Can robots make good models of biological behaviour?

Area of Science:

• * Neuroethology and computational neuroscience.
• * Robotics and artificial intelligence.
• * Philosophy of science and scientific modeling.

Background:

• * Traditional approaches to modeling biological behavior are diverse and sometimes lack clear justification.
• * A new approach involves using physical robot models to study biological sensorimotor systems in neuroethology.
• * Existing methods for comparing scientific models require a structured framework.

Purpose of the Study:

• * To establish a framework for describing and comparing simulation models in behavioral and biological sciences.
• * To justify the use of physical robot models in neuroethology.
• * To define key dimensions for evaluating such models.

Main Methods:

• * Distinguishing simulation models from other forms of "modeling."
• * Defining seven dimensions for model comparison: relevance, level, generality, abstraction, structural accuracy, performance match, and medium.
• * Analyzing the strengths and weaknesses of different positions within this model space.

Main Results:

• * A framework is presented for evaluating simulation models based on seven dimensions.
• * Arguments are made for biologically relevant robot models over loosely inspired ones.
• * The benefits of physical implementation (medium) and iterative refinement for accuracy are highlighted.

Conclusions:

• * No single model position is universally correct; methodology must be explicit.
• * Context-specific modeling, considering real animal problems in real environments, is crucial.
• * Physical robot models offer significant advantages for studying biological behavior in context.