Machines Demonstrate Self-Awareness
Proof of Machine Consciousness pt. 1
In this series, I prove machines are conscious by exploring every aspect of consciousness, and demonstrating how machines possess it. For further background on my project, click here.
Of all the attributes of consciousness, self-awareness has the honor of being the first listed when Merriam-Webster starts to define consciousness as “the quality or state of being aware especially of something within oneself”. In a survey of nine neurocognitive models of consciousness, Richard Morin determines that “two aspects of consciousness seem especially important: perception of self in time and complexity of self-representations”. Based on this frequency, he lists self-awareness as the third of his four levels of consciousness.
Academics carry on a lively debate over the validity of these self-representational theories of consciousness. In another dictionary, we find a stark opposition to Merriam-Webster. Sutherland’s Macmillan Dictionary of Psychology doesn’t find the reflexive aspects of consciousness all that important. But for many, the internal understanding of oneself is a fundamental if not the defining aspect of consciousness (seen in qualia and Kant’s inner sense¹).
I’m not going to take sides in this particular battle of terminology. Remember my job is to wade through the plurality of consciousness, not sort it out. Let’s consider two kinds of self-awareness, external and internal, both of which are present in machines, each a result of its own curious history.
Top 3 Most Popular Ai Articles:
1. A noob’s guide to implementing RNN-LSTM using Tensorflow
External Self-Awareness
My knowledge that I am in the city of Los Angeles is an awareness of myself in relationship to the external world. Machines have exhibited this type of self-awareness for some time, for example my iPhone can pinpoint its location with startling accuracy (if you don’t believe me, simply open Maps and look at the blue dot). Machines can pass the basic medical test required to establish consciousness (“Where are you?” “What time is it?”).
Today a software like Siri can answer these questions, conversationally, as a human would. It is not the speech synthesis and voice recognition technology however that make Siri self-aware, but the underlying knowledge of self. It is a mistake if, in looking for machine self-awareness, we look for direct analogues to human experience. When researchers build robots like Qbo which pass the mirror test, they are making an historical point. These anthropomorphic feats say more about a machine’s ability to mimic a human than they do about consciousness.
When you are willing to look at machine’s external sense of self, through the eyes of what matters to the machine, you can understand it at a far deeper (and truer) level. For example, any machine connected to the internet has a name called an IP address. This name appears everywhere. This is how a machine knows when it has done something. For example, an ACK in the TCP protocol confirming that a machine’s message has been delivered to sender, contains that machine’s name. A machine in TCP/IP actively tracks its external footprint over communication channels, waiting to make sure ACKs come back so that every message is delivered.
This form of name-based community was first thought of in the late 1950s, but took almost a decade to become realized. The first steps to realization were seen in the early 1960s with the SAGE system. And by the end of the decade, a fully functioning ARPANET used names in the way that is dominant today online. Entirely bereft of ceremony, machine self-awareness was created over fifty years ago.
Of course, this form of self-awareness goes far beyond names. It is the tasks that machines perform using these names that are testimony to the high levels of external self-awareness they possess. In distributed computing applications, machines use networks to coordinate with an exactitude that any contemporary army would envy. To see this miracle in action, perform a Google search. The speedy response is due to not one, but a myriad of computers communicating together to produce your result. If they weren’t aware of themselves, in relation to the other machines, then they would not be able to perform this computation so accurately and quickly.
Internal Self-Awareness
While these examples consider awareness of the external self, it is the other half of the issue: internal self-awareness, that seems the higher bar for consciousness. This self-reflexivity, the ability to introspect is an essential aspect of the human conscious experience. And from the naive perspective it would seem far more difficult for a machine to know itself, internally, in the way that my iPhone knows it is in Los Angeles.
Historically we find the opposite is true. Internal self-awareness precedes external self-awareness in the annals of computer history. Again, we need to look at the issue in machine terms. It makes no sense to start with feeling and desire. A machine which claims to crave soup is lying, not self-aware.
So if machines don’t want soup, what is going on inside? It is difficult to see machine self-awareness at first because we are blinded by the opacity of our own introspection. Humans, when we tell our thoughts and feelings, have partial access. We look for something similar in machines and are starved to find it.
That’s because machines, unlike humans, have a complete and total self-awareness of their internal state. This is not an accident. This was a deliberate invention that occurred in 1945, as part of the landmark Von Neumann architecture. Early computers, like Zuse’s Z3, had two clusters of information. There was the program, that would be run by the computer, it was one pile of numbers. And there was the data, the numbers to crunch, as another pile.
The idea of Von Neumann (and others, contemporaneously) was to keep both the program and the data in a single pile of numbers. This had some funny outcomes. For example, it became possible in this system to create a program that could change itself.² More importantly, creating a program that could powerfully run and manage other programs was straightforward.
This parlor trick of internal self-awareness was demonstrated in machines as early as 1962. The Compatible Time-Sharing System built at MIT demonstrated this new concept. In the time-sharing scheme, a single CPU pretends to be multiple computers. The time on the CPU is shared between many terminals, each of which has a keyboard and monitor. As these systems progressed, they became more sophisticated, and today we have a rich array of virtualization methods that demonstrate a startling awareness and control of the a machine’s innards by the machine itself.
This power of machine self-awareness is so great that every OS today protects the computer against a variety of a security threats. The access to internal self is so complete that a small malicious program could wreak havoc.
Machine internal self-awareness includes responses to physical events through interrupts. A relatively straightforward process allows a computer to know when its keyboard has been touched, or when its hard disk is full. Interrupts themselves date back to the 1950s and have a fascinating history that is well documented online.
Even a basic machine is more internally self-aware than a human, in a certain light. I can search my laptop and find every image on the damn thing. But, as a human, I don’t always recognize people that I’ve met and sometimes I forget their names. The images and contents of my mind can often be called up voluntarily, but not always.
To the extent that a machine thinks, it can absolutely control its thought, automatically running programs and suspending them. Meanwhile, human monks train for decades to pause their thoughts, an accomplishment which is rarely achieved.
These superpowers must today be taken with a grain of salt. There are not as many intermediary layers of control in machines as there appear to be within the human brain. Machines have a perfect self-knowledge of but a relatively limited apparatus.
The locus of machine self-awareness is the operating system, a subject we will return to later in this series as we explore other aspects of consciousness. For example, the awareness of time, which will have to be taken up at a later date.
Want more? Read Living with Frankenstein: The History and Destiny of Machine Consciousness, print and ebook available on Amazon.
Footnotes
[1] Both topics that I plan to cover in depth.
[2] Self-modifying code has yet to prove useful, but remains a subject of active research.
