We’ll know that robots are self aware when they tell us they are.

There is no coming to consciousness without pain - Carl Jung

What is self awareness? That’s always been a fairly ambiguous question; there’s no hard limit to sentience. The ability to distinguish yourself from a group? Having a sense of your physical body? Consciousness? Something else?

Consciousness is part of being self aware, certainly. And Max Tegmark, a theoretical physicist at MIT, thinks consciousness may be another state of matter, like a liquid or a solid. And he’s got other people agreeing with him, and the math to prove it. His work is based on proposals by neuroscientist Giulio Tononi, who in 2008 said that a system displaying consciousness must be able to store and process large amounts of information, and that this information must be integrated into a whole that can’t be divided into parts.

Like humans, robots already have the first part down. In fact, for a robot, conscious and unconscious states are clearly defined. The second part is trickier - it deals with self awareness, or seeing yourself as an individual. Even though humans are rationally aware of the different, independent biological mechanisms that make our existence possible, we still see our ‘selves’ as unique and special snowflakes, with dreams, thoughts, and abilities all our own.

I’m interested in consciousness and self awareness for a number of reasons, and seeing hard science like this makes me think we’re getting closer to robotic self awareness, or as I like to think of it, the machine awakening.

One of the reasons I've always been fascinated by the lack of proprioception in cephalopods is because it demonstrates that self awareness is varied based on evolutionary need, and can be vastly different among organisms.

We interpret self awareness as something anthropomorphic—“I think, therefore I am.” But an octopus doesn't necessarily have the same type of self awareness as we do - it experiences a very different body, relationship with other beings, and environment than we do. By the same token, we really have no idea what a robot’s sense of self awareness might entail. But we love to imagine that it will mimic our own.

As an evolutionary advantage, it makes sense for an organism to have a perception of themselves as an individual. In animals that must work both independently and as part of a social group, an awareness of a self is critical. In the case of the cephalopod, this awareness is limited to the visible parts of oneself, may be distributed in places other than the brain, and may not include key differences between itself and objects around it. Self awareness, therefor, does not require full knowledge of one's body or a complete understanding of what is 'other than self'. It can be argued that self awareness boils down to the ability to distinguish one’s own mind. 

A robot’s self awareness will likely evolve based on survival needs; at a point where it must develop an ability to differentiate between functional requirements (must recharge battery; falling down stairs could cause personal damage) and the needs of those around it (I have been told to sit; I am too heavy to sit on the baby; I have been asked to stop playing Taylor Swift songs). Some robot survival needs are obvious, but we don't really know what the others will be; we are not robots.

The Mirror Test or the MSR (Mirror Self Recognition) test, was designed to identify the age at which human and animals become self-aware. If the subject recognizes their own reflection and responds to seeing something they would not be able to see without a mirror, (usually a red dot painted on their forehead), they identify themselves as a unique individual.

Here's Morgan Freeman talking about how elephants respond to The Mirror Test. Interestingly, not all elephants pass the mirror test, suggesting that the species is at a point where self awareness may be emergent, or some kind of mutation.

Now consider getting robots to do this test. Here’s a robot named QBO seeing itself in a mirror.

The problem is that many robots have a far higher spatial memorization ability than humans and animals, so the comparison is not entirely accurate. A robot named Nico is believed to be one step closer to self awareness because he can recognize his own arm in a mirror. But that doesn't mean he wouldn't confuse another, identical robot's arm as his own.

In both examples, the robot is trained to look for the shape of itself, and learns to recognize itself based on stored information. But it doesn't know itself as separate from other things, such as a copy or photograph of itself. That would require self awareness. And it may not be something based on visual identification. It could come down to esoteric differences in programming or unique RFID identification. But if we can exactly clone a robot that is said to be self aware, does that negate the awareness? We don't know.

Self awareness is an emergent quality. We can provide an environment that will make it more likely to emerge (for example, by creating robot groups that must work socially to accomplish tasks, and therefore need to distinguish themselves from their copies.) But we can't predict exactly how a robot's sense of self will manifest, any more than we can know for sure what an elephant is really thinking when it sees it's teeth in the mirror for the first time.

We can only measure this stuff by human standards, and we don't even understand our own sense of self awareness very well. Robots are likely to become self-aware without our direct participation, despite our best efforts to model their consciousness after what we experience as humans.

Consider some of the Hollywood themes around the topic of robot self-awareness and you’ll find that as humans we both love and fear the moment when robots start thinking ‘for themselves’. If Tegmark and Tononi are correct, self awareness will emerge as an element of survival regardless of our hopes and fears.

We will know that robots are self aware when they act independently, in the interest of their own survival, regardless of the will of their creators. For all we know, the machine awakening could already be underway.