Artificial intelligence: decades of exploration
R. Ramanujam, Azim Premji University, Bengaluru

This is the second article of a series on Artificial Intelligence.

Today Artificial intelligence (AI) is a buzzword. We associate it with "cutting edge" technology of the "latest" kind, and even more with futuristic vision. For many, the very term AI evokes a vision of robots taking control of the world and running amok.

In reality AI systems have evolved over a long period of time, over more than six decades. Research in AI started in the days when one communicate with computers only through punched cards and paper tapes, and main memories were very minuscule compared to today: 16K or 16000 bytes, as opposed to today's 16G or 16 billion bytes. Yet, researchers were fascinated by the *idea* of AI, and were already working on getting computers to recognise patterns, to attempt conversations with human beings, to play games, to prove mathematical theorems and to solve problems.

Why does the size of memory matter? Before 1949 computers lacked a key prerequisite for intelligence: they could not *store* commands, only execute them. That is, computers could be told what to do, which they did, but they couldn’t remember afterwards what they had done. Can you imagine what it would be like for you to do things each day but forget how the very next day and needing to learn anew each day? That was how it was for computers at that time. The idea of stored program computer, initiated in the 1950's by the great Hungarian American mathematician John von Neumann, based on brilliant work by English mathematician Alan Turing in the 1930's, showed how computers could store their "programs", fetch them from memory and execute them when needed. This was a major breakthrough. But the memory size was very smal, this was a technological limitation.

Further, building computers was a very expensive activity. Only prestigious universities and big technology companies could afford to dillydally in these uncharted waters.

Meanwhile, scientists were already studying mathematical models of the human brain, in the hope that it would lead to building computers along similar lines. In 1943, the first mathematical model of a *neuron* was proposed by McCulloch and Pitts. In this view, the brain is a vast network of millions of interconnected neurons. Such *neural networks* are the essential components of the technology that has led to most of the major recent breakthroughs in AI. Based on these, John von Neumann proposed *cellular automata* in the 1960's, which became important in physics later on. But while this line of research went on in parallel, computers were mostly built on the stored program concept outlined above.

However limited the memory, people went ahead. In 1952, Arthur Samuel wrote one of the earliest programs to play a *board game:* the game of Checkers, where counters are placed on a Chess board subject to some rules.

In 1956 Allen Newell, Cliff Shaw, and Herbert Simon wrote a program called The Logic Theorist, to prove theorems formally in logic. It was a program designed to mimic the problem solving skills of a human being. 

That year, Marvin Minsky and John McCarthy organised a conference in which McCarthy coined the term *artificial intelligence*. There was failure to agree on standard methods for the field, but everyone aligned with the sentiment that AI was achievable.

Meanwhile in 1957 Frank Rosenblat showed that neural networks could "learn" by adjusting weights on edges of neurons adaptively. This was a stunning idea as such learning would be different from conscious inference based learning.

From mid-50's to mid-70's AI flourished. During that period, computing technology changed dramatically. Computers became faster, cheaper, more accessible and they could store more information. Machine learning algorithms on neural networks also improved and people got better at knowing which algorithm to apply to their problem. Newell and Simon’s General Problem Solver attempted to construct a strategy to try any given problem. 
Weizenbaum’s ELIZA program could talk like a psychiatrist to people. Governments in many countries started to give funding for AI research in universities.

A major challenge arose: the need for a machine that could transcribe and translate spoken language as well as perform high volume data processing. It was clear that there was still a long way to go before the end goals of natural language processing, abstract thinking, and self-recognition could be achieved. 

The biggest problem was the lack of computational power to do anything substantial: computers simply couldn’t store enough information or process it fast enough. It was with the advent of the first microprocessors at the end of 1970 that AI took off again and entered the age of *expert systems*, which mimicked the decision making
process of a human expert.

An expert system would ask an expert in a field how to respond in a given situation, and once this was learned for virtually every situation, non-experts could receive advice from that program. Expert systems began to be  widely used in industries. Expert systems were built for molecular chemistry, for diagnosis of blood
diseases and prescription drugs, and so on. The Japanese government announced a major project for AI research based on expert systems.

Scientists realised that building machines with all AI capabilities would be difficult, so they divided up the work: some studied computer vision, some computer speech, some game playing, and so on. This proved to be very effective, each could develop in parallel. By now, storage was no longer a problem. 

In 1996 the computer program Deep Blue beat grandmaster Gary Kasparov in a chess match. By the turn of the century AI research had made great advancements in each subarea like vision, robotics, planning etc but combining these seemed a daunting task. 

Meanwhile computer science changed the world: the *Internet* arrived! Use of search engines such as Google and Yahoo changed people's every day lives. In the first decade of the 21st century, *social media* became prominent, in which people could share images, videos etc. Thus started the age of *Big Data*. Suddenly computers had access to huge amounts of data provided voluntarily by half of humanity, and fast algorithms to automatically process them became available. There was no longer any need to code up behavioural rules (as in expert systems) but let computers "discover" rules from volumes of data. 

The last decade has been one of dramatic developments in AI. In 2012, AlexNet by Krizhevsky and Hinton gave the first significant results in *deep learning*, a layered technique that made machine learning much faster and more effective. In 2016 the program DeepMind (based on deep learning) could beat a professional Go player. In 2022, ChatGPT, based on a Large Language Model, was released. ChatGPT can write essays and poetry, carry out long conversations with human beings.

Today we see AI applications advancing not only into all areas of science and technology but also arts and the humanities. AI carries the potential of tremendous benefits but also terrible social harm. Whether the modern human will be wise enough to use this technology and at the time retain control over its use, is unclear.