Communicative Capital for Prosthetic Agents

Abstract

This work presents an overarching perspective on the role that machine intelligence can play in enhancing human abilities, especially those that have been diminished due to injury or illness. As a primary contribution, we develop the hypothesis that assistive devices, and specifically artificial arms and hands, can and should be viewed as agents in order for us to most effectively improve their collaboration with their human users. We believe that increased agency will enable more powerful interactions between human users and next generation prosthetic devices, especially when the sensorimotor space of the prosthetic technology greatly exceeds the conventional control and communication channels available to a prosthetic user. To more concretely examine an agency-based view on prosthetic devices, we propose a new schema for interpreting the capacity of a human-machine collaboration as a function of both the human's and machine's degrees of agency. We then introduce the idea of communicative capital as a way of thinking about the communication resources developed by a human and a machine during their ongoing interaction. Using this schema of agency and capacity, we examine the benefits and disadvantages of increasing the agency of a prosthetic limb. To do so, we present an analysis of examples from the literature where building communicative capital has enabled a progression of fruitful, task-directed interactions between prostheses and their human users. We then describe further work that is needed to concretely evaluate the hypothesis that prostheses are best thought of as agents. The agent-based viewpoint developed in this article significantly extends current thinking on how best to support the natural, functional use of increasingly complex prosthetic enhancements, and opens the door for more powerful interactions between humans and their assistive technologies.

Pavlovian Signalling with General Value Functions in Agent-Agent Temporal Decision Making

“…At a high level, the results in this work support the use of Pavlovian signalling as a lens to study certain agent-agent relationships. As identified by Pilarski et al [25], it is possible to frame dyadic partnerships between agents in terms of the agency and capacity of the parties engaged in the interaction; while capacity of a partnership might be limited by having reduced agency by one of the parties (here the co-agent), the simplicity of that partner provides the opportunity for fast learning of its behaviour by the other party [24,25]. This is the case with Pavlovian signalling: we see that as a main benefit useful co-agent predictions could be learned rapidly from a blank slate (less than 500 steps in the abstract domain, or less than a minute in the VR domain) (Figs.…”

“…In addition to rapid co-agent learning, and the learnability of co-agent signals by the main agent, we see evidence that there are further benefits to Pavlovian signalling in that it assumes nothing about the internal structure or mutability of the control agent; human and machine agents alike can approach co-agent signals from an ungrounded perspective. Further, there is no reason to limit the co-agent to only observations of the environment as we have done in these studies; it is natural to think that information pertaining to the agent and its adaptability can also be used as inputs to the co-agent, further increasing its ability to make relevant predictions and to begin to build what has previously been termed as communicative capital [25]. As such, Pavlovian signalling appears to be a viable stepping stone between between fixed agent-agent signalling and bidirectionally learned signalling relationships.…”

Examining the Use of Temporal-Difference Incremental Delta-Bar-Delta for Real-World Predictive Knowledge Architectures

“…In particular, anticipation and adaptation are highly important given the world and tasks encountered by a prosthetic limb are continuously changing. Therefore, the arm is an interesting showpiece as an autonomous learner (Pilarski et al, 2017).…”

“…Detailed demonstrations of other potential of predictive knowledge architectures in real-world domains have been offered in industrial laser welding (Günther et al, 2016), robot navigation (Kahn et al, 2018), animal models of partial paralysis (Dalrymple et al, 2018), and artificial limbs (Pilarski et al, 2013;Sherstan et al, 2015;Edwards et al, 2016;Pilarski et al, 2017). Recently, work has focused on using predictive knowledge to construct representations of state that capture aspects of the environment that cannot be described by current observations alone (Schlegel et al, 2018), and on accelerating the learning of predictive knowledge through use of successor representations (Sherstan et al, 2018).…”

Learned human-agent decision-making, communication and joint action in a virtual reality environment

“…In what has come to be defined as joint action, humans coordinate with other humans to achieve shared goals, sculpting both their actions and their expectations about their partners during ongoing interaction (Sebanz et al 2006;Knoblich et al 2011;Pesquita et al 2018). Humans now also regularly make decisions in partnership with computing machines in order to supplement their abilities to act, perceive, and decide (Pilarski et al 2017). It is natural to expect that joint action with machine agents might be able to improve both work and play.…”