Review
Action observation and robotic agents: Learning and anthropomorphism

https://doi.org/10.1016/j.neubiorev.2011.03.004Get rights and content

Abstract

The ‘action observation network’ (AON), which is thought to translate observed actions into motor codes required for their execution, is biologically tuned: it responds more to observation of human, than non-human, movement. This biological specificity has been taken to support the hypothesis that the AON underlies various social functions, such as theory of mind and action understanding, and that, when it is active during observation of non-human agents like humanoid robots, it is a sign of ascription of human mental states to these agents. This review will outline evidence for biological tuning in the AON, examining the features which generate it, and concluding that there is evidence for tuning to both the form and kinematic profile of observed movements, and little evidence for tuning to belief about stimulus identity. It will propose that a likely reason for biological tuning is that human actions, relative to non-biological movements, have been observed more frequently while executing corresponding actions. If the associative hypothesis of the AON is correct, and the network indeed supports social functioning, sensorimotor experience with non-human agents may help us to predict, and therefore interpret, their movements.

Highlights

► The ‘action observation network’ responds more to human than non-human movement. ► Stimulus form and kinematics influence tuning, but not knowledge about identity. ► Associative learning is likely to explain the origins of the biological tuning. ► Sensorimotor experience with robots may help us to predict their movements.

Introduction

Observation of actions activates the motor codes required for their performance. For example, it has been shown behaviourally that we automatically imitate others, when there is no intention to have done so, and no reported awareness of having done so (e.g. Chartrand and Bargh, 1999). In line with such behavioural data, neuroimaging studies have shown that observing action activates an ‘action observation network’, including ventral and dorsal premotor cortices, primary motor cortex, and inferior parietal lobule (Rizzolatti et al., 1996, Buccino et al., 2001, Grezes and Decety, 2001, Gazzola and Keysers, 2008, Kilner et al., 2009). Some of these areas, namely ventral premotor cortex and inferior parietal lobule, correspond to those in which ‘mirror neurons’ have been found in the macaque monkey: These neurons discharge not only when the monkey executes an action of a certain type (e.g. precision grip), but also when it observes the experimenter performing that action (Di Pellegrino et al., 1992, Gallese et al., 1996, Gallese et al., 2002; note that neurons with similar properties have also been found in primary motor cortex and dorsal premotor cortex (Cisek and Kalaska, 2004, Dushanova and Donoghue, 2010)).

The AON is biologically tuned, such that it responds more to the observation of human, than non-human, movement (either defined by form or kinematic profile). This biological tuning may be crucial for sociocognitive functioning, which the AON is hypothesized to support (e.g. Gallese and Goldman, 1998). On the basis of such hypotheses, some have suggested that when the AON is active during the observation of non-human agents like humanoid robots, it is a sign of the ascription of human properties such as mental states to these agents (e.g. Oberman et al., 2007, Gazzola et al., 2007, Chaminade and Cheng, 2009). For example, Oberman et al. (2007) claim that ‘the implication is that the human mirror neuron system may be activated as a result of the human interactant anthropomorphising these robots. Indeed, by activating the human mirror neuron system humanoid robots could potentially tap into the powerful social motivation system inherent in human life, which could lead to more enjoyable and longer lasting human-robot interactions’ (p. 2195). Similarly, Gazzola et al. (2007) say ‘now we know, that our mirror neuron system may be part of the reason why, when in Stars Wars, C3PO taps R2D2 on the head in a moment of mortal danger, we cannot help but attribute them human feelings and intentions, even if their physical aspect and kinematics are far from human’ (p. 1683). Furthermore, Chaminade and Cheng (2009) state ‘the underlying assumption is that the measure of… (AON activation) indicates the extent to which an artificial agent is considered as a social inter-actor’ (p. 289).

This review will outline evidence of biological tuning in the AON. It will consider the AON to be a mechanism which translates an observed action into motor codes required for execution. It will therefore cover behavioural studies indicating operation of such translation processes (see Heyes, in press) and neurological studies suggesting activation of components of the motor network when observing actions, including primary motor cortex, and ventral and dorsal premotor cortices; both BA6 and BA44. This range of coverage is not assuming that activations in different components of the motor network will all necessarily signal the same processes; it simply reflects that, on the basis of present theorizing about the AON, the components cannot be divided functionally with confidence. It will examine the features of observed actions which generate biological specificity, and conclude that there is evidence for tuning to both the form and kinematic profile of observed movements, and little evidence for direct tuning to belief about identity. It will subsequently propose that biological tuning in the AON is a result of more frequent and systematic observation of human actions while executing corresponding actions. If the AON develops through learning, and it indeed supports social functions such as action understanding, sensorimotor experience with agents may help us to predict, and therefore interpret, their movements.

Section snippets

Biological tuning in the AON

Kilner et al. (2003) showed that the execution of sinusoidal arm movements in a vertical or horizontal plane was subject to interference from simultaneous observation of another human performing arm movements in the opposite plane; if participants executed vertical arm movements while observing horizontal movements, there was greater variance in the horizontal dimension, compared with conditions where they observed vertical movements (Fig. 1a). This ‘interference effect’ is thought to be a

Theories about the origin of biological tuning

There are two prominent theories concerning the origins of the AON. The first posits that the network evolved through natural selection to support higher level sociocognitive functioning, such as theory of mind (Gallese and Goldman, 1998, Rizzolatti and Sinigaglia, 2010). In contrast, the associative sequence learning model (ASL, e.g. Heyes, 2001, Heyes, 2010) suggests that the AON acquires its mirror properties through sensorimotor learning (see Chaminade et al., 2008 for a simulation).

Anthropomorphism

It has been suggested that when the AON is active during the observation of non-human agents like humanoid robots, it is a sign of the ascription of human mental states to these agents (e.g. Gazzola et al., 2007). If the AON evolved through natural selection to support mental state inferences (Gallese and Goldman, 1998, Rizzolatti and Sinigaglia, 2010), that is, it is the function for which the network evolved, it may be possible to predict that its activation signals mental state inference.

Conclusion

The present review has outlined evidence to suggest that the AON, which is thought to translate an observed action into the motor codes required for its execution, responds more to the observation of human, than non-human, movement. It examined the features which generate this biological specificity, and concluded that there is evidence for tuning to both the biological form and kinematic profile of observed movements. It found little evidence for tuning to beliefs about stimulus identity. It

Acknowledgments

CP was supported by an interdisciplinary postdoctoral fellowship awarded jointly by the Medical Research Council and the Economic and Social Research Council. I am grateful to Geoffrey Bird, Cecilia Heyes and James Kilner for comments on an earlier version of the manuscript.

References (88)

  • V. Gallese et al.

    Mirror neurons and the simulation theory of mind-reading

    Trends Cogn. Sci.

    (1998)
  • V. Gazzola et al.

    The anthropomorphic bra the mirror neuron system responds to human and robotic actions

    Neuroimage

    (2007)
  • E. Gowen et al.

    Exploring visuomotor priming following biological and non-biological stimuli

    Brain Cogn.

    (2010)
  • E. Gowen et al.

    Movement interference in autism-spectrum disorder

    Neuropsychologia

    (2008)
  • C.M. Heyes

    Causes and consequences of imitation

    Trends Cogn. Sci.

    (2001)
  • C. Heyes

    Where do mirror neurons come from?

    Neurosci. Biobehav. Rev.

    (2010)
  • C.M. Heyes et al.

    Experience modulates automatic imitation

    Cogn. Brain Res.

    (2005)
  • M. Jonas et al.

    Do simple intransitive finger movements consistently activate frontoparietal mirror neuron areas in humans?

    Neuroimage

    (2007)
  • S.S. Jones

    Exploration or imitation? The effect of music on 4-week-old infants’ tongue protrusions

    Infant Behav. Dev.

    (2006)
  • K. Kessler et al.

    Investigating the human mirror neuron system by means of cortical synchronization during the imitation of biological movements

    Neuroimage

    (2006)
  • J.M. Kilner et al.

    An interference effect of observed biological movement on action

    Curr. Biol.

    (2003)
  • F. Lacquaniti et al.

    The law relating the kinematic and figural aspects of drawing movements

    Acta Psychol. (Amst.)

    (1983)
  • R. Liepelt et al.

    When do we simulate non-human agents? Dissociating communicative and non-communicative actions

    Cognition

    (2010)
  • L.M. Oberman et al.

    EEG evidence for mirror neuron activity during the observation of human and robot actions: Towards an analysis of the human qualities of interactive robots

    Neurocomputing

    (2007)
  • L.M. Oberman et al.

    EEG evidence for mirror neuron dysfunction in autism spectrum disorders

    Cogn. Brain Res.

    (2005)
  • D. Perani et al.

    Different brain correlates for watching real and virtual hand actions

    Neuroimage

    (2001)
  • C. Press et al.

    Robotic movement elicits automatic imitation

    Brain Res. Cogn. Brain Res.

    (2005)
  • M. Romani et al.

    Motor facilitation of the human cortico-spinal system during observation of bio-mechanically impossible movements

    Neuroimage

    (2005)
  • R. Ramsey et al.

    Triangles have goals too: understanding action representation in left aIPS

    Neuropsychologia

    (2010)
  • R. Saxe

    Against simulation: the argument from error

    Trends Cogn. Sci.

    (2005)
  • S. Shimada

    Deactivation in the sensorimotor area during observation of a human agent performing robotic actions

    Brain Cogn.

    (2010)
  • J. Stanley et al.

    How instructions modify perception: an fMRI study investigating brain areas involved in attributing human agency

    NeuroImage

    (2010)
  • Y.F. Tai et al.

    The human premotor cortex is ‘mirror’ only for biological actions

    Curr. Biol.

    (2004)
  • K. Biermann-Ruben et al.

    Right hemisphere contributions to imitation tasks

    Eur. J. Neurosci.

    (2008)
  • G. Bird et al.

    Automatic imitation of human and robot actions in autism spectrum disorders

    Proc. Biol. Sci.

    (2007)
  • G. Buccino et al.

    Action observation activates premotor and parietal areas in a somatotopic manner: an fMRI study

    Eur. J. Neurosci.

    (2001)
  • B. Calvo-Merino et al.

    Action observation and acquired motor skills: an fMRI study with expert dancers

    Cereb. Cortex

    (2005)
  • M. Candidi et al.

    Virtual lesion of ventral premotor cortex impairs visual perception of biomechanically possible but not impossible actions

    Soc. Neurosci.

    (2008)
  • A. Casile et al.

    Neuronal encoding of human kinematic invariants during action observation

    Cereb. Cortex

    (2010)
  • P. Cisek et al.

    Neural correlates of mental rehearsal in dorsal premotor cortex

    Nature

    (2004)
  • R. Cook et al.

    Acquisition of automatic imitation is sensitive to sensorimotor contingency

    J. Exp. Psychol. Hum. Percept. Perform.

    (2010)
  • T. Chaminade et al.

    Motor interference between humans and humanoid robots: effect of biological and artificial motion

  • T. Chaminade et al.

    Artificial agents in social cognitive sciences

    Interact. Stud.

    (2006)
  • T. Chaminade et al.

    Mutual benefits of using humanoid robots in social neuroscience

  • Cited by (87)

    • Animacy and the prediction of behaviour

      2022, Neuroscience and Biobehavioral Reviews
      Citation Excerpt :

      A part of this network responds both during the observation and the production of a particular action and is called the ‘mirror neuron system’ (Cook et al., 2014; Kemmerer, 2021; Rizzolatti and Craighero, 2004). Perhaps a result of associative learning between observed and executed actions (Cook et al., 2014; Press, 2011), the system is sensitive to different aspects of biological actions (Kemmerer, 2021) ranging from simple characteristics such as adherence to the “two-thirds power law” (Dayan et al., 2007; Casile et al., 2010) to complex kinematic cues indicative of goals and intentions (Koul et al., 2018; Savaki et al., 2022). Activation of the action observation network and /or the mirror neuron system may thus trigger recognition of the moving object as an agent capable of independent action.

    • Human but not robotic gaze facilitates action prediction

      2022, iScience
      Citation Excerpt :

      During the past two decades, research examining the cognitive and psychological principles facilitating human–robot interaction for recreational (Palinko et al., 2016), assistive (Melkas et al., 2020), therapeutical (Langer et al., 2019), and educational purposes (Senft et al., 2019) has rapidly increased. Designing autonomous agents whose form and motion are modeled after humans is thought to facilitate the tendency to attribute human qualities to these agents (Fink, 2012; Press, 2011). Furthermore, modeling robots’ behaviors after human social behavior is thought to increase human acceptance.

    • A direct test of the similarity assumption — Focusing on differences as compared with similarities decreases automatic imitation

      2021, Cognition
      Citation Excerpt :

      These theories assume that perception-action links are learned responses that develop as a result of self-observation. Based on this idea, observed stimuli should activate corresponding motor plans to the extent that they are perceptually similar to the stimuli with which observers have learned sensorimotor associations (Press, 2011). In other words, perception-action links and imitative tendencies hinge on the similarity between actor and observer.

    View all citing articles on Scopus
    View full text