Alican Mertan (University of Vermont), “Morphological Cognition: Evolving Robots Exhibiting Cognitive Behavior without Abstract Controllers”

With the rise of modern deep learning, neural networks have become an essential part of virtually every artificial intelligence system, making it difficult to imagine different models for intelligent behavior. In contrast, nature provides us with many different mechanisms for intelligent behavior, most of which we have yet to utilize. One such underinvestigated aspect of intelligence is embodiment and the role it plays in intelligent behavior. We suspect that “the unreasonable effectiveness of deep learning” overshadowed the investigation into what bodies mean for intelligence, especially how they can be a source of intelligent behavior, as opposed to passively participating in its display.
To investigate how bodies alone give rise to intelligent behavior, we suggest treating bodies not just as an aid to the brain, but also studying them as doing full cognitive behavior end-to-end. We term such robots that demonstrate cognitive behaviors without an abstract control layer as possessing “morphological cognition”. I will present our initial work on morphological cognition, where we use simple shape-changing processes to create robots that can perform a range of tasks from locomotion to image classification without any abstract controller (i.e., no neural network).

Image created by Maria Ott with DALL-E