3 Rise of the AI Phenomenon

For a lot of people, their first recollection of the possibility of computers being smarter than humans harks back to 1997 when IBM supercomputer Deep Blue beat then reigning world chess champion Garry Kasparov. Up until then, the idea that computers would excel at tasks that required strategic and analytical thinking had mostly been the stuff of science fiction—or academic research. Deep Blue’s victory over the Russian grandmaster changed that. Suddenly, the idea that humans could train computers to undertake tasks to an extent where they could then outperform humans was no longer a nebulous concept but reality.

What most people don’t realise is that this victory was decades in the making. Deep Blue was a result of a college project that started in 1985, which then underwent several iterations. The earlier versions of the program were good but not good enough to outplay a human being. Kasparov defeated an earlier iteration of Deep Blue as late as 1996, less than a year before he lost to the computer program.1

This event firmly established the notion that it was possible to train machines to perform tasks that went beyond simple automation and instruction and required strategic thinking and analysis. Deep Blue’s triumph built upon decades of work that went into a broad field of research that has come to be known as AI, the earliest seeds of which were sown in the 1950s.

Before going any further, it is important to be clear on what AI means. In simple terms, AI is a simulation of the human intelligence process by computer systems and machines. The Oxford English Dictionary defines it as ‘the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages’.2 Harvard University’s Marvin Minsky, who is known as one of the fathers of AI, defines it as ‘the science of making machines do things that would require intelligence if done by men’.3 Practically, this translates into the ability of computer systems to carry out tasks that normally require human cognition and understanding. This can mean helping a person buy the appropriate insurance plan for their specific needs or speeding up the early stages of new drug development.

The genesis of AI can be traced back to 1950. Noted British computer scientist and mathematician Alan Turing published a paper in 1950, Computing Machinery and Intelligence, where he introduced the world to the Turing Test or what was then called the Imitation Game. This is the ability of a machine to exhibit intelligent behaviour similar to a human being without being found out. The way to test this is if a computer program can hold a conversation with a human being without the person realising that they are talking to a computer.4 This test, in many ways, has come to be symbolic of the first time AI entered the broader scientific discourse. Over the years, only a handful of supercomputers have come close to passing this test.

A lot of the foundational work in the field of AI can be traced back to the early 1950s in the United States of America, a time before AI was a recognised term or a field of study. At Carnegie Mellon University, Marvin Minsky and Dean Edwards went on to develop the world’s first artificial neural network. This is a system of artificial neurons that mimics the neurons in the human brain. The development and use of an artificial neural network paved the way for the introduction of a type of machine learning process called deep learning. Over the next few decades, this was fundamental to the advances made in the field of AI and how machines could be taught to train themselves.5

Building upon this same line of reasoning, Arthur Samuel created the Samuel Checkers-Playing Program at IBM, which was the world’s first self-learning program to play games. The system was designed such that it could learn how to play checkers based on the results of games that had already been played. Over time, this ability of the machine to learn from experience rather than explicitly fed in data came to be known as machine learning, and Samuel is credited with coming up with this system of training.

All these were scattered instances of experiments that were building up towards the establishment of AI as a separate, stand-alone field of scientific research. Notably, in the early 1950s, all this development was still being referred to as synthetic intelligence.6

The pivotal moment for AI came in 1956 at Dartmouth College in New Hampshire, United States of America. It was the Dartmouth Summer Research Project on AI that led to the birth of this field of research. The project was the brainchild of John McCarthy, who was then a mathematics professor at the college and who later went on to set up the AI Lab at Stanford University and at Massachusetts Institute of Technology. The foundation for this project was laid by McCarthy and Minsky, along with Nathaniel Rochester of IBM and Claude Shannon of Bell Telephone Laboratories, all of whom were among the pioneers in the AI movement and came to be known as the fathers of AI. This is the first time the term artificial intelligence was used, and McCarthy has been credited with naming the then nascent stream of science. The summer of 1956 is also recognised as the formal birth of this stream of science and research.7

Around this same time, parallel advances were being made in what would come to be known as automation, which would deeply impact the manufacturing industry. Automation, at its most basic level, has been core to any kind of industrial development in the past century. Even as it is currently understood, automation is not a new concept. In the simplest of terms, automation is the use of any kind of technology that enables a process to happen with minimal human assistance or inputs. This, in turn, is supposed to result in improved efficiency, productivity and the ability for machines to perform increasingly complicated tasks. In the current context, the term automation is most often used in two instances: the use of machines on the shop floor and robotic process automation (RPA), which is using software to perform repetitive tasks.

There is also business process automation, which uses software to automate end-to-end processes within the organisation.8 However, all of this only gained steam in the 1980s and 1990s as computers and software programs became more ubiquitous in offices. Up until then, automation almost exclusively referred to the use of technology and machines in the manufacturing sector.

The mid-1950s marked the start of a period of high excitement and exuberance in the AI research community. Fresh off the success of the conference, John McCarthy went on to develop LISP, the first AI programming language. LISP (list processing) became the foundation of AI research and continues to remain in use till date among AI researchers. Through the decade, scientists across the US were working on different projects to demonstrate how computers could be trained to think and process information like humans. Two standout examples are ELIZA, the world’s first chatbot, and Shakey, the world’s first intelligent mobile robot. Unlike other robots that were in use at that time that needed step-by-step directions on how to undertake a large task, Shakey had the ‘intelligence’ to analyse and assess for itself the series of steps needed to complete a larger task. It used a combination of NLP and robotics and was a rare example of bringing together logical processing and physical action.

Then expert systems were created. Expert systems were designed to replicate human behaviour, to the extent that it was possible, and were modelled on human knowledge. Such a system could, hypothetically, analyse a person’s medical history and reports and suggest a diagnosis to a physician.9 For the first few decades, all the work being done in developing the field of AI was restricted to research laboratories. It still hadn’t reached a stage where it had real-world applications and remained a ‘promising area of research’ for several years. Progress in AI research coincided with a period where computers were gradually getting better, faster and more accessible. Government interest in AI solutions was also increasing, leading to more research and development in the field. Scientists were optimistic that within the next decade, they could build a machine with the general intelligence of an average human being. Computers were still prohibitively expensive, and while storage capacity was improving, it still had to get a lot better for machine learning algorithms to work in the way scientists envisaged they would.