We’ve been talking a little bit about the degree to which AI will have to be embodied if it’s ever going to achieve something that resembles consciousness. The theoretical question can probably be phrased as “to what degree does consciousness exhaust behavior,” but the practical questions are about to what extent an AI has to interact with the environment the way we do to get the tools to simulate consciousness.
Now the gird above is a basic implementation of White’s illusion. When humans look at the squares where the gray lines meet, the perceived brightness seems lighter than the rest of the line. Of course the lines are solid – the illusion is an artifact of trying to learn to differentiate between light and dark. Turns out, digital “brains in vats” are vulnerable to this very very same illusion:
Now for the sting: the brain in question was an artificial neural network (ANN) that only ever existed inside a computer. It was trained to successfully perform on a lightness constancy task. Most excitingly, when trained to discern between overlapping layers, the ANN sees White’s illusion (Box E). White’s illusion has been problematic to model as the lightness perception goes “the other way” from the stimuli shown here. Thus, the by-product of learning to see lightness and depth is a susceptibility to these illusions. This also tells us something about how animal brains, including our own, work.
If AI’s can simulate the results of walking around and interacting in the world without ever having to walk around and interact in the world, then that’s one less hurdle toward building something that most people most of the time would agree is conscious (via Cognitive Daily)
* A Peircean Checklist For Conscious Artificial Intelligence
* Seeing things similarly [Auntie Em]