Over the past three years, it seems as though the entire tech world has become obsessed with a single topic, and that topic is… AI! Although artificial intelligence agents have only become widespread since 2022, they are so ubiquitous that it sometimes feels as though they’ve always been around, as if there were no history of artificial intelligence.

It’s no wonder: we are in the GenAI era, as defined by a 2024 McKinsey study. Just as baby boomers were taken aback by how easily Generation X handled computers, and just as Generation X was amazed by how easily millennials navigated the Internet, the generation being born today has come of age after the widespread adoption of artificial intelligence.

To put it another way, AI is the kind of innovation that defines an era. Just as we can no longer imagine life without electricity, fast transportation, telephones, computers, and the Internet, it will be impossible to imagine life without artificial intelligence.

Since it has become virtually the sole topic of discussion in the tech industry and the corporate world, AI seems like a brand-new phenomenon. But the history of artificial intelligence is far from short. Join us in this article to understand where it comes from and, consequently, what paths lie ahead!

Concept

Before the story, the reason why. We can contrast the term “artificial intelligence” with “natural intelligence.” Given that the human species is the most intelligent on the planet, when we talk about natural intelligence, we are talking about our ability to think and construct complex reasoning.

AI as we know it today exists only because, more than 70 years ago, humanity imagined it might be possible to emulate the functioning of the brain. The most iconic paradigm in computing revolves precisely around the hypothesis that machines can think like people. In 1950, Alan Turing suggested that computers had evolved to the point where it would be impossible to distinguish whether we were interacting with humans or programs. In his article “Computing Machinery and Intelligence, ” the English mathematician proposes the imitation game, an exercise in which an evaluator interacts, separately and via text messages, with a human and with a machine. If, at the end of the exercise, the evaluator could not distinguish who is who, the machine would have won.

To win the imitation game, a program would need to “think.” Based on this premise, scientists drew inspiration from neural networks to develop ANNs (artificial neural networks): computational models capable of recognizing patterns, classifying them, and learning from them.

Just like biological networks, artificial neural networks have layers. And just like any computer program, they have an application architecture. In practice, the number and organizational structure of the layers define the architecture of the ANN. The deeper the network, the greater its learning potential.

The History of Artificial Intelligence: Nearly 100 Years of Research

Even before the Turing test, scientists believed that machines would be capable of emulating human thought. Even before the term “artificial intelligence ” existed, Walter Pitts and Warren McCulloch, in a 1943 article titled “A Logical Calculus of the Ideas Immanent in Nervous Activity, ” proposed a simplified model to try to illustrate how the human brain supposedly works, mathematically speaking.

Pitts and McCulloch’s ideas paved the way for the development of artificial intelligence. In 1952, Marvin Minsky introduced Snarc, the first machine built with an artificial neural network and, therefore, capable of learning. That same year, Arthur Samuel demonstrated “Checkers-Playing” at IBM, his software for playing checkers. It was the first self-learning software in history.

Just as in biological neural networks, artificial neural networks also start with the “neuron.” The “artificial” version was introduced in 1958 by Frank Rosenblatt and was named the perceptron: a linear classifier that maps input values to an output value.

The perceptron algorithm calculates an output value from a set of input values, which are weighted to express the importance of each input value. The calculation performed by the perceptron is a weighted sum of the input values and takes into account a bias value, a constant term that does not depend on the input values.

The output value of each perceptron is necessarily one of the input values of all perceptrons in the next layer. In this new layer, the perceptrons perform the same calculation, the weighted sum of the previous values, taking into account a bias value—and so on—until reaching the final value of the artificial neural network.

In the 1960s, artificial intelligence moved beyond universities and corporate research centers to find industrial and commercial applications. Using Lisp, a programming language developed by John McCarthy—the creator of the term “artificial intelligence”—a team of scientists led by Edward Feigenbaum, Joshua Lederberg, and Carl Djerassi programmed Dendral: a system that, by analyzing data on chemical elements, was able to identify the structures of large organic molecules.

The 1960s also saw the emergence of the first chatbot to use natural language, the first industrial robotic arm, and the first robot capable of sensing its surroundings to move around… AI was becoming increasingly part of everyday life.

However, it was in the 1980s that a pivotal milestone was reached in the history of artificial intelligence, paving the way for the AI we know today: the backpropagation algorithm, described in the paper “Learning Representations by Backpropagation of Errors” by David Rumelhart, Geoffrey Hinton, and Ronald Williams.

The algorithm allows a signal to be sent backward—or “backpropagated”—through the neural networks, indicating how much and in which direction each perceptron contributed to the final error. The signal indicates an error sensitivity, that is, how much a difference in values—whether in the inputs or the weights—would alter the output value. This signal is used to calculate the adjustment of the weights for each connection.

Thanks to the backpropagation algorithm, it has been possible to develop artificial neural networks with a greater number of layers and output values that exhibit a much lower error rate.

As we mentioned at the beginning of this article, artificial intelligence marks the dawn of a new era. Just as we’ve electrified everything we could—from showers to cars—and seen the internet make its way into even the smallest devices, we’ll see the same thing happen with AI.

If your company hasn’t yet implemented artificial intelligence in its processes, you can reach out to us to take this transformative step for your business. Here at Inmetrics, we have a Consulting, Data, and Artificial Intelligence unit that is fully equipped to apply AI in the context of your company. To learn about all the possibilities, click here and get in touch!