History of AI

AI is not new: the idea is more than 70 years old, and it has gone through several ups and downs.

The Early Ideas (1950s)

In 1950, the British mathematician Alan Turing asked a famous question: "Can machines think?" He suggested a test: if a person chatting by text cannot tell whether the other side is a human or a machine, the machine could be called intelligent. This became known as the Turing Test.

In 1956, a group of researchers met at a summer workshop at Dartmouth College in the USA. The term "artificial intelligence" was used there, and AI became a field of study.

Rule-Based AI (1960s to 1980s)

Early AI programs used hand-written rules. Programmers wrote "if this, then that" logic. Later, expert systems stored the knowledge of human experts as hundreds of rules, for example to suggest a medical diagnosis.

A tiny rule-based "expert" might look like this:

symptom = "fever"
if symptom == "fever":
    print("Advice: rest, drink water, and see a doctor if it lasts.")
elif symptom == "cough":
    print("Advice: warm drinks and rest.")

Advice: rest, drink water, and see a doctor if it lasts.

The problem: real life has too many cases. Writing rules for everything was slow and the systems broke easily.

AI Winters

Twice, people expected too much from AI too soon. When results were disappointing, funding was cut. These quiet periods (around the 1970s and late 1980s) are called AI winters.

Machine Learning Grows (1990s to 2000s)

Instead of writing rules, researchers let computers learn rules from data. This is machine learning. In 1997, IBM's Deep Blue beat the world chess champion Garry Kasparov. Spam filters and recommendation systems began learning from millions of examples.

# The machine learning idea: learn a rule from data
hours = [1, 2, 3, 4]
marks = [40, 55, 70, 85]
# The pattern: each extra hour adds about 15 marks
rate = (marks[-1] - marks[0]) / (hours[-1] - hours[0])
print("Each hour of study adds about", rate, "marks")

Each hour of study adds about 15.0 marks

The Deep Learning Boom (2010s)

Three things came together: huge amounts of internet data, fast graphics chips (GPUs) and better neural network methods. In 2012, a deep neural network won a major image recognition competition by a large margin. Soon deep learning powered voice assistants, translation and self-driving car research. In 2016, DeepMind's AlphaGo beat a world champion at the game of Go.

In 2017, researchers at Google introduced the Transformer, a neural network design that became the base of modern language models.

The Generative AI Era (2020s)

Large language models trained on enormous amounts of text learned to write, answer questions and write code. When ChatGPT launched in late 2022, millions of people used AI chat for the first time. Since then, models like ChatGPT, Gemini and Claude have added image understanding, voice, the ability to use tools and to act as agents that carry out multi-step tasks.

Timeline

Exercise