Achieving AI: Part 1 - The thinking machine

This is an introduction to a series of entries on the topic of Artificial Intelligence. The approach, for better or worse, is more philosophical than academic and is simply a series of vague snapshots of my thoughts on the subject that I hope to express in a coherent manner. Part One begins here...

Ask any John or Jane with a passing interest about how they envisage human-computer interaction in the distant future and many will paint a picture of something akin to the droids in Star Wars or the ship computer in Star Trek: Machines possessing a reasonable understanding of human dynamics even if they themselves struggle to express them (Though I've always viewed that agreed upon hypothetical as impractical. Machines have proven themselves to be very good at expressing what they understand).

By and large we are conditioned into this view by science fiction writers and the like. However one would think that even without such direct influence, we would still hold an expectation of machines being able to understand and relate to us in a similar fashion to humans. After all, sci-fi writers had to get their inspiration from somewhere.

The Holy Grail of AI, if it is to ever be achieved, will be a melding of the mechanical logic of an electronic device coupled with the emotional intelligence of a human being. The question then becomes; what is it that makes us human? What aspect of sentience sets us apart from algorithmically controlled machines and can they be mutually inclusive?

A little self analysis

People often say that what separates them from machines is the ability to make a decision regardless of their current state. Selecting Left instead of Right without consideration to any preconditions and being, in essence, in full control of the decision. However, is any decision we make truly uninhibited? Isn't there, at some level, a point of reference that we must use, even if only as an antithesis, to go Left instead of Right? Aren't our actions just as algorithmically controlled as machines?

The benefit we have in analyzing the 'thought process' of machines is that we are the ones that created it. We have a full picture of all inputs and processes that produce any observed output and can mathematically prove what any variation in the provided input would subsequently produce. We can, as a result, proclaim that machines cannot arrive at any conclusion without following the prescribed steps or some algorithmically controlled mutation of those steps.

Contrast this with our attempts to analyze our own decision making prowess. For the most part we waddle in ignorance. This ignorance has led us to believe that our decisions are, if we so choose, capable of being undefined right up to the point that they are made. But is this really so?

Granted, there may never come a day when when a human or machine will be able to mathematically deduce what your next thought or action will be, however this does not negate the possibility that these thoughts and actions are rule based. Our experiences (memory), immediate environment and culture play a very strong role in enforcing the rules which govern us. The problem is that in many instances these guiding hands are invisible; leading us to think that we and those around us are autonomous. This is a fallacy that we must acknowledge if we are to begin understanding our decision making ability in our efforts to recreate them. What you know, consciously or subconsciously, define your decisions. Any attempt to break form simply creates an inverse of those rules which then continue to define you. Any decision we make, no matter how obscure and random, can be traced back to some base idiosyncrasy. The only thing we lack is the time and insight to fully analyze it.

The discussion could go off into many areas, however I'm simply trying to draw a parallel between how we as sentient beings and machines with simplified logic processing utilize predefined rules to come to decisions.

Once we begin to appreciate this, the problem then becomes the further development and refinement of a machine's decision making process. What are the factors that make our processing so much more human?

If we can do it, why can't they?

It's one of the pillars of Artificial Intelligence and a major step into having human like machines. A sufficiently advanced artificial intelligence, to transparently interact with humans, must know our nuances. But how do we achieve this? What is required to truly understand human emotion. How does a machine discern that a statement that begins with "a paster in a church" is an axiom but "a paster in a bar" is the beginning of a punchline? This may not be as difficult as one may initially believe once you realize that it comes down to the very same things that will help it make any other decision; experience, environment and culture.

As far as experience goes, it's easy enough. The one thing a machine has no problem with is memory, and as resources such as the Internet continue to grow and the semantic web continues to develop, soon machines will not only have to rely on its own prior experiences, but on anyone that has chosen to contribute to this vast pool of knowledge. The line of separation when it comes to memory is that humans utilize our memory in a very inefficient manner. With the exception of a few savants we are incapable of perfect memory and many times unable to recall needed information in a given time. In fact, out ability to recall information is severely impacted by our ability to understand it. Think of a sentence spoken to you in your mother tongue, followed by one of equal length in an unknown foreign language. The known sentence can be recalled much easier because you're able to ascertain the meaning behind the statement, while the other can only be recalled via phonetics; a noticeably more difficult task. This may be one of our humanizing factors. Should we try and create a machine which duplicates this imperfection? Do we even want to?

The environment aspect is simply an issue of context. Different things take on different meanings depending on where you are and what you're doing. As experiences grow, so does the ability to decipher what is the appropriate interpretation of received input. The ability to identify and process a myriad or environmental factors is key. Sensors need to be broad enough to gather big picture information while acute enough to pick up contextual discrepancies.

Culture is a lot more complex than the previous two requirements. So complex in fact it may not be possible for humans to implement it. The idea of culture touches on a very interesting topic as far as machine intelligence goes. For a machine culture to exist, there must be a machine society. This need not be a physical manifestation as it is for human communities, but simply a medium for multiple artificial entities to be aware of and interact with each other. This culture will need to evolve over time as more entities become aware and begin to share experiences and knowledge of their environment.

The sharing of information, depending on the particular manifestation of the exchange medium, may lead to a very interesting and needed evolutionary step for true AI: assimilation. If machines, with the exception of their physical housing, are to be seen as information patterns which include their rules for input processing, then the sharing of information between entities quickly becomes the merger of entities. In essence, a new entity is created that represents a coming together of two or more different information patterns. Redundant processing instructions would be excluded based on which serves best in the current environment; this knowledge drawn from its assimilated memory. It doesn't take much imagination to see where such a development could potentially and, if science fiction writers are to be believed, eventually lead.

From understanding our own humanity, we can then begin the process of recreating it. From recreating it we may stumble on ways to improve it; creating something that not only understands and mimics us, but surpasses us. Moving past human evolution and instead creating a Human Revolution. The thinking machine may indeed be our next step.