Artificial Intelligence (AI) – Grey cells, anyone?

The Evolution of Intelligence

Intelligence, as we understand it, refers to the ability to comprehend a given set of events, to analyze the results thereof, and to use the knowledge thus gathered to interpret, predict or conceptualize other similar or dissimilar occurrences. Sounds simple, yes? Trust this humble author, there is a lot more at work here than mere comprehension and analysis. After all, even supposedly unintelligent beings such as plants can comprehend and analyze, if we go by Acharya Jagdish Chandra Bose’s experiment which demonstrated that plants react in different fashions to different people, depending on their previous experiences with the person. But can we term them as intelligent? You’ll be surprised, but yes, even they are intelligent, though in their own botanical fashion. For, the evolution of intelligence does not depend on the size or content of the brain but on necessity. Take for instance your pet dog. As Anton Pavlov had demonstrated, if a bell is rung regularly before your pet’s lunchtime, a day will come when your pet would begin to salivate on the mere sound of the bell, regardless of the fact that there is no food on its plate. This is a simple form of association – the dog associates food with the sound of the bell. Since the evolution of the dog did not move in the same direction as of us humans, its (and most of the animal kingdom’s, for that matter) intelligence is reflected towards meeting the more basic necessities – that of food and shelter.

So where did we humans go wrong (or right!)? Surely we have evolved in a totally different fashion, food and shelter being only two of our several basic necessities? Well, the trick lies in the word necessity. While weeds propagate in leaps and bounds and monkeys can leap while cheetahs can bound, us humans have no such inherent self-preservation module. We therefore, had to rely on other, external devices to save us from extinction. And as our reliance on such tools and devices grew, our intelligence too began to manifest itself towards that goal. We learned to assimilate complex information and to process it logically to develop more and more sophisticated scientific wizardry.

The second most important thing in the evolution of human intelligence was language. You heard right, language. You see, we humans can speak, not merely communicate, but speak. With a rich vocabulary, we have words to express and explain almost anything under the sun. And this development of language leads to exchange of thoughts and ideas, which in turn lead to scientific thinking. For all other animals, communication has remained a collection of signals or a limited range of sounds which is definitely prohibitive so far as development of schools of thought is concerned.

From Intelligence to Artificial Intelligence

As man progressed up the ecological chain and gained a place of import, his thirst for knowledge was fuelled by his desire to create. You see, since eternity (read for the past five thousand years ) human intelligence has been skewed towards creation – from the first stone clubs to the modern day skyscrapers. Combined with the genetic need of living beings to create their own replicas, this desire naturally progressed towards replicating his own image, not through reproduction alone but also through scientific gimmickry. And with the rapid development of computing and robotics in the past few decades, man today suddenly finds himself on the doorstep of his long cherished dream – to assemble a non-human device that can understand, analyze and interpret information and use the same to predict or conceptualise similar or dissimilar occurrences. In other words, a device that is Artificially Intelligent.

Artificial Intelligence

So what is Artificial Intelligence? A device that can successfully imitate humans? Well, while the ultimate aim is to construct something of the kind, present AI research is directed towards creating fragmented forms of the ultimate prototype - machines that can make the day to day human tasks simpler, that can take on the role of guide, if not the ones of friend and philosopher. The primary aim as of now is to replace input devices like the keyboard and the mouse in computing machines by vision and speech. For example, Project Oxygen being conducted at the Massachusetts Institute of Technology aims at creating human centered computing through a combination of specific user and system technologies. The idea is to use a combination of embedded computational devices (devices that are embedded in the immediate environments, e.g., homes, cars, etc.) and handheld devices to create mobile intelligent spaces that the user has access to at any point of time from any point in space. Even today we see a lot of similar semi-intelligent devices – Remote Sensing Satellites, Televisions that adjust the brightness by themselves depending on the extent of light from other sources in the room – all tiny examples of AI.

The Hurdles

While we have come a long way since the days of the first auto-pilots, a lot remains to be done in the as yet science-fiction world of Androids (Robots that are almost identical to humans – remember Terminator-2?!!) and the like. One must remember that the animal brain is a highly complex network of transmitters, receivers and semiconductor pathways. Developing purely logic-based programs (as we see in the computers of today) can only create machines, and not intelligence. For example, the animal brain is capable of storing seemingly insignificant data which can be retrieved decades later if they come in handy. A simple computer, however, cannot store any data on its own; it reads and records only that information which it has been programmed to assimilate – no more and no less. Even for highly advanced chess playing computers like Deep Blue/ Deep Thought, the mechanism of "thought" follows the same pattern – these machines only retrieve information relating to the thousands of moves that have been incorporated into them and would, in all probability, stand helpless in front of a really novel approach to the game.

The second problem that needs to be overcome is that of inference. The machines of today are bounded by rational thinking – an out of the box way of thinking is not their forte. But for there to be intelligence, irrational thinking is quite frequently called upon. We can take of example of the discovery of the structure of benzene – August Kekule had a dream of snakes devouring themselves, tail first and thus forming rings; and that was how Kekule chanced upon the possibility of the benzene molecule being circular in structure. We can easily understand how his brain had stored the information relating to benzene (which was his immediate project) and juxtaposed the possible structure on a real life example which, however irrational, pointed Kekule in the right direction. However, similes of a similar nature cannot be expected from the computing devices of today.

Thirdly, there is the problem of Integration – to be successful, an AI must be able to integrate two or more dissimilar events. For example, through imaging and subsequent analysis, a computing device can understand, say, the normal gait of a John Doe. However, if John jumps up in amazement on receiving news of getting a fellowship, the computer would fail to understand the change in John’s gait and might decide that he is sick which would be an utterly incorrect assumption. The problem is mainly due to the fact that while in the case of humans, most assumptions and decisions are based on common sense gained through practical experience as also emotional trends, in the case of logical machines decisions are based on a set of "yes" "no" type protocols. Thus, logical intelligence leaves precious little space for ambiguity, while ambiguity perhaps is the most common occurrence in the animal universe. One way to override this problem (as proposed by John McCarthy, 1959) would be to create programs which enable the AI device to arrive at several possible conclusions by observing one single event. For example, when John jumps up on hearing about the fellowship, the AI should arrive at the following illustrative conclusions:

a) John is sick
b) John has lost his marbles
c) John is practicing for the high jump event in college
d) John is happy

It should then correlate these reasons with John’s facial expressions and other factors such as the tone of his voice, the event immediately preceding his jump, etc. Thus formalized, common sense can be induced into the AI device.

Passing Shot

So where do we stand now? Is Artificial Intelligence in the nature of Arnold Schwarzenegger in Terminator 2 a real and distinct possibility? In this author’s opinion, it is, though not in the near future. As discussed earlier, we are already seeing bits and pieces of AI around us; the problem that remains to be solved is how to integrate these pieces to make them into a whole that has similar capabilities as the human brain. And while the task is difficult, it is not insurmountable, what with new programming theories and proposals coming up everyday. Perhaps in another fifty years every middle class household would boast of possessing an R2D2 of the Star Trek fame, if not a Terminator.