History of Artificial Intelligence

 Artificial intelligence, sometimes called machine intelligence, is intelligence demonstrated by machines, unlike the natural intelligence displayed by humans and animals. Leading AI textbooks define the field as the study of "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals. Colloquially, the term "artificial intelligence" is often used to describe machines (or computers) that mimic "cognitive" functions that humans associate with the human mind, such as "learning" and "problem-solving

    So where it all started? To understand these let's get into the history of it.

•  Rossum’s Universal Robots (R.U.R.): In 1920 the Czech writer Karel Čapek published a science fiction play named Rossumovi Univerzální Roboti (Rossum’s Universal Robots), also better known as R.U.R. The play introduced the word robot. R.U.R. deals with a factory, which creates artificial people named as robots. They differentiate from today’s term of the robot. In R.U.R. robots are living creatures, who are more similar to the term of clones. The robots in R.U.R. first worked for humans, but then there comes are robot rebellion which leads to the extinction of the human race. The play is quite interesting, because of different reasons. First, it is introducing the term robot, even if represents not exactly the modern idea of robots. Next, it is also telling the story of the creation of robots, so some kind of artificial intelligence, which first seems to be a positive effect on humans, but later on the is the robot rebellion which threatens the whole human race. Artificial Intelligence in literature and movies is a big topic for its own. The example of R.U.R. should have shown the importance and influence of Artificial Intelligence on researches and society.
• Alan Turing: Alan Turing was born on 23rd June 1912 in London. He is widely known because the encrypted the code of the enigma, which were used from Nazi Germany to communicate. Alan Turing’s study also led to his theory of computation, which deals with how efficient problems can be solved. He presented his idea in the model of the Turing machine, which is today still a popular term in Computer Science. The Turing machine is an abstract machine, which can, despite the model’s simplicity, construct any algorithm’s logic. Because of discoveries in neurology, information theory, and cybernetics in the same time researches and with them Alan Turing created the idea that it is possible to build an electronic brain. Some years after the end of World War 2, Turing introduced his widely known Turing Test, which was an attempt to define machines intelligent. The idea behind the test was that are machine (e.g. a computer) is then called intelligent, if a machine (A) and a person (B) communicate through natural language and a second person (C), a so-called elevator, cannot detect which of the communicators (A or B) is the machine.
• The Dartmouth conference: In 1956 there was probably the first workshop of Artificial Intelligence and with it the field of AI research was born. A researcher from Carnegie Mellon University (CMU), Massachusetts Institute of Technology (MIT) and employee from IBM met together and founded the AI research. In the following years, they made a huge process. Nearly everybody was very optimistic.

Machines will be capable, within twenty years, of doing any work what man can do.” – Herbert A. Simon (CMU)

“Within a generation … the problem of creating ‘artificer intelligence’ will substantially be solved” – Marvin Minsky (MIT)

That was in the 1960s. The progress slowed down in the following years. Because of the failing recognizing the difficulty of the tasks promises were broken.

·       The first AI Winter: Because of the over-optimistic settings and the not occurred breakthroughs the U.S. and the British government cut off exploratory research in AI. The following years were called (first) AI Winter. The enthusiasm was lost, nobody wanted to fund AI research. The interest of publicity on Artificial Intelligence decreased. This was around 1974.

·       Expert Systems: After the first AI Winter, Artificial Intelligence came back in a form of so-called “expert systems”. Expert systems are programs that answer the question and solve problems in a specific domain. They emulate an expert in a specific branch and solve problems by rules. There are two types of engines in expert systems: First, there is the knowledge engine, which represents facts and rules about a specific topic. Second, there is the inference engine, which applies the rules and facts from the knowledge engine to new facts.

In 1981 an expert system named SID (Synthesis of Integral Design) designed 93% of the VAX 9000 CPU logic gates. The SID system was existing out of 1,000 hand-written-rules. The final design of the CPU took 3 hours to calculate and outperformed in many ways the human experts. As an example, the SID produced a faster 64-bit adder than the manually designed one. Also, the bug per gate rate, which was around 1 bug per 200 gates from human experts, was much lower at around 1 bug per 20,000 gates at the final result of the SID system.

·       The second AI Winter: The second AI Winter came in the later 80s and early 90s after a series of financial setbacks. The fall of expert systems and hardware companies who suffered through desktop computers built by Apple and IBM led again to decreasing AI interest, on the one hand, aside from publicity and on the other aside from investors.

·       Deep Blue: After many ups and downs Deep Blue became the first chess computer to beat a world chess champion, Garry Kasparov. On 11 May 1997 IBM’s chess computer defeated Garry Kasparov after six games with 3½–2½.

Deep Blue used tree search to calculate up to a maximum of 20 possible moves. It evaluated positions by a value function mainly written by hand, which was later optimized by analyzing thousands of games. Deep Blue also contained an opening and endgame library of many grandmaster games. In 1997 Deep Blue was the 259th most powerful supercomputer with 11.38 GFLOPS. In comprising: The most powerful supercomputer in 1997 had 1,068 GFLOPS and today (December 2017) the most powerful supercomputer has 93,015 GFLOPS. FLOPS stand for floating-point operations per second and the ‘G’ in GFLOPS stands for Giga. So the equivalent of 1 GFLOPS is 10⁹ FLOPS.

·       21st Century: Deep learning, Big Data, and Artificial General Intelligence: In the last two decades, Artificial intelligence grows heavily. The AI market (hardware and software) has reached $8 billion in 2017 and the research firm IDC (International Data Corporation) predicts that the market will be $47 billion by 2020. This all is possible through big data, faster computers, and advancements in machine learning techniques in the last years. With the usage of Neural Networks complicated tasks like video processing, text analysis, and speech recognition can be tackled now and the solutions which are already existing will become better in the next years.

·       Atari Games: In 2013 DeepMind, one of the world’s foremost AI research, introduced an AI which could play a couple of Atari games on top of a level of human players. This first seems not very expressive, but they just used reinforcement learning and neural networks to let the AI self-learn these games. Also, they just used the pixels as an input to the agent, so there was no direct reward score given to the agent depending on the moves he did. In 2015 they further introduced a smarter agent, who successfully played 49 classic Atari games by itself.

Next to classic games from old retro consoles DeepMind is developing an AI for more complex games, like e.g. StarCraft 2. StarCraft 2 is a Real-Time Strategy (RTS) game, which is the most popular 1 vs.1 E-Sport title. StarCraft 2 is very popular in South Korea and the best StarCraft 2 pro player comes from South Korea. Nevertheless, there are many European and North American pro player who plays for a living. StarCraft 2 is a much more complex game than classic video games: There are many more possible actions you can do, you do not know everything about your opponent and you have to scout him to explore what he is doing. In StarCraft 2 there are also dozens of strategic decisions to choose from every minute and in general much more to care about comparing to classic video games. The current AI is not very good at the moment and it only can play mini-games like building units. About the StarCraft 2 AI I am very excited about because I am a big StarCraft 2 fan and I am exciting about how the AI will change the StarCraft 2 Metagame and what new tactics it will explore.

·       AlphaGo: Next to classic Atari games, DeepMind also managed to defeat the world's best human Go player with his AI AlphaGo. In October 2015 they first defeated the European Go champion Fan Hui five to zero. After the match, there was a lot of skeptics in the Go scene about AlphaGo, because Fan Hui is ‘only’ a 2-dan (out of 9-dan, which is best) European Champion. Therefore the DeepMind team flew to South Korean to face Lee Sedol, a 9-dan Go Player. Lee Sedol is known as one of the best Go players in the world. After DeepMind managed to win the first 3 matches Lee Sedol seemed very desperate. But in the fourth game, AlphaGo lost after it made an obvious mistake. In the last match AlphaGo could win again. In the end, AlphaGo managed to win with 4-1 against Lee Sedol. If you are more interested in the story about AlphaGo I recommend the movie about it. In my opinion, the movie is great and shows, next to the technical impact of the AI, the impact on the Go community. In 2017 DeepMind published the next generation of AlphaGo. AlphaGo Zero is built upon reduced hardware and just learned Go to play against itself. After three days of training, AlphaGo Zero was stronger than the version of AlphaGo who defeated Lee Sedol and won against his younger version with 100-0. After 40 days of training, it also defeated his former version of AlphaGo Zero.

There are already Artificial Intelligence systems who can outperform humans in specific areas, like e.g. playing GO or data analysis. Today, if we talk about Artificial Intelligence systems in production we refer to specialists. But there is no Artificial General Intelligence (AGI) yet, who can perform like a human, and neither there is a superintelligence, who is smarter than a human being.

In the last few decades, we have seen great hype about Artificial Intelligence among students. But AI is the big thing and (as we learned from our Subaru) it's not just hype, it's real. Well, maybe there is still a bit of hype. But we're beginning to see real function come out of machine learning models and natural language processing and it isn't just marketing fluff.