Imagine a baby just a few weeks into life, looking up at a mobile hanging above its crib. With tightly clenched fists, it moves its arms and legs in every direction with no control, no awareness that those limbs are even part of its own body.
No one tells the baby "those are your hands, use them this way." It couldn't understand the instruction if they did. So instead, something else happens: around 4 months of age, the baby starts a process called hand regard, through hand-eye coordination and physical self-exploration, that it is the one causing certain changes in the environment. It learns to separate "my signal" from "external agent." That's the discovery of the body.
It's a process so organic in humans that we never notice it happening; we simply wake up one day already in control.
Which raises the question this piece is really about:
Can an AI or something we might call an "AI Mind", discover its own physical body the same way? Through experimentation and observation alone, with no pre-programmed configuration for that specific hardware, or any hardware at all?
A note from me:
This is an adapted excerpt from my master's thesis in Computer Engineering (AI specialization). Full bibliographic references will be up once the thesis is complete. I'm still deep in it.
If embodiment, self-modeling, or the boundary between "controller" and "body" is something you think about too, I'd genuinely like to hear your take.
What's your intuition, Is body discovery a hardware problem or a learning problem?
Image from Pavlina Baudysova on Unplash
Top comments (0)