A number of websites recently reported on a breakthrough in which a robot was alleged to have successfully passed the mirror test. In turn, many of these news sites declared that the machine had achieved "self-awareness," proclaiming it to be an important step forward in the development of advanced robotics.
Unfortunately, this robot isn't actually "aware" of anything — nor did it pass anything even closely approximating a true mirror test. Here's what actually happened.
The study, which was conducted by Justin Hart and Brian Scassellati at Yale University, involved a robot named "Nico" that was successfully able to identify the location of its arm in space by referring to its reflection in the mirror. This task required some fairly sophisticated spatial recognition software, and it did indeed mark the first time that a robot was able to reference the location of a part in three-dimensional space by using a reflection.
Okay, so that's the experiment — nothing more and nothing less. Yet that didn't stop much of the media from presenting it as being something much more than that.
New Scientist's headline declared that the robot had learned to identify itself in the mirror, and that it was a "unique experiment to see whether a robot can tackle a classic test of self-awareness called the mirror test." Likewise, the BBC fell into the same trap, noting that "such self-awareness would represent a step towards the ultimate goal of thinking robots."
Worst of all was KurzweilAI's coverage, which went so far as to report that Nico had in fact achieved self-awareness, and that we could now officially discard the myth that only humans have that capacity (never mind, of course, that many nonhuman animals are self-aware).
What all these accounts fail to recognize was that Nico did not pass the mirror test, nor did it truly identify "itself." Strictly speaking, Nico was only able to determine the location of its extended appendage based on incoming visual data — and very poorly at that; Nico was only able to pin down the location of its arm within a two-centimeter margin of error.
In terms of true self-recognition, humans, like great apes, dolphins, and elephants, are self-aware in that they're capable of introspection — an internal mental state — which is accompanied by the ability to distinguish oneself as an individual, separate from the environment and other individuals.
Nico, on the other hand, has none of these attributes. And the fact that it has a pair of googly eyes and an extruded tongue doesn't change any of that.
And what's worse is that, through this misinterpretation, much of the media is both undermining and understating the complexity required for true self-awareness. It's an extremely higher-order capacity that has only been observed in a handful of species. Neuroscientifically speaking, the capacity for self-awareness, and the ability to cognitively identify oneself based on a representation in an external medium, involves a number of brain areas, including the brainstem, thalamus, and posteromedial cortices — not to mention the other areas of the brain required for conscious function.
Until software developers and roboticists like Hart and Scassellati start to develop correlates to those kinds of functions, it's safe to remain skeptical of any claim that a robot has become "self-aware."