The early years of AI saw rapid progress and the creation of several iconic programs.
The Perceptron (1958)
In 1958, psychologist Frank Rosenblatt built an electronic device called the Perceptron – the first model of an artificial neural layer with learning capabilities [citation:1][citation:9]. It was a program that made decisions like the human brain and could teach itself to distinguish between punch cards marked on the left and right [citation:7]. Rosenblatt described it as “the first machine which is capable of having an original idea” [citation:7].
The Perceptron is regarded as the first artificial neural network and the ancestor of modern deep learning systems [citation:7]. However, its performance was limited, and it could only solve simple problems where input categories were “linearly separable” [citation:1]. This limitation would lead to criticism and eventually contribute to the first AI winter.
ELIZA – The First Chatbot (1966)
In 1966, MIT computer scientist Joseph Weizenbaum created ELIZA, a program designed to simulate a psychotherapist [citation:1][citation:7][citation:11]. ELIZA used scripted responses and repurposed the user’s answers into questions that prompted further conversation – a technique known as the Rogerian argument [citation:7].
Weizenbaum believed this back-and-forth would prove the simplistic state of machine intelligence [citation:7]. Instead, many users came to believe they were talking to a human professional [citation:7][citation:1]. ELIZA is widely considered the first chatbot and sparked early excitement about language-based AI [citation:11].
Shakey the Robot (1969)
Shakey is recognized as the first general-purpose mobile robot [citation:7]. It could perceive its surroundings, create plans, recover from errors, and communicate using ordinary English [citation:7]. Shakey had a television monitor and whiskers to detect when it came close to objects [citation:7].
These early milestones demonstrated the potential of AI, but they also revealed significant limitations. The technology of the time simply could not support the grand visions of AI researchers.