The history of AI is primarily about technical innovations. However, an in-depth understanding of this history also needs the consideration of many financial, political, and cultural factors.

As marked in Fig. 2.2, the fertilization of AI started in myths, stories, and rumors in which artificial beings were endowed with consciousness or intelligence by master craftsmen. Next, the implantation occurred with further developments in science fiction, e.g., during the Golden Age of science fiction between 1938 and 1946 [8]. Then, AI's fast maturation into a fetus was driven and marked by breakthroughs like philosophers' effort at describing the process of human thinking and materialized as mechanical manipulation of symbols was realized in the invention of programmable digital computers in the 1940s.

During this time, the confluence of several closely related ideas from different areas provided theoretical support for constructing the electronic brain, which cast the foundation for modern AI. These include research in neurology that showed the brain is a network of neurons fired in all-or-nothing electrical pulses. In particular, Walter Pitts and Warren McCulloch reported the use of networks of idealized artificial neurons for performing simple logical functions in 1943 [9], which opened the door to artificial neural networks as well as the ups and downs of connectionism in the later AI history. In 1950, Alan Turing made the first serious proposal in the philosophy of AI by presenting his famous Turing Test, in which a machine is thought to be "thinking" if it could conduct a conversation that was indistinguishable from a conversation with a human being [10]. This study allowed Turing to convincingly argue that a "thinking machine" was plausible and answered the most common objections to the proposition.

Other breakthroughs included the first stored program computer in 1948, i.e., the Ferranti Mark 1 machine [11], the use of this machine to write a checkers program, and Arthur Samuel's checkers program developed in the mid 1950s and early 1960s, which gained skills comparable to respectable amateur players [12]. When access to digital computers became possible in the mid 1950s, a few scientists identified a new approach to creating thinking machines: a machine that manipulates numbers could also manipulate symbols, and such a manipulation of symbols could be the essence of human thoughts. As a result, the first AI program, i.e., Logic Theorists [13], was created in 1955.

Between 1952 and 1956, the advent of computers inspired a handful of scientists to seriously discuss the possibility of building an electronic brain. Usually, it is believed that the milestone where AI becomes a field of study (or an academic discipline) was marked by a workshop held on the campus of Dartmouth College, USA in 1956 [14]. Most event attendees later led AI research with millions of dollars of financial support, and many of them predicted machines as intelligent as human beings would be created in no more than one generation.

Figure 2.2: Historical development of AI

After the Dartmouth Workshop, fast developments in AI programs, especially symbolic AI, were achieved for applications such as solving algebra problems, proving theorems in geometry, and learning to speak English. There were many efforts at maze-alike games in the paradigm of "reasoning as search". Attempts at enabling computers to communicate in natural languages yielded programs like "Student", AI programs written using a semantic net (conceptual dependency theory), and chatterbots (later clipped to chatbots) like ELIZA [15, 16]. The MIT AI Laboratory proposed to focus on artificially simple situations known as micro-worlds based on a perception that, in successful sciences like physics, basic principles were often best understood using simplified models like frictionless planes or perfectly rigid bodies. In Japan, the world's first full-scale "intelligent" humanoid robot or android was created via the WABOT (WAseda roBOT 1) project.

Such successes led to over-optimism among the first generation of AI researchers as well as some funding agencies. Many researchers believed that, within ten to twenty years, the problem of creating AI and generating machines that can handle most of human beings' work would become possible. Meanwhile, the optimism prompted major research funding, such as that from the Advanced Research Projects Agency in the U.S. (later known as DARPA).

In the 1970s, AI started receiving major critiques and financial setbacks. In particular, AI researchers failed to appreciate the difficulty of the AI problems they faced. Meanwhile, the over-optimism had raised expectations impossibly high, and when the promised results failed to be delivered, funding for AI was withdrawn. During this time, the field of connectionism (or neural nets) was shut down almost completely for 10 years partially due to Marvin Minsky's devastating criticism of perceptrons [17]. Despite the recession and criticisms against AI in the late 1970s, new ideas were explored in logic programming, commonsense reasoning, and many other areas.

In the 1980s, a form of AI program called "expert systems" gained popularity in the industry worldwide [18], and knowledge became the focus of mainstream AI research. In the same period, the government of Japan aggressively funded AI through its fifth-generation computer project [19]. Another remarkable advance in the early 1980s was the revival of connectionism represented by the work of John Hopfield and David Rumelhart [20]. Once again, AI gained success in a variety of ways during this relatively short blooming period.

The rise and drop of AI in the 1980s, especially the involvement of industries and governments, exhibited a clear correlation with the economy. The AI collapse in the later 1980s was partially due to the failure of commercial vendors to
develop a wide variety of workable solutions and the burst of the economic bubble during that time. The AI technology was deemed not viable as the public was also discouraged by the failures of dozens of companies. Despite the setbacks and pessimism, the AI field continued to advance in multiple ways. For example, many researchers, including robotics developers Rodney Brooks and Hans Moravec [21], advocated approaching AI in other ways.

After more than half a century of development, the field of AI finally achieved some of its oldest goals. In particular, widespread use of many AI techniques in industries finally became realistic. Some of the successes were due to the increasing computer power, while some others were achieved by focusing on specific problems and pursuing them with the highest standards of scientific accountability. Despite the progress, the reputation of AI, at least in the business world, was less than pristine. Within the field, there was little agreement on the reasons why AI failed to fulfill the dream of human-level intelligence in the 1960s. All these factors led to the evolvement of AI into competing subfields focused on particular problems or approaches, sometimes even under new names to blur, rebrand, or dissociate their AI tattoo. This is a time when AI became both more cautious and more successful than it had ever been.

The third wave marked the explosive development of deep learning, which may be attributed to three factors: improvements in deep learning architectures, especially for addressing the vanishing gradient issue, increases in computational power represented by the use of GPU, and growth of data in a "big data" era especially image data. Important milestones in these three aspects include the publication of the Deep Belief Networks [22], the advocacy for the use of GPUs for training Deep Neural Networks [23], the launch of the ImageNet database ( 14 million labeled images) that led to later ImageNet annual competitions (ILSVRC) [24], and the development of ReLU and other techniques for the vanishing gradient problem [25]. Then in 2012, AlexNet's victory in the ImageNet competition triggered a new deep learning bloom globally and attracted industry giants' attention [26]. Deep learning started gaining more momentum and making impacts in or even sweeping many disciplines such as computer vision and natural language processing. Advancements in algorithms such as Generative Adversarial Network (GAN) [27] and the continuous development of deep learning architectures for various purposes, platforms, and tasks have been continuing. This led to the dramatic growth of the market for AI-related products, which was called an AI "freeze" by the New York Times. During this time, the rise of reinforcement learning, especially its integration with deep learning, also generated astonishing breakthroughs, and further developments along this direction gained more momentum and popularity. This freeze continues as large language models and generative AI impact and reshape many business areas [28].