StudyAI traces AI ideas back long before computers, from Hero to Geniac

StudyAI has put together a vivid retrospective showing that the idea of artificial intelligence appeared much earlier than electronic computers. Long before neural networks, people were already trying to build mechanisms that looked like "thinking" machines and carried out actions according to set logic.

A dream older than the computer

The article's central point is simple: if you treat AI not just as a program on a chip but as a system that follows an algorithm and imitates intelligent behavior, its history does not begin in the 20th century. The author draws a line from ancient automata to mechanical toys and teaching machines that worked without processors, screens or cloud infrastructure. These devices did not think in the modern sense, but they show that people have long tried to turn rules, sequences and feedback into machine behavior.

One of the earliest examples is an Egyptian proto-automaton from the Middle Kingdom, depicting the figure of Hathor with a simple internal mechanism. Later, Hero of Alexandria took the idea to a different level: his devices could stage scenes, move figures and even dispense holy water on a principle similar to a vending machine. In the Middle Ages, Arab engineers picked up the baton: Al-Jazari described complex automata, including a hand-washing device that offered soap and a towel on its own. All of this looked like mechanical theater, but with clearly defined logic behind the actions.

Mechanics as algorithm

The best-known example in the selection is the Antikythera mechanism. It is usually called the first analog computer, but the text presents it more broadly: as an ancient machine that turned astronomical observations into useful guidance. The device tracked the cycles of celestial bodies, helped predict eclipses and calculate dates. In essence, it is a compact decision-making system built from gears: the user enters input data, and the mechanism returns a meaningful result. For an era without electronics, this looked almost unbelievable.

The selection matters not for its exoticism but because it shows several distinct models of machine behavior. Some devices replayed a predetermined script, others calculated cycles and helped with forecasting, and still others reacted to obstacles almost like autonomous robots. These were not just toys for an audience but practical experiments in logic, prediction and control. That is why these historical examples read today not as curiosities but as early versions of engineering ideas familiar from modern robotics and AI.

• Hero of Alexandria's automata staged little scenes and ran on a pre-built script
• the Antikythera mechanism calculated astronomical cycles and indicated dates of important events
• Leonardo da Vinci's robot knight could move, sit, stand and raise its visor
• Geniac taught Boolean logic, played tic-tac-toe and even assembled simple musical sequences
• the Soviet "Umka" steered around obstacles and did not fall off a table, thanks to pure mechanics

Leonardo da Vinci's robot knight holds a place of its own. Judging by surviving sketches, it was built on cables, pulleys and gears that mimic human biomechanics. The machine could move its limbs, change the position of its torso and perform a set of spectacular actions at a carnival at the court of Ludovico Sforza. For a modern reader, it is less a Renaissance curiosity than an ancestor of humanoid robotics. Projects like this break the familiar picture in which the history of AI starts only with the mathematical models of the 20th century.

Toys before neural networks

Closer to the present, the selection moves from spectacular automata to devices that already resemble computing hardware. Especially telling is Geniac, a mid-20th-century mechanical educational computer with lights, dials, switches and jumpers. It had no memory and ran on combinational logic, but it let users build circuits, play games, solve simple problems and reprogram the machine's behavior by hand in a few minutes. It is an important bridge between the age of automata and the era of mass cybernetics.

The Soviet "Umka" is no less intriguing: a toy car that could steer around obstacles and avoid falling off a table without cameras, lidars or depth sensors. Its behavior came from a feeler, a movable support and a wheel mounted at an angle. When the car hit an obstacle or neared an edge, mechanical feedback changed its trajectory. By today's standards it is a very simple system, but that is exactly the strength of the example: useful autonomous behavior is possible even without computing power, if an engineer sets the rules of reaction to the environment correctly.

What this means

The StudyAI selection makes a clear point: the history of AI is not only neural networks, GPUs and large language models. It is a long evolution of ideas in which people spent centuries trying to teach machines to perceive a signal, follow an algorithm and react to the world without direct human involvement.