AI: a technological revolution built on long-standing foundations

Artificial intelligence (AI) is arguably one of the most widely used terms today, both in the media and in business. Virtual assistants, self-driving cars, machine translation, facial recognition: so many applications that feed the idea that we are living through an unprecedented revolution.

Yet behind this now-ubiquitous term, the reality is more nuanced. Most of the technologies grouped under the label of “artificial intelligence” are not recent. They have existed for several decades and have evolved gradually, in step with advances in computing power, data availability and algorithm efficiency.

This article offers an overview of the main technological building blocks of AI, tracing their origins and their evolution.

Beginnings in the 1950s

The history of modern AI begins in the 1950s, a period when the first computers appeared. Alan Turing then posed a question that has remained famous: “Can machines think?”.

From that time onwards, researchers attempted to simulate certain human capabilities using programs: recognising sounds, analysing images, or solving logical problems. A few examples:

• Speech recognition : in 1952, Bell Labs presented “Audrey”, a machine capable of recognising digits spoken by a human voice.

• Computer vision : in the 1960s, MIT launched the “Summer Vision” project, a pioneering attempt to have a computer interpret simple images.

• Expert systems : in the 1970s, software such as MYCIN was already assisting doctors in diagnosing infectious diseases, based on hand-coded rules.

These technologies remained limited by the computing power available at the time, but they laid the foundations of what would become AI.

Machine Learning: a long journey

The machine learningis often presented as the essence of modern AI. It involves designing algorithms capable of learning from data, without being explicitly programmed for each task.

Its history is marked by key advances:

• 1957 : Frank Rosenblatt developed the perceptron, a primitive form of neural network.

• 1980s–1990s : the emergence of statistical methods still widely used today, such as decision trees, Bayesian networks and support vector machines (SVM).

• 2006 : Geoffrey Hinton and his colleagues revived research into deep neural networks (deep learning), showing that they could outperform traditional approaches.

Since then, machine learning has established itself as an essential pillar of modern technologies: from speech recognition to recommendation systems, as well as demand forecasting and predictive maintenance.

Image recognition: from the laboratory to everyday use

The Image recognition is now everywhere: smartphones, security cameras, biometrics, industrial inspection.

This technology has evolved gradually:

• In the 1990s, Yann LeCun developed LeNet-5, a convolutional neural network designed to recognise handwritten digits.

• For nearly 20 years, these models remained limited by computing power and restricted access to data.

• In 2012, a decisive turning point came with AlexNet, which won the ImageNet competition by shattering the performance of classical approaches.

Since this breakthrough, convolutional neural networks have dominated the field, enabling applications ranging from medical diagnosis to autonomous driving.

Natural language processing: from statistics to deep learning

The Natural language processing (NLP) is one of the most visible branches of today’s AI, with virtual assistants and chatbots.

But here too, the origins go back a long way:

• In 1954, the Georgetown-IBM experiment automatically translated around sixty sentences from Russian into English.

• In the 1990s, probabilistic statistical models became dominant.

• In 2017, the Transformer architecture changed the game, paving the way for giant models such as BERT, GPT and LLaMA.

These models are capable of generating coherent texts, translating, summarising and even answering complex questions. They rely on learning from billions of sentences and on unprecedented computing power.

Speech recognition: a discreet but steady evolution

The reconnaissance vocale is another field that has evolved considerably:

• In 1952, “Audrey” could only recognise digits.

• In the 1980s, models based on Hidden Markov Models (HMM) considerably improved accuracy.

• In the 2010s, neural networks revolutionised the field, giving rise to Siri (2011), Alexa (2014) and Google Assistant (2016).

Today, speech recognition powers both personal assistants and industrial applications, particularly for operators in production.

AI in games: historic milestones

Games are a privileged testing ground for AI:

• In 1997, Deep Blue from IBM beat Garry Kasparov at chess.

• In 2016, AlphaGo from DeepMind astonished the world by beating Lee Sedol, one of the best Go players.

• In 2020, AlphaFold solved a major problem in biology by predicting the structure of proteins with unprecedented accuracy.

These milestones show that AI is not only an applied technology, but also a tool for scientific research.

Why is AI so visible today?

If AI occupies such a place in our lives today, it is not because it is “new”, but because several favourable conditions have come together:

1. Computing power : GPUs and specialised processors make it possible to train complex models.

2. Data availability : the internet, sensors and connected devices produce colossal volumes of data.

3. Modern algorithms : in particular deep neural architectures such as Transformers.

4. Democratisation : the tools are accessible to the general public, via APIs, open-source software and even smartphone applications.

It is this combination that explains the recent explosion of visible AI applications.

Conclusion

Artificial intelligence is often presented as a sudden disruption, but it is in fact the culmination of more than seventy years of research. Speech recognition, computer vision, natural language processing, machine learning: all these building blocks already existed, sometimes in rudimentary form, as early as the 1950s.

What is changing today is the scale, the maturity of the algorithms and their integration into our daily lives. Rather than a passing trend, AI is the result of a gradual evolution, one that is now profoundly transforming industry, services and scientific research.