The cognitive mechanisms that are currently implemented in robotic systems focus mainly on the spatial aspects of the world, resulting in artificial agents which are “stuck in time” (a phrase borrowed from Roberts, 2002).
Temporal cognition: a key ingredient of intelligent systems
Computational Vision and Robotics Laboratory, Institute of Computer Science, Foundation for Research and Technology Hellas, Heraklion, Crete, Greece
The investigation of the brain mechanisms involved in the perception and processing of time has attracted significant research interest in brain science during the last decade. Contemporary review papers and special journal issues have summarized the new and burgeoning scientific findings in the field (Allan and Church, 2002; Szelag and Wittmann, 2004; Meck, 2005; Crystal, 2007; Ivry and Schlerf, 2008; Tarlaci, 2009; Wittmann and van Wassenhove, 2009).
Definition: Temporal Cognition encompasses the set of brain functions that enable experiencing the flow of time and processing the temporal characteristics of real world phenomena, accomplishing (i) the perception of synchrony and ordering of events, (ii) the formation of the experienced present, (iii) the perception of different temporal granularities, (iv) the conceptual abstraction and processing of durations, (v) the mental traveling in future and past time, (vi) the social sharing of temporal views about the world.
It is now well established that, despite the fundamental role of time in our life, there is no region in our brain that is solely devoted to the sense of time (this contrasts to the exclusive representation of audition, vision, touch, proprioception, taste, and other senses in specific cortical regions). However, over the past decade, a number of different brain areas have been implicated to contribute in time-experiencing including (among others), the cerebellum, right posterior parietal cortex, right prefrontal cortex, fronto-striatal circuits, and insular cortex for duration perception (Lewis and Miall, 2003; Hinton and Meck, 2004; Bueti et al., 2008; Ivry and Schlerf, 2008; Wittmann, 2009), the inferior frontal and superior temporal lobes, hippocampus, medial prefrontal, medial parietal and posterior cingulated cortex for past–future distinction, and mental time travel (Botzung et al., 2008;Suddendorf et al., 2009; Viard et al., 2011), the prefrontal, inferior parietal cortex, superior colliculus and insular cortex for synchronous, and asynchronous event distinction (Dhamala et al., 2007; Kavounoudias et al., 2008), the posterior sylvian regions, posterior parietal, and temporo-parietal networks for temporal order judgment (Woo et al., 2009;Bernasconi et al., 2010; Kimura et al., 2010). The involvement of many brain areas in TC is explained by the significant contribution of multiple cognitive processes such as attention, working-memory, decision making, emotions, etc., in experiencing and processing time (Livesey et al., 2007). Therefore, slight perturbations on these processes may affect our time experiences, explaining why subjective time (how each one of us is perceiving the flow of time) is in principle different than the objective, physical time (Searle, 1992).
Experiencing the flow of time is an important capacity of biological systems that is involved in many ways in the daily activities of humans and animals. However, in the field of robotics, the key role of time in cognition is not adequately considered in contemporary research, with artificial agents focusing mainly on the spatial extent of sensory information, almost always neglecting its temporal dimension. This fact significantly obstructs the development of high-level robotic cognitive skills, as well as the autonomous and seamless operation of artificial agents in human environments. Taking inspiration from biological cognition, the present work puts forward time perception as a vital capacity of artificial intelligent systems and contemplates the research path for incorporating temporal cognition in the repertoire of robotic skills.
The majority of existing computational models dedicated to time processing focus on the duration estimation aspect of TC (Matell and Meck, 2004; Zakay and Block, 2004; Machado and Arantes, 2006;Arantes, 2008). Additional models have been recently implemented to address mental time travel into the past (Hasselmo, 2009; Polyn et al., 2009; Hasselmo et al., 2010), without however considering future time traveling.
In practical terms, we can identify at least three dimensions in which TC can improve robotic cognition.
• Advance internal cognitive processes: There are many mechanisms with an important role in shaping cognitive dynamics, such as learning, memorization, forgetting, attention, association, and others, that can significantly benefit by considering temporal information. For example, new learning algorithms may be implemented that consider the details of past events when adjusting decision making procedures, time-based association mechanisms may be used to enable future conflict prediction, while directing attention on a particular time period in the past will enable considering relations between a specific set of events.
• Develop skills dealing with the manipulation of time: Artificial TC will provide robotic agents with the capacity to process all different aspects that time is involved in our daily life, accomplishing tasks which are currently out of their scope. For example, robots may be capable of (i) synchronizing with natural human actions (currently humans are mainly synchronized to robots); (ii) abstracting and categorizing the time scales required for the evolution of different processes; (iii) being aware of the temporal order of their own experiences; (iv) considering the causal relationship linking the present and future with past events that may have occurred many hours or days ago, and others.
• Develop skills that implicitly involve time processing: Time is an important parameter for many low and high-level skills. This is because even simple actions (e.g., object grasping) include a critical “when” component (Battelli et al., 2008) that links a given behavior with the ongoing real world processes. Moreover, high-level cognition that is typically less related with the here and now of the world, requires the association and reasoning on events that occurred, or will occur at different times (e.g., mind reading links past knowledge with future actions). Therefore, both low and high-level cognitive skills can gain significant efficiency through artificial TC.
Organisms are equipped with value systems that signal the salience of environmental cues to their nervous system, causing a change in the nervous system that results in modification of their behavior. These systems are necessary for an organism to adapt its behavior when an important environmental event occurs. A value system constitutes a basic assumption of what is good and bad for an agent. These value systems have been effectively used in robotic systems to shape behavior. For example, many robots have used models of the dopaminergic system to reinforce behavior that leads to rewards. Other modulatory systems that shape behavior are acetylcholine’s effect on attention, norepinephrine’s effect on vigilance, and serotonin’s effect on impulsiveness, mood, and risk. Moreover, hormonal systems such as oxytocin and its effect on trust constitute as a value system. We invite papers on research involving neurobiologically inspired robots whose behavior is: 1) Shaped by value and reward learning, 2) adapted through interaction with the environment, and 3) shaped by extracting value from the environment.