Artificial Intelligence
A Short History of AI
The concern of society with machines dominating the human race has existed for a long time. This fear or concept of “intelligent machines” was first elaborated by the philosopher and mathematician René Descartes in the book “Treatise on Man” which was published in 1637 and is one of the most important books in modern philosophy. Descartes makes some important reflections that have influenced the development of “thinking machines” or automatons, concepts that would be used in the future in the development of AI.
The Automatons
In his view, the human body was a complex machine, where emotions and movements could be explained by physical parts interacting, and that the machines could not think, but could mimic certain aspects of human actions. In his time, he was aware of “automatons”, which are mechanical contraptions that mimic and simulate human and animal movements and behaviour. But he highlighted that unlike humans such devices had no soul or true intelligence but are just a copy, an imitation of human intelligence and actions, with no conscience and intelligence. This is very much the same concept as our current AI, or Narrow AI, that we will talk about as we continue in the module.
These machines could execute complex functions in an automated way, but without any cognitive or reflexive capacity that is fundamental for genuine intelligence.
Descartes’ vision of automated machines had an important influence on the later development of mechanics and computer science. During the seventeenth century, the concept of machines copying human behaviour started to gain prominence and attention, especially through some devices created by Jacques de Vaucanson. These were machines that simulated natural movements.
In the nineteenth century, the Cartesian view to analyse and treat humans as machines led to the work of Charles Babbage, who produced the Analytical Machine (seen as the first modern computer). This led in turn to the development of computers capable of performing tasks that were previously only done by humans, like complex calculations and eventually data-based decisions.
The concept of a machine being able to mimic human behaviour, copying behaviour without conscience, cognitive capacity or common sense, all started in 1637 with Descartes.
Alan Turing was another important figure in the development of AI, developing the concept of how smart a machine is and how much it resembles human intelligence. Turing was famous as the scientist that broke the code of the Nazi Code Enigma Machine. He is also considered one of the fathers of computer science and AI. He contributed not only to the theory of AI, but also to the debate about what intelligence means, and how it can be mimicked or replicated by machines.
The Turing Machine and Turing Test
Turing created the “Turing Machine”, a theoretical computer that could simulate any computer algorithm. He proved that any problem that a human could solve only by using a set of rules or algorithm could be done by a machine, as long as there is enough processing capacity and computing power. This concept is fundamental for computer science development and AI, when we talk about the building of algorithms and machine learning.
Another very important contribution from Turing was the Turing Test, which is still used to measure how “intelligent” a computer is. The test consists of an interaction in natural language between a machine and a human being. If the human being cannot identify that he or she is interacting with a machine, the machine can be considered intelligent. The test does not measure if the machine has a conscience, common sense or empathy, but if the machine is capable of convincingly mimicking human intelligence.
Turing never used the term Artificial intelligence, but there is no doubt that his ideas, studies and contributions were fundamental for the development of AI as we know it today. (The term AI was created in 1950 by John McCarthy.) Turing also predicted that machines would be able to learn from experience, like machine learning concepts and capabilities of today. In an article in 1950 he suggested that instead of explicit programming, a machine would be able to “teach to itself”, thus executing tasks based on experience, which is a basic system in current machine learning. We will talk more about machine learning as we progress in the module. His concept of machine self-teaching helped the build of the concepts of unsupervised machine learning.
From 1950 to 1990, alongside the development of computers and databases, the development of AI has had many iterations. Machine learning has improved as data storage and processing capacity improved. This involved the concept of “big data”, translated into the much-improved capacity of data storage, cleaning, and treatment, the availability of big digital data centres, and the improvement of computational processing capacity of the machines. Companies can now scale up storage and processing in cloud centres rather than on-site machines.
In the 2010s, AI started to expand in the areas of voice recognition (Siri and Alexa), image recognition (Facebook and Google) and self-driven cars (Tesla).
In the 2020s AI had another boost via generative models. The big breakthrough in Large Language Models was the ability to mimic human interaction by understanding natural language. These LLMs are trained on massive dataset before being used for specific tasks, allowing them to learn patterns and context in language, using transformer architecture. The transformer is a deep learning architecture that enables the model to understand the relationships between words in a sequence. ChatGPT creates human-like text and Dall-E creates images from simple descriptions. These advances mean AI is in everyday use in homes, schools and companies.
AI Definition – What is AI?
The best definition of AI is the technology that enables computers and machines to simulate human intelligence and problem-solving capacity, not only by copying the human way of dealing with a problem or subject but adapting and learning.
Today, in essence, what AI does is to process a huge amount of data and seek common parameters to provide the answer to the question. However, it does that without a sense of the quality of the output provided, a subject for us to discuss further.
Much of what we have in AI today has existed since 1990, as we have described in the “history of AI session”. We have had AI since the moment that companies were able to deal with large amounts of data (big data) and create hypotheses and learn from common outputs.