Full Automation: Phenomenology, Embodied Cognition, and the Road to Full Artificial General Intelligence

This paper argues that artificial intelligence is on course to surpass human intelligence not only in the cognitive domain—through the combination of neural networks, large language models, attentional architectures, and data scaling according to power laws governing model parameters, compute, and training data—but in the embodied domain as well, through the development of large behavioral and visual learning models. I argue that there are no theoretical limitations impeding these embodied models from surpassing human sensorimotor intelligence in the same way that language models are poised to surpass human beings in abstract cognitive operations, though embodied intelligence follows a staggered developmental course in contrast to the comparatively smooth scaling of cognitive generative intelligence. Issues of consciousness are orthogonal to questions of intelligence: the structural architecture for embodied consciousness is distinct from that of embodied intelligence, and embodied intelligence can advance without it. Hubert Dreyfus—far from being a defeated opponent of AI—is the most precise available diagnostician of why the path to full embodied automation has the specific staged developmental structure it does. Dreyfus is vindicated descriptively while being refuted normatively: his phenomenological analysis of embodied temporal intelligence specifies the engineering challenges behavioral models must solve, but his conclusion that these challenges may constitute a permanent theoretical barrier is wrong. The essay unfolds in three parts. The first section outlines contemporary techniques for scaling neural networks, their particular success in cognitive labor, and how these techniques are being applied in robotics. The second shows how Dreyfus, while making correct arguments against symbolic AI and identifying structural facts about human embodiment, conflates the package of human conscious intelligence with embodied intelligence as such. The third, drawing from interoceptive views of consciousness, outlines how consciousness is distinct from the structures needed for embodied intelligence and why they need not be packaged together, establishing the separability of intelligence and consciousness and diagnosing the precise nature of Dreyfus’s error.