Embodied AI and cognitive robotics
Explanation
Embodied AI and cognitive robotics are research currents that emphasise the importance of body, environment and physical interaction in the emergence of genuine intelligence. Their philosophical roots lie in embodied cognition (Merleau-Ponty, Varela, Thompson), in the critique of classical symbolic AI, and in the rediscovery of the body as the foundation of cognition. Key pioneers include Rodney Brooks (MIT), Rolf Pfeifer, Luc Steels, Cynthia Breazeal.
Rodney Brooks articulated the founding critique in articles such as Elephants Don't Play Chess (1990) and Intelligence Without Representation (1991). He argued that classical AI, obsessed with abstract symbolic representations, was a dead end: insects with tiny brains navigate complex environments very effectively without symbolic representations; we humans actually use abstract representation only for a few tasks, most of our intelligence is bodily and situated.
Brooks proposed the subsumption architecture: simple layers of reactive behaviour (avoid obstacles, follow walls, explore, etc.) combined without a central symbolic controller. His insect-like robots (Allen, Herbert, Attila, Cog) demonstrated that robust and adaptive behaviour could emerge without complex symbolic representations. This marked a paradigm shift in robotics and helped rehabilitate corporeality in cognitive science.
Luc Steels has explored how language and concepts can emerge in populations of robots that interact with each other and with the world, without explicit programming of language (talking heads experiments). Rolf Pfeifer (How the Body Shapes the Way We Think, 2007, with Josh Bongard) has developed principles of morphological computation: the body itself (its form, elasticity, materials) performs computation, reducing the load on the computational controller.
Cognitive robotics seeks to produce robots with advanced cognitive capacities: multimodal perception, episodic representation, reasoning, language, emotion, social interaction. Projects such as Cog (MIT, 1993-2003), iCub (a child-sized humanoid robot, Europe, 2004-), NAO, Pepper (Softbank), Kismet (Breazeal's emotional robot) have explored various aspects. Recent years have seen enormous progress with deep-learning-based robotics (robots learning manipulation by self-supervision, videos, etc.).
For the theory of consciousness, embodied AI is relevant because if human consciousness is profoundly embodied, situated, sensorimotor (as Varela, Thompson, Gallagher, O'Regan, Noë defend), then AI systems without a body, without interaction with the world, will probably not achieve genuine consciousness. Embodied AI aims to produce systems whose representations are grounded in bodily experiences. Whether this is sufficient or not for consciousness is a debated matter. But it offers a particularly fruitful bridge between robotics, embodied neuroscience and philosophy of mind. Future robots with sophisticated bodies, rich sensors, continuous learning systems and self-modelling capacities could raise urgent ethical and philosophical questions: do they have experience? are they moral patients? what do we owe them? Embodied AI forces us to take these questions seriously.
Strengths
- Incisive critique of disembodied symbolic AI.
- Constructive agenda of successful bioinspired robots.
- Productive dialogue with phenomenology and enactivism.
- Conceptual foundation for modern embodied AI.
- Morphological computation as a genuinely new paradigm.
Main critiques
- Historical difficulty in scaling to complex abstract cognition.
- The 'no representation' slogan requires nuance today.
- Does not solve why sensorimotor coupling generates qualia.
- Integration with deliberative reasoning still incomplete.