This paper provides a broad introduction to artificial intelligence (AI), tracing its origins from early theoretical work by McCulloch, Pitts, and Turing through the development of expert systems and neural networks. It defines intelligence and its machine counterpart, outlines the key capabilities researchers seek to build into AI systems—including reasoning, knowledge representation, machine learning, and natural language processing—and contrasts the bottom-up and top-down development approaches. The paper then situates AI within a near-future American context, exploring anticipated applications in healthcare, autonomous vehicles, fraud prevention, manufacturing robotics, weather prediction, and space exploration, arguing that AI is poised to become the United States' next major technological frontier.
If we look at definitions of intelligence, there are probably infinite ones. Put most precisely, however, intelligence is one's ability to acquire and gain knowledge — the capacity to learn from one's environment and, in turn, contribute back to it. When we speak of "the ability to learn," we are using the term in very broad terms. It encompasses characteristics such as the ability to adapt to new or changing environments and the ability to reason, think, comprehend, judge, and evaluate situations. Intelligence determines how one achieves goals in the world. The degree of intelligence in people, animals, and machines differs significantly (Hawkins and Blakeslee).
Artificial intelligence is intelligence dedicated solely to machines. It is a multidisciplinary field involving psychology, cognitive science, neuroscience, and, most importantly, computer science. It is typically regarded as a branch of computer science concerned with the study and design of intelligent agents — systems that analyze and monitor their environment and then take actions to maximize their chances of success. We can define artificial intelligence as the ability of any hardware or software to perform tasks that human beings would characterize as intelligent. These activities include:
Searching: finding relevant information from a large database after being given very few directives.
Surmounting constraints: finding techniques to fit things into limited spaces and building upon complex scenarios and objects.
Recognizing patterns: identifying items with similar characteristics or spotting entities that do not match a given profile.
Making logical inferences: drawing conclusions based on what is known, using reasoning methods such as deduction and induction.
The word artificial intelligence comes from technology, a term that itself branches from the Greek word techno, meaning art and skill. When any technology is called sophisticated, it implies a combined structure of refined and learned skills and processes. In artificial intelligence, these processes have manifested in a number of ways, including neural networks, expert systems, and automatic speech recognition (Hawkins and Blakeslee).
The history of artificial intelligence began in earnest when Warren McCulloch and Walter Pitts, in 1943, proposed a model of artificial neurons. Each neuron was characterized as being either on or off, turning on when a significant number of neighboring neurons were stimulated. Evidence of interest in artificial intelligence can be traced as far back as ancient Egypt, but the technology only became practically available with the development of the electronic computer in 1941. Computers were the necessary platform for making artificial intelligence viable, yet it was not until the 1950s that a meaningful link between computers and human intelligence could be established.
Norbert Wiener was among the first to observe how human intelligence might be imitated by machines. He grounded his observations in the principles of feedback theory, using the thermostat as an illustrative example. A thermostat controls environmental temperature by measuring the actual temperature, comparing it to a desired level, and responding by turning heat up or down. What made Wiener's research significant was that he translated intelligent behavior into a formal, describable process — one that directly influenced the early development of AI (Hawkins and Blakeslee).
Mechanical reasoning had long fascinated philosophers and mathematicians, and their work in logic led directly to the invention of digital electronic computers. Building on the work of mathematician Alan Turing, who suggested that the electronic computer could shuffle simple symbols — 0s and 1s — to reproduce any conceivable mathematical computation, combined with discoveries in neurology, information theory, and cybernetics, researchers paved the way for what was then called an "electronic brain" (Scheier and Pfeifer).
Formal research in artificial intelligence began in 1956. Early leaders and their students wrote papers demonstrating computers solving algebraic word problems and understanding spoken English. Those founders were enthusiastic about AI's future and foresaw machines eventually matching or surpassing many human capabilities. Research slowed considerably in 1974 and remained limited until the early 1980s, when expert systems — AI programs capable of replicating the knowledge and analytical skills of human experts — achieved notable success. By the mid-1980s, the artificial intelligence market had reached enormous proportions (Scheier and Pfeifer).
When developing an artificial intelligence system, a central question is whether a machine can truly act intelligently. Researchers have pursued several key capabilities they wish to embed in intelligent machines.
Reasoning and problem solving: Early AI researchers developed algorithms that imitated human reasoning for solving problems and making logical deductions. By the 1990s, such algorithms could handle incomplete information and concepts from probability and economics. As problems grow more complex, however, the algorithms grow correspondingly complex, and the memory and processing time required can become prohibitive. The search for more efficient algorithms remains ongoing (Scheier and Pfeifer).
Knowledge representation: Many problems that machines must solve require thorough knowledge of the world. One of the hardest aspects of this challenge is default reasoning — knowing what assumptions to make in the absence of complete information — and the qualification problem, which concerns the difficulty of listing all preconditions for an action to succeed.
Planning: An intelligent machine must be able to set particular targets and subsequently achieve them, just as the human mind plans ahead (Jones).
Machine learning: The machine must find distinct patterns in a flow of inputs, classify information into the correct categories, and identify relationships between inputs and outputs. Reinforcement learning is also central: an agent is rewarded for correct behavior and penalized for incorrect behavior, guiding it toward better performance over time.
Natural language processing: Giving machines the ability to read and understand human language is a critical dimension of AI. A powerful processing system enables a natural-language user interface that can acquire knowledge from human-written sources. One approach to extracting meaning from natural language is semantic indexing in computers (Jones).
Computer vision and social intelligence: Machines must also interpret input from sensors — the ability to analyze visual input is called computer vision. Beyond this, a machine should be capable of reading human emotions and adapting its behavior accordingly, as well as exhibiting a degree of creativity. When all these capabilities are unified in a single system, researchers hope to achieve artificial general intelligence — a machine whose skills meet or exceed human abilities across most or all domains.
"Neural networks versus expert systems approaches compared"
"AI in traffic, finance, smartphones, and more"
"AI's future role in health, manufacturing, and society"
Artificial intelligence has reached a point at which it can provide human beings with invaluable assistance and speed up tasks still performed manually. It has enabled the automation of numerous systems, and the most successful examples are those that people use without even being aware of it.
You’re 50% through this paper. Sign up to read the remaining 3 sections.
Sign Up Now — Instant Access Already a member? Log inAlways verify citation format against your institution’s current style guide requirements.