Even the most abstruse of mental gymnastics are (at some level, in some form) reified in eqully pretzelian off-line motor cortex contortions. Naturally this sheds some light on the current stuck-in-the-mud state of general AI. For us to perceive an agent as “intelligent” it must foremost be “intelligible” (to us humans in particular). That means its perceptive world must be pretty close to our own. Embodied approaches to AI development acknowledge that human intelligence is chthonic; derived from our terrestrial heritage. We think the way we do because of an interplay tween the realities of living on the surface of the earth (ie, hearing thx to atmo, visible light thx to sol spectrum, trichromatic thx to colorful fruits, etx) and our haphazard evolutionary path. If an AI is to seem intelligent to us, it must be versed in our environs. To aliens feats routinely performed by computers and well beyond the ken of h. sapiens may be more deserving of awe and accolade than the programmers themselves. It is a mistake to assume that the programmers must, a priori, be more “intelligent” than the programs.
Consider: evolution gave rise to humans. The “blind” algorithmic processes of evolution have created (what H. sapiens likes to enshrine as) a wonderfully intelligent being. A “very dumb” system relying solely on chemical/biological interactions has produced “intelligence”. Thus we see the “creator” need not be (even remotely) as intelligent as the creation.
