The History of Artificial Intelligence
A journey through the milestones of artificial intelligence — from 1943 to ChatGPT
1943 — McCulloch-Pitts neuron
The McCulloch-Pitts neuron is a neuron model proposed by Warren McCulloch and Walter Pitts. It regards an (artificial) neuron as an inactive and active switching element that can have the value 0 or 1, i.e. is binary. The McCulloch-Pitts neuron model is the simplest neuron model in neuroinformatics. Today the model is regarded as the first work on artificial intelligence, or artificial neural networks respectively.
1950 — Turing test / Imitation Game
In this year Alan Turing presented the test in his essay “Computing Machinery and Intelligence”. This test requires three spatially separated terminals, two of which are operated by humans and one by a computer. One of the humans is the questioner and asks the other two questions. Both the human and the computer must convince the questioner via the terminals (in writing) that they are human. If in the end the questioner cannot say which of the two is the computer, the machine has passed the Turing test.
1956 — “Artificial Intelligence”
The term “Artificial Intelligence” was officially mentioned for the first time in the summer of 1956 at the Dartmouth Conference. The full name of this conference was “Dartmouth Summer Research Project on Artificial Intelligence” and it was applied for, planned and carried out by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon. At this conference, ten participants met in order to find out (among other things) how they could get machines to use language, play games, learn and form abstractions and concepts.
1958 — Perceptron
The “perceptron” (from the English “perception”) is a mathematical model of an artificial neural network presented by Frank Rosenblatt. In its simplest variant it consists of a single neuron with one output and several inputs. The binary output function (0 or 1) is determined by the weighting of the inputs and by threshold values. Learning changes the weightings. With this model, a machine becomes capable of learning on the basis of the principle of trial and error. Typical applications of the model are the analysis of measured data (e.g. weather data) or image recognition.
1958 — LISP
In 1958 John McCarthy developed the programming language LISP (acronym for “list processing”), which is still used to this day — among other things in the field of fundamental programming and artificial intelligence through to machine learning and quantum computing. Historically, LISP is, alongside “Prolog”, one of the most important programming languages for AI and, after Fortran, the second-oldest language still in widespread use.
1966 — Chatbot ELIZA
Joseph Weizenbaum’s ELIZA program is regarded as one of the first chatbots. The aim was to show that a true dialogue between human and machine is not possible and always remains superficial. However, the users often did not care that they were communicating with a machine, as long as the answers seemed “human”. Feelings and understanding that it does not even possess were attributed to the machine — the so-called ELIZA effect was born. Through his experiences with ELIZA, Weizenbaum became a critic of “thoughtless faith in computers” and from then on repeatedly warned about the critical handling of computers, emphasising that control should always remain in human hands and that AI should merely be used as an aid for obtaining information.
You can still talk to ELIZA today: eclecticenergies.com/deutsch/psyche/eliza
1968 — Film “2001: A Space Odyssey”
Marvin Minsky, one of the founding fathers of AI and one of the participants at the Dartmouth Conference, served Stanley Kubrick as a technical adviser on the film with the supercomputer HAL 9000. Thanks to Minsky’s input, a portrayal of artificial intelligence that was coherent for the viewer and additionally scientifically substantiated could be shown. The film reflects the predictions of AI researchers at the time and is furthermore regarded as a cinematic milestone of artificial intelligence that inspires many technology visionaries to this day.
1966–1972 — Shakey, the first universal mobile robot
Shakey, who got his name from his shaking, was created by researchers at the SRI research institute and has a legendary status for its first-ever unification of robotics and AI in a single system. The robot is able to carry out its tasks independently, without step-by-step instructions, with the help of its tools (camera, radio antenna, detectors, bumper, etc.) — albeit slowly.
1970s — “AI winter”
In the early 1970s, AI research stalled after research, following multi-million investments, fell short of expectations (or rather could not meet the excessive expectations). The consequence was a cut in funding and the onset of the so-called “AI winter” due to the reduced research.
1972 — PROLOG
PROLOG (derived from the French “programmation en logique”) is one of the most important logical programming languages, enabling declarative programming. It was developed by Alain Colmerauer in the 1970s and is particularly suited to problems based on logical relationships, e.g. in the field of artificial intelligence and computational linguistics.
1972 — SHRDLU
SHRDLU is a program developed by Terry Winograd (deliberately named without any deep meaning) that was able to arrange blocks according to given criteria with the help of a robot hand. SHRDLU is regarded as the first program to combine the understanding of language with the simulation of planned actions. However, its success fell far short of expectations, as systems of that time had difficulties, e.g. in understanding language and forming categories (a task that humans solve with ease).
1972 — MYCIN system
The system is an expert system (one of the first) for the medical diagnosis and therapy of infectious diseases by means of antibiotics and was developed in the programming language LISP. On the basis of health data and further inputs from the doctor, a diagnosis is created, but the final medical decision still remains with the human, whereby the MYCIN system essentially takes on the role of a consultant (and was thus also better accepted by contemporaries).
1982 — Film “Tron”
Tron was far ahead of its time. The film already touched on topics back then that are much discussed today: artificial intelligence, digital identity, privacy, personal data and the dominance of Big Tech. Tron was also the first film to visualise the digital world — the “cyberspace” (today perhaps rather “metaverse”). Furthermore, Tron was the first film to use longer sequences of computer-generated imagery (CGI) and is thus regarded as a pioneer for many digitally generated films.
1983 — Film “War Games”
This film is about a supercomputer (WOPR) that almost triggers a nuclear war between the USA and the Soviet Union, but could be stopped through the game of Tic-Tac-Toe. Through the game, the computer learned that there can be games without a winner and, after further analyses, that “the only winning move is not to play”. The film appeared during the Cold War, but even afterwards the vision of the supercomputer that wants to eliminate the human factor as a source of error has retained lasting relevance.
1984 — Film “Terminator”
SKYNET is the next acronym of cinematic origin that stands for the eradication of humanity by artificial intelligence — a topic that reflects the attitude, or rather fear, towards artificial intelligence at the time. Unlike WOPR, SKYNET decides in favour of nuclear war, followed by Terminators, which (unlike the bombs) can strike more precisely.
1991 — Dr. SBAITSO
Dr. SBAITSO (SoundBlaster Acting Intelligent Text-to-Speech Operator) is an AI speech synthesis program from 1991 that communicates with a user with the help of digitised voices as if it were a psychologist — often with questions like “Why do you feel that way?“.
1991 — Loebner Prize
The Loebner Prize is a competition that was held for the first time in 1991 and honours the “most human-like” program. The basis for the tests is the Turing test.
1994 — “ChatterBot”
The term “ChatterBot” was coined by Michael Mauldin — a mixture of the words “Chat” and “Bot” (as an abbreviation for “Robot”). From “ChatterBot”, the term “ChatBot” was ultimately derived, which is still used today for virtual assistants.
1995 — A.L.I.C.E.
A.L.I.C.E. (Artificial Linguistic Internet Computer Entity) was developed from 1995 onwards by Richard Wallace on the basis of AIML (Artificial Intelligence Markup Language) and won the Loebner Prize (which honours the “most human-like” program) for the first time in 2000, followed by 2001 and 2004.
1997 — Deep Blue
On 11 May 1997, the IBM computer called “Deep Blue®” defeated the world chess champion Garry Kasparov. This victory is regarded as a milestone in AI research that was significant in the media and, among other things, caused hysteria and fear of soon being replaced by machines. Deep Blue was not a learning system, but used its enormous computing power to analyse up to 200 million chess positions per second.
1997 — Jabberwacky
Jabberwacky was released by Rollo Carpenter in 1997 and is a chatbot that interacts with the user via natural language in a humorous, entertaining way. The chatbot uses so-called pattern recognition to calculate suitable answers. In 2005 and 2006, Jabberwacky also won the Loebner Prize and in 2008 the bot was renamed “Cleverbot”.
2005 — Mitsuku
Mitsuku was developed from 2005 onwards by Steve Worswick and won the Loebner Prize in 2013, 2016, 2017, 2018 and 2019. It uses a supervised learning model in which the developers actively adjust the rules in such a way that Mitsuku seems more human-like.
2010 — Siri
Siri, Apple’s well-known voice assistant, was originally developed for the iPhone 4S in October 2011, although it was originally a separate app for iOS, which was released in February 2010. It is software that can recognise and process natural language and can respond accordingly to the user’s requests or commands. In addition, Siri can recognise patterns in recurring user inputs, i.e. it masters “machine learning” in order to gather “experience” and thus become better. It is not the first voice assistant of this kind, but through Siri digital assistants became better known and more suitable for the mass market — and thus represents an important milestone for AI research.
2011 — Watson and Jeopardy
In February 2011, IBM’s computer Watson, named after the IBM founder Thomas John Watson, took part in a quiz show called “Jeopardy!” and won against two human opponents — Ken Jennings and Brad Rutter. In doing so, the computer showed that it could process natural language and answer complex questions (and do so faster than its human fellow players). A (later) goal for Watson was to use the technology for areas in which time is critical — for example doctors for possible diagnoses, in order to facilitate the decision for the final diagnosis and treatment; retailers, in order to be able to give their buyers better suggestions; travellers, to have their best route calculated; and so on.
2012 — AlexNet
AlexNet is a Convolutional Neural Network (CNN) architecture that won the ImageNet Large Scale Visual Recognition Challenge (LSVRC) on 30 September 2012. AlexNet is regarded as the cornerstone of the success of deep learning in the fields of image recognition.
2013 — Film “Her”
“Her” is a film about a love story between human and AI — a completely different perspective than a few years earlier on a possible future with artificial intelligence.
2014 — GAN
GAN (“Generative Adversarial Networks”) is a machine learning system developed by Ian Goodfellow. The task of a GAN is to create its own creations on the basis of a handful of example data — mainly for image creation. The aim is that it should not be visible in the generated creations that they were not created by humans. In order to accomplish this, two neural networks are used that compete with each other: on the one hand a generator network, which produces the forgery; and on the other hand a discriminator network, which checks the received data.
2014 — Alien: Isolation
Alien: Isolation is a survival horror game by Creative Assembly. The paths and behaviours of the alien are monitored by two AIs that control the alien. On the one hand, the director AI, which keeps track of both the location of the player and of the alien at all times and regularly gives the alien hints about the location of the player (though the exact location is never handed over). The second AI, of the alien itself, consists of a so-called “behaviour tree” — a tree diagram that regulates under which conditions the alien shows which behaviour. In this way, the alien can “react” to the player’s inputs, or rather adapt its behaviour pattern accordingly.
2014 — Alexa / Amazon Echo
Alexa was introduced by Amazon in 2014 and is used today by devices such as the Amazon Echo to carry out voice commands from users with the help of the internet. Alexa can thus be used as a personal assistant, for example to play music, create to-dos, read out news or weather reports and so on. Due to the wide distribution of Alexa, one can say: AI is moving into the living room.
2016 — Tay
Tay was an artificial intelligence that was launched by Microsoft on 23 March 2016 and learned through interaction with other users via Twitter. But only 16 hours after its launch, Tay was already switched off again, because it absorbed in an unfiltered way whatever it was fed and began to spread racist and sexist statements. Tay thus represents a good example of what problems and dangers can arise with the use of artificial intelligence, unless the appropriate countermeasures are put in place.
2018 — CIMON (Crew Interactive Mobile Companion)
CIMON (Crew Interactive Mobile Companion) is the first robot that uses AI to interact with the astronauts of the ISS while floating freely. It is intended to explore whether the work of the astronauts can be made more efficient and stress can be reduced by the machine interacting with the astronauts in a positive way.
2021 — EU guidelines on AI
In 2021, the European Commission proposed the first EU legal framework for artificial intelligence (AI). The goal was and is to regulate AI with a risk-based approach, in order to create better conditions for a human-centred and ethical development and use, and to ensure that AI systems used in the EU are safe, transparent, traceable, non-discriminatory and environmentally friendly.
2022 — ChatGPT
GPT stands for “Generative Pre-Trained Transformer”. ChatGPT is a language model that uses artificial intelligence to communicate with users via text-based messages. The model is based on large amounts of data from the internet, on which it was trained, in order to be able to answer in a human-like, or rather natural-sounding, way. ChatGPT was released by the American company OpenAI in November 2022.
Originally published at SEQIS Blog