How Artificial Intelligence Became the Ultimate Productivity Partner

For centuries, people have been developing various tools that could facilitate human thinking. The abacus, for instance, was created to reduce the possibility of errors when performing arithmetic operations. In addition, in the 17th century, Blaise Pascal and Gottfried Wilhelm Leibniz improved the design of the abacus and created a geared machine that could perform additions and subtractions and, eventually, multiplications. However, the most unexpected source of inspiration came from the textile industry, when in the early 19th century Joseph Jacquard developed a loom that could weave complex patterns by reading punched cards. The latter invention could be considered an early example of a stored program.

It was Charles Babbage who further expanded upon the idea, though none of his projects, namely the Difference Engine and the Analytical Engine, were completed at the time of his death. Nevertheless, his designs incorporated some of the most critical elements of modern-day computers, including memory for storing results. Yet, the most notable contribution to the field was made by someone who was not a scientist or an engineer by profession – Ada Lovelace, the first ever computer programmer. In 1843, she translated a French article about Babbage’s Analytical Engine and supplemented her notes, which were nearly three times longer than the article itself. In particular, Lovelace’s Notes provided the first-ever description of a computer algorithm, which was meant for calculating Bernoulli numbers and could, therefore, be considered the first computer program. She also predicted that such a machine would have applications beyond numerical calculations, thus foreshadowing the future potential of computers almost a century ahead of her time. Moreover, Ada noted the similarities between the Jacquard’s invention and Babbage’s Analytical Engine, specifically that both were capable of producing intricate patterns via sequences of coded instructions.

It took almost a century for the concept of computer programming to evolve and lead to the creation of modern-day machines. Early 20th-century computers utilized relays, which were large and unreliable electric switches, as their essential elements. However, von Neumann and Turing soon realized that a substantial leap in performance could be achieved by storing programs in memory, thus allowing computers to be controlled by their own set of instructions. These ideas, along with the development of transistors and integrated circuits, eventually led to the creation of modern-day computers that utilized silicon wafers for etching electronic circuits. Personal computers followed soon after and, eventually, they became small enough to fit in the palm of one’s hand. Such miniaturization of technological devices allowed them to become powerful, flexible, and accessible to ordinary people, which led to their unprecedented proliferation.

The similarities between computers and neural networks have prompted the comparison between the two, as they are both effectively information processors in spite of the differences in their design. More specifically, external stimuli received by the sensory organs are processed by the human brain and converted into actions by the body, whereas the information received by peripheral devices is processed by the computer’s CPU and stored in its memory. However, for the majority of the last century, computers and their close relatives, the programmable logic controllers, have been rather rigid in their operations, as they had to be explicitly told how to perform a certain task. At the same time, human thinking is more flexible, as it allows people to acquire new knowledge and apply it in unforeseen circumstances. In particular, when a specific algorithm for processing information is absent or too unreliable to be used, humans can guess an approximate answer, as opposed to computers, which require explicit instructions. As a result, computers are rather slow at tasks that require flexibility, and an attempt to tackle them with conventional programming tools results in either incorrect results or excessive processing time. However, the development of artificial intelligence and machine learning algorithms made it possible to equip computers with some of the flexible aspects of human cognition.

The first AI applications that were actually useful were expert systems, examples of which include programs capable of recognizing patterns in data and making decisions according to these patterns. The first expert system, called DENDRAL, was created in Stanford in 1965 and was designed to read mass-spectrometry data in order to determine the molecular composition of organic compounds. It was quickly followed by another Stanford-created expert system that was used for medical purposes and was called MYCIN. The latter project began in the early 1970s and was designed to aid physicians in diagnosing bacterial blood infections such as meningitis. It could also recommend courses of treatment and doses of medication and was later approved by the FDA. Its effectiveness was comparable to that of human experts, as it was able to diagnose ailments correctly in approximately 90% of cases. In addition, it could also explain the logic of its recommendations and decisions, which significantly contributed to its credibility, albeit it was never actually used for that purpose. Thus, the main challenge of AI development was to encourage medical professionals to adopt such technologies, which was not achieved at the time, most likely because of ethical concerns that were raised.

The turning point in the development of AI came when it became capable of learning from data by using training sets. The most prominent example of such machine learning algorithms is deep learning, which uses artificial neural networks arranged in multiple layers. When shown images, the first few layers of such computers detect simple patterns, such as lines and corners, and more complex patterns are identified at deeper layers. There are three major reasons why such computer programs could be trained when such programs were not available a few decades ago. First, there was a sufficient amount of data that could be used to train such algorithms. Second, graphics cards that were designed for video processing became essential for AI development, for they enabled the necessary computational power. Third, new methods in machine learning allowed these algorithms to be trained.

The most notable milestone in the history of AI, which demonstrated its capabilities to the general public, occurred in February 2011, when IBM’s supercomputer Watson competed in the Jeopardy! game show and defeated its two best human opponents, Ken Jennings and Brad Rutter. In particular, contestants were required to formulate questions to particular answers. However, Watson’s work was far from being as simple as it appeared, as it had to analyze various levels of ambiguity in the answers. Moreover, Watson had to evaluate the confidence level in its answers, which took just a few seconds. Similar AI algorithms were later used in medicine in order to identify relevant studies on potential treatment options. Besides such methods that are beneficial in particular fields, AI also became able to assist in more creative ways. In particular, computers were able to suggest unusual food pairings and new recipes.

Today, most people use various AI-powered gadgets and applications on a daily basis. For example, recommendation systems suggest new videos to watch or products to purchase, whereas other programs, such as spam filters and GPS navigation devices, function with limited AI capabilities. Some AI methods are so advanced that they can process unlabeled data, such as images of various objects, and group them together according to certain similarities. Moreover, some computers can actually teach themselves how to play a game by analyzing the screen and attempting to maximize their scores. However, in most cases where creativity and extensive information processing are required, computer programs function as assistants that help humans, rather than replace them. For instance, AI-generated articles, music, or paintings can be used as a starting point for actual creative works by people, which is especially useful when the existing information is overwhelming.

Even now, these systems remain specialists. Software that reads a medical scan doesn't also write poetry or negotiate a contract, and machines that match human breadth across every domain remain a long way off. What's changed, and what keeps changing, is how naturally people can interact with these tools, through voice, text, and images, and how much of the unglamorous, information-heavy work they can absorb. The more durable skill isn't using any particular tool. It's the judgment to ask a good question, check the answer, and know where a machine's output ends and a person's own judgment needs to begin.