the future ability of robots to understand and interact socially with humans will rely on memory systems that allow these robots to accumulate and integrate experience over extended periods of interaction, and to be capable of enriching this experience through narrative input from humans, in a developmental approach. We have begun to address this by developing autobiographical memory systems that (a) encode extended interaction experience between robots and humans, and (b) allow for the consolidation and extraction of semantics from this experience inspired by human memory systems. More recently we have introduced a narrative language capability, such that graph-based representations called situation models are constructed from experience encoded in the ABM, and can then be enriched by narration from the human partner, thus allowing the robot to have a deeper understanding of unseen causal relations between mental motivational states, and resulting actions. This understanding is reflected in the robot’s ability to generate simple narrative, as well as understand it. In return, we demonstrate how these enhanced memory systems could be used by humans who have deteriorated memory in normal and pathological aging.
12:30 Discussion
Moderator : Rachid Alami
12:45 Lunch
14:00 Joint Actions, Commitments and Need to Belong
Víctor Fernández Castro - University of Granada / Jean Nicod Institute
Among the myriad of mechanisms involved in joint actions, commitments appear to play a key role. Commitments can dramatically improve the performance of joint actions by stabilizing expectations, reducing the uncertainty of the interaction, providing moral motivations to cooperate or improving coordination in cooperation/coordination problems (Michael and Pacherie, 2015). However, the credibility of commitments proves not to be a straightforward matter, especially when one notes that credibility depends upon the motivation of the committed agent to honor her commitment in many situations where alternative options that maximize her interests are available. Let us call this challenge the human credibility problem.
Two available approaches to the human credibility problem claims that human preferences for fulfilling commitments relies on reputation management and socio-emotional mechanisms. After presenting several empirical and theoretical concerns about these approaches, we present an alternative in terms of social belonging. According to this alternative, human preference for honoring their commitments lies in the need of individuals for having frequent, positively valenced interactions with other people within a framework of long-lasting concern for each other’s welfare. We argue that such alternative can avoid the central problems of its contenders and account for some recent empirical findings regarding commitments.
14:35 Discussion
Moderator : Elisabeth Pacherie
14:45 Ingredients of Joint Action: A View from Cognitive Psychology
Guenther Knoblich - Central European University
What are the key processes and mental representations involved in joint action? One approach to address this question prominent in philosophy and developmental and comparative psychology is to define benchmarks for interpersonal constellations of knowledge states that distinguish joint action from other forms of interaction. For HRI this approach is of limited use because robots interacting with humans do not mainly need to determine whether they and their partners have the appropriate knowledge states to perform a ‘real’ joint action. Rather they need to get things done with humans. This involves 1) identifying and evaluating opportunities for acting together, 2) dynamically specifying and adjusting plans involving their own and others’ contributions, 3) swiftly performing, monitoring, and adjusting actions that require a high degree of spatial and temporal coordination, 4) learning to perform coordinated actions from observing others’ coordinated actions, and so on. Thus, getting things done with others can involve a variety of perceptual, decision, planning, and motor processes operating on different kinds of representations. A useful approach for HRI could be to construct robots that perceive, think, and act like humans who perform actions together. In my talk I will provide an update of relevant research on humans and identify knowledge gaps that work against a better integration of cognitive psychology and HRI research.
15:15 Discussion
Moderator : Aurélie Clodic
15:30 Coffee break
16:00 Panel
17:30 End
