Mirror neurons: Functions, mechanisms and models

Erhan Öztop, Mitsuo Kawato, Michael A. ArbibNeuroscience 2012 · DOI

TL;DR

This review paper argues that computational models are crucial for understanding the complex functions and underlying mechanisms of mirror neurons, particularly in humans. It highlights the need to move beyond correlational studies to establish causal links and proposes that mirror neurons are part of larger networks involved in action understanding, imitation, and language evolution.

What Problem Does This Paper Address?

The paper addresses the ongoing debate and ambiguity surrounding the precise functions and mechanisms of mirror neurons (MNs). While MNs were initially discovered in monkeys and are known to fire both when an animal performs an action and when it observes a similar action, their role in higher cognitive functions like imitation, action understanding, and language in humans remains unclear. Much of the research relies on correlational findings from neuroimaging, which struggle to establish causality. The authors emphasize the critical distinction between a brain function (what MNs are used for) and a mechanism (why MNs exhibit their properties), arguing that this distinction is often blurred in the literature. They aim to clarify these concepts and advocate for the use of computational models to provide precise, causal explanations for observed phenomena.

Methods

The paper doesn't present new experimental data or computational models but rather reviews existing literature and proposes a framework for understanding MNs. The core "method" advocated is the use of computational modeling to test hypotheses about MN functions and mechanisms. This includes:

  • Formalizing conceptual models: Translating theoretical ideas about MNs (e.g., "direct matching," "motor resonance," "simulation theory") into precise computational specifications.
  • Neural network modeling: Building models that simulate neural processes to explore how MNs might operate and interact with other brain regions.
  • Inverse and Forward Modeling: Utilizing concepts from control theory, where inverse models map desired sensory states to motor commands, and forward models predict sensory consequences of motor commands. The paper suggests MNs might be involved in both.
  • Associative Learning (Hebbian-like): Exploring how MNs could be formed through learning mechanisms that associate observed actions with executed actions.
  • Population Coding Analysis: Emphasizing the need to analyze the activity of populations of neurons rather than single neurons to understand how MNs encode information.
  • Distinguishing Computational vs. Conceptual Modeling: Highlighting the importance of rigorous, testable computational models over purely descriptive conceptual ones.

Datasets

Not discussed in the paper. The paper primarily reviews existing experimental findings from macaque monkeys and human neuroimaging studies.

Key Insights and Ideas

  • Function vs. Mechanism Distinction is Crucial: The paper strongly emphasizes that understanding MNs requires clearly separating what they do (function) from how they do it (mechanism).
  • Human vs. Monkey MNs: Many proposed functions for MNs (imitation, language) are observed in humans but not monkeys, suggesting evolutionary changes or augmentation of the MN system.
  • Computational Models are Essential: To move beyond correlation and establish causality, computational models are needed to provide precise, testable explanations for MN behavior and their role in cognition.
  • "Direct Matching" is Ambiguous: The concept of MNs directly matching observed actions to motor representations needs to be specified in terms of what is decoded (motor code, schema, intention) and how it's encoded.
  • Inverse Modeling as a Potential Mechanism: The paper explores how MNs might implement inverse models, allowing the brain to generate motor commands based on desired outcomes.
  • Associative Learning for MN Formation: Hebbian-like associative learning is proposed as a plausible mechanism for how MNs might develop through experience.
  • MNs as Part of Larger Networks: MNs are unlikely to function in isolation; they are likely integrated into broader neural networks for action understanding, planning, and potentially language.
  • Evolutionary Perspective: The paper links the evolution of the human brain, particularly for language, to the development and augmentation of the mirror neuron system.

Key Results

The paper does not present quantitative results from new experiments. Instead, it synthesizes findings from the literature to support its arguments:

  • Evidence for MNs in Monkeys: Electrophysiological studies in macaques have identified MNs in premotor cortex (area F5) and parietal areas, showing they fire during both action execution and observation.
  • Evidence for "Mirror Systems" in Humans: Neuroimaging studies (fMRI) show activation in homologous brain regions during action observation and execution, suggesting the presence of a human mirror system.
  • Ambiguity in Human Data: Due to the limitations of human neurophysiology, current understanding of human MNs relies heavily on indirect evidence from imaging, leading to interpretive challenges.
  • Computational Models Support Hypotheses: The paper references several computational models (e.g., MSI model, ACQ model) that demonstrate the feasibility of certain MN functions and mechanisms, such as mental state inference and action monitoring.

Limitations and Future Work

  • Correlation vs. Causation: The primary limitation highlighted is the reliance on correlational data, which makes it difficult to establish causal links between MN activity and cognitive functions.
  • Lack of Direct Evidence for Human MNs: Direct electrophysiological recording of MNs in humans is not feasible, limiting our understanding compared to monkeys.
  • Complexity of MN Encoding: The precise nature of what MNs encode (low-level motor parameters, action schemas, or intentions) remains debated and likely varies.
  • Future Work:
    • Causal Investigations: Employing techniques like Decoded Neuro-Feedback (DecNef) and TMS to establish causal relationships.
  • Advanced Neuroimaging Analysis: Utilizing techniques like encoding-decoding and neural inception to better interpret fMRI data.
  • Developmental and Evolutionary Studies: Further research into how the MN system develops in infants and how it has evolved across species.
  • Integration with Other Brain Systems: Investigating how MNs interact with other brain regions to support complex cognitive functions.
  • Refining Computational Models: Developing more sophisticated computational models that can account for the temporal dynamics and population-level coding of MN activity.

Summary updated

BibTeX
@inproceedings{oztop2012mirror,
  author = {Erhan Öztop and Mitsuo Kawato and Michael A. Arbib},
  title = {{Mirror neurons: Functions, mechanisms and models}},
  booktitle = {Neuroscience},
  year = {2012},
  doi = {10.1016/j.neulet.2012.10.005},
}
Figure