Artificial Intelligence: The past, the present and the future
Humans have always striven for greater things. The evolution of man has seen inventions and discoveries, the rise of civilizations, massive feats of engineering and in the recent past exponential advances in technology. The inception of all this was in the human mind. So there is no surprise that for thousands of years, people have tried to understand how the human mind works. But it does not stop there. Attempts are now being made to not only understand, but also to build intelligent entities that replicate this intelligence. This has given rise to the field of Artificial Intelligence.
Although the phrase Artificial Intelligence seem modern and recent, the concept can be traced back to classical philosophy where attempts have been made to link the human mind to a symbolic system. Since then many different attempts and advances have been made in laying a slow but solid foundation that would eventually provide a strong base for the field of Artificial Intelligence. A very recent example is the work of the mathematician Alan Turing. Widely thought to be a man ahead of his time, his ideas proved to be of profound impact in the field of AI. Even at an age where computing was at its dawn, he was pursuing the question, “Can machines think?” With no clear visibility as to where computing was heading at this time, he argued that conjectures are of great importance to new lines of research. He then published a set of criteria that he believed was necessary to determine if a machine was genuinely intelligent. This is now known as the Turing Test which essentially states that a machine could be judged intelligent if it can fool a human examiner into thinking the machine is human. The Turing Test is still considered the holy grail of AI researches and remains a useful method to evaluate the progress of AI.
However, the field of AI was not formally founded until 1956, at a conference at Dartmouth College where the term Artificial Intelligence was first used. With lots of optimism it was stated at this conference that, “every aspect of learning or any other feature of intelligence can be so precisely described that a machine can be made to simulate it’ . This principle kick started an AI revolution that many believed would lead to fully fledged thinking machines by the turn of the millennium.
The early years of AI was highly successful. Lots of enthusiasm surrounded the community and even minor advancements by AI researches in early computing was seen as breakthroughs. The primitive nature of computers and the rapid advancements in technology only made them more frequent. To everyone in the intellectual community this was rapid progress which lead to bold (and on hindsight overconfident) statements such as this;
“It is not my aim to surprise or shock you but the simplest way I can summarize is to say that there are now in the world machines that think, that learn and that create. Moreover, their ability to do these things is going to increase rapidly until — in a visible future — the range of problems they can handle will be coextensive with the range to which the human mind has been applied.”
But the bubble would soon burst as the reality of the ever growing engineering requirements would add to the complexity of the problems to be solved. The earliest strategies of AI problem solving was exhaustive; which basically meant that possible sequences of solutions were tried out until a solution was found. This worked at that time due to the simplicity of the then existing problems. It was also wrongly assumed that scaling up a problem was only a matter of faster hardware and more memory. But with time as theorems and ‘fact’ failed to hold their own, the realization set in within the community that all ‘progress’ made was just the tip of the iceberg. True Artificial Intelligence was still beyond reach.
The first decade of AI research problem solving did not scale up with problem complexity as expected. Clearly, new ideas were required going forward. One approach was to use more powerful, domain-specific knowledge that allowed larger reasoning steps during problem solving. The DENDRAL program was an influential pioneer project in AI that followed this approach. The software program is considered the first expert system because it automated the decision-making process and problem solving behavior of organic chemists. The Dendral project lead to the derivation of many other systems that had widespread application in real-world problems. This lead to a higher demand for workable knowledge representation schemes and as a result a large number of different representation and reasoning languages were developed. AI soon became an industry. Commercial expert systems were helping companies save millions every year, competition was brewing between countries and within a decade, a few million dollar industry was transformed into a billion dollar industry.
Modern Artificial Intelligence has seen significant advancements in the past decade due to greater use of the scientific method in experimenting with and comparing approaches. Many parallels have been drawn and sub fields of AI have found common ground with other disciplines as well. Neural Networks and Genetic Algorithms are two such areas that has had important theoretical advancements recently. Artificial Neural Networks (ANN) attempt to simulate the conditions that exist within the human brain, which is without doubt the ultimate artificial intelligence machine. The concept behind the natural brain is the idea of connection strength between individual synapses and dendrites. In ANN, A similar principle is followed where artificial neurons used most are assigned a higher weight, indicating a stronger connection. However, no artificial neural network exists as of now that can match the complexity of the human brain. Genetic algorithms on the other hand, mimics the process of Darwin’s natural selection where computers/algorithms are allowed to evolve through multiple iterations, and the result is a population of fit or highly effective computers/algorithms that are chosen to be the best fit for a particular problem. Likewise, there are many other algorithms such as Bayesian networks, Support Vector Machines, and most importantly, Natural Language Processing (NLP). An NLP module is able to process fuzzy human language that would otherwise make no sense to a machine. It could analyze a paragraph consisting of spoken language, analyze it and produce required results, even predict the sentiment behind the language, something an ordinary machine is not able to do. The above mentioned techniques and algorithms, also brings cognitive psychology, neuroscience, evolution, sociology, etc. into the picture. What this means is the AI venture cannot be successful without a thorough grounding on the aforementioned areas because they dive into the study of the human mind on an individual and collective level.
So the question is, where are we currently in AI development? Or how far are we from a truly intelligent machine? To answer this we need to understand the three calibers of AI development that most experts agree on.
Artificial Narrow Intelligence (ANI): AI’s that have a specialization such as gaming, stock market prediction, healthcare, etc. They have no purpose beyond the highly specialized task for which they were created.
Artificial General Intelligence (AGI): An AI that passes the singularity of human intelligence and is able to think, reason, think on an abstract level (a feature that separates humans from chimps), solve complex problems, and even to learn by experience.
Artificial Super Intelligence (ASI): An AI smarter than anything humanity has ever known, even smarter than the collective intelligence of humanity. What we would otherwise call a god. Having mental capacity and power trillions of times that of a human.
Artificial Narrow Intelligence is basically a machine that surpasses human intelligence at a specific thing. The world currently has successfully conquered ANI. Artificial Narrow Intelligence is everywhere; from Google’s search, through self-driving cars, to chess agents able to beat the best human chess player. ANI systems are increasingly being used at commercial level by internet giants such as Google, Facebook and Youtube. The advancement through these gauntlets of AI research is increasing exponentially in speed and knowledge, simply because knowledge and our technological capacity is increasing exponentially according to Moore’s Law. This leads us to the million dollar question, does each new innovation in ANI pave the way albeit slowly towards AGI and ASI?
Man has walked on the moon. Science and engineering together have gone well beyond what was previously thought possible. All of this at its infancy, an idea within the human brain. Yet no brain on earth is even remotely close to replicating an entity that can mimic the functionality of the human brain. AGI is what everyone imagine AI to be. AGI is the Artificial Intelligence of science fiction. It is the idea of Artificial Intelligence that is as intelligent as the human brain. But ironically the difficulties lie in what is actually easy or second nature for us human beings.
“Nothing will make you appreciate human intelligence like learning about how unbelievably challenging it is to try to create a computer as smart as we are. Build a computer that can multiply two ten-digit numbers in a split second — incredibly easy. Build one that can look at a dog and answer whether it’s a dog or a cat — spectacularly difficult. Make AI that can beat any human in chess? Done. Make one that can read a paragraph from a six-year-old’s picture book and not just recognize the words but understand the meaning of them? Google is currently spending billions of dollars trying to do it.”
It was suggested that a successful AI is not one that tries to behave like just another human being. It is one that behaves like a child. Rather than being preprogrammed with instructions, the machine is allowed to evolve and learn through experience much a like a child does. The things that seem so simple to humans; vision, motion, perception, prediction, abstraction, are that way because humanity has collectively evolved to be good at those. But areas such as highly complicated mathematics are not a natural feature of collective humanity. It has to be learnt by every human being if at all. But that very thing is what a computer can be so good at. So maybe in order for a machine to be good at what it isn’t, and at the same time be intelligent, the best way is to make this whole thing the computer’s problem. What this means is to program the computer to be student and teacher at the same time. Let it be an AI researcher allowing itself not only to learn, but also to improve and figure out how to make itself smarter. Let it evolve.
The final two calibers would be massive milestones for AI scientists. As of now they are highly theoretical and pose both solutions and problems. A super intelligent machine could do things we couldn’t even dream of doing, with intelligence a trillion times greater than ours, it could do science that is way beyond human capability. On the other hand, such an intelligence could also come to think of humanity as an unevolved population of primates. The consequences are unpredictable and scary in that case. But the trajectory of AI is such. These are leaps and bounds ahead from where we are now; the incipient stages of AI application. But based on the rate of technological advancement, they aren’t too out of reach. If they come to pass, the criteria for ‘intelligence proper’ would not only be met, it would far surpass it to the point of redefining intelligence for machines whose thinking would be on a different class altogether.
REFERENCES
1. Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig.
2. http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html
