This question has long preoccupied mankind. From early automata to HAL 9000 to Ava from the film Ex Machina, the idea of artificial, man-made intelligence permeates our culture and our stories. In reality, we are still far from this general artificial intelligence, but the field has made rapid progress in recent years. But how did we get to where we are today?

1950: TURING TEST

Intelligence is notoriously difficult to define. In 1950, the British mathematician and computer scientist Alan Turing formulated the test named after him to determine when one can speak of machine intelligence. If in a (written) dialogue a person cannot tell whether they are talking to a machine or to another person, one could speak of machine intelligence.

1956: DARTMOUTH CONFERENCE

The birth of the academic discipline, this conference was the first to speak of artificial intelligence. The term caught on. The initiators John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon requested 13,500 US $ from the Rockefeller Foundation to host the conference. They started with big plans: in two months, with ten participants, all aspects of learning and intelligence were to be described so that a machine could be built that simulates these processes.

1966: ELIZA THE FIRST CHATBOT

ELIZA was a computer program developed by Joseph Weizenbaum to show the possibilities of how computers can enter into a dialogue with people via so-called “natural language”. The best-known form of ELIZA simulated a psychotherapy in which the computer searched the statements people typed in for key words and played them back in a modified form. The program was surprisingly successful, although users were able to push ELIZA to the limits of its capacity quite quickly, because the program did not learn anything new, but was pre-programmed according to certain logics.

1972: MYCIN AND THE BEGINNING OF EXPERT SYSTEMS

Expert systems are computer programs that support people in solving complex problems. MYCIN was developed at Stanford University to support the diagnosis and treatment of infectious diseases with antibiotics. MYCIN analyzed numerous variables to identify the pathogens and recommend the best antibiotics – tailored to the patient’s individual parameters. However, despite the high success rate, MYCIN was not used in practice, as scepticism was too great and the technical basis for successful scaling was not yet in place.

1997: DEEP BLUE WINS AGAINST WORLD CHESS CHAMPION

Deep Blue, developed by IBM, beat the reigning world chess champion Garry Kasparov in 1997. Unlike today’s systems, Deep Blue did not “learn” the game, but beat its human opponent through sheer computing power.

2011: WATSON WINS QUIZ SHOW

In the American quiz show Jeopardy, players not only have to answer questions, the questions sometimes include word games. Watson, developed by IBM, competed against the two best Jeopardy players in 2011 and won. The AI showed that they could understand and answer questions.

2016: ALPHAGO CRACKS GO

For a long time, Go was considered the game that would take artificial intelligence years to master. This is mainly due to the complexity of the game. If the first player in chess has 20 possible moves to choose from, Go has 361. Sheer computing power, as in Deep Blues chess game, would not crack Go. With Reinforcement Learning, Google’s AlphaGo learned the game and in 2011 it faced Lee Sedol, a South Korean who is considered one of the best players. AlphaGo beat Lee Sedol 4-1.

2018: DUPLEX MAKES AN APPOINTMENT

Google is demonstrated duplex, an artificial intelligence, and had it call a hairdresser to make an appointment. The voice and style of speaking are indistinguishable from a humans.

AI WINTER AND AI SUMMER

The history of artificial intelligence ran in waves. Great enthusiasm (summer) was followed by disappointed expectations and cuts in research funding (winter). Often challenges were underestimated and progress overestimated. For example, the AI researcher Marvin Minsky said in an interview in 1970: “In three to eight years we will have a machine with the general intelligence of an average human being”. Expectations were not fulfilled, among other things due to the amount of data available and the comparatively weak computing power at that time. Known as Moravec’s paradox, researchers repeatedly found that things that are very difficult for a human being, such as solving complex mathematical problems, are easy for a computer, while things that are easy and self-evident for humans, such as recognizing images, understanding speech or performing movements, are extremely complex and difficult for machines.