Artificial intelligence (AI) is the ability of a computer system to process information and perform tasks that would otherwise require human intelligence. As systems perform these tasks, they may appear to exhibit humanlike behavior. AI is also a branch of computer science that studies how to design systems with such capabilities. A computer with AI could perform such tasks as learning, understanding language, working toward goals, and perceiving the world through vision, hearing, and other senses.
Artificial intelligence as a scientific field began in 1956, when the term was coined at a workshop at Dartmouth College. Dartmouth professor John McCarthy and his colleagues organized the workshop. At the meeting, scientists presented the first computer programs capable of logical reasoning, learning, and playing board games. To develop the capabilities of AI, computer scientists use a combination of knowledge representation, reasoning, and machine learning.
How AI systems process information
Knowledge representation and reasoning
are two core challenges in AI. They can be handled in various ways. One way is using a logic-based approach, where precise, logical rules represent knowledge. Such rules might take the form, “If x is true and y is true, then z is true,” where x, y, and z represent various statements. In this approach, reasoning involves computing the consequences implied by such rules. The logic-based approach has led to developing expert systems, software systems that solve problems in narrow fields of expertise with well-established processes. For example, scientists have trained expert systems to identify different types of cancers.
A second approach is a probability-based approach—representing knowledge using numerical probabilities. In this approach, reasoning involves computing the probability of various conclusions, given specific evidence.
In a third approach, called the neural network approach, knowledge is represented as a network of interconnected units that perform certain tasks by exchanging information. This approach mimics the behavior of neurons (nerve cells) in the brain.
Machine learning
involves computer programs that learn from examples and from experience. The 1956 Dartmouth workshop presented the first program that learned to play checkers by competing against a copy of itself. Other programs have learned to recognize human speech and handwriting. In 2011, a machine-learning system called Watson, developed by International Business Machines Corporation (IBM), demonstrated an advanced ability to understand human language. It did so by defeating two human champions on “Jeopardy!,” a question-and-answer quiz show.
How artificial intelligence is used
Artificial intelligence has several major areas of application. They include (1) planning and problem solving, (2) speech and natural language processing, (3) computer vision, and (4) robotics.
Planning and problem solving
require programs to identify sequences of actions that accomplish specified goals. Such programs often search for many alternative plans guided by heuristics, rules that direct the search toward promising solutions. A common type of problem solving is required for playing board games. AI programs can play many such games at the level of the best human players. In 1997, Deep Blue, a computer developed by IBM, won a chess match against the reigning world chess champion, Garry Kasparov. Programmers developed an unbeatable AI checkers program in 2007.
In 2015, a Google-developed computer program called AlphaGo defeated the Chinese-born French Go champion Fan Hui. Go, a board game played in Asia for thousands of years, is known for its complexity due to the large number of plays possible during each turn. AlphaGo’s surprising success was achieved through a type of machine learning called deep learning. Deep learning uses multiple interconnected “layers” of representations. The program learns complicated concepts by developing them from layers of simpler concepts. In this way, deep learning programs can build complex knowledge through experience.
Speech and natural language processing
involves developing computer programs that communicate in a human language, such as English, instead of in a specialized programming language. Scientists have developed logical, probabilistic, and neural network systems for processing natural language. Simple chatbots can successfully carry on conversations about a narrow topic—for example, making an airline reservation. Advanced chatbots can discuss a wide variety of topics with a human user. Some advanced systems can translate web pages from one human language to another. Some generative systems can even generate longer, more complex texts, such as short stories and news articles, or produce computer code according to specific instructions. The most advanced systems can generate text that is difficult to distinguish from text written by a human. But such systems cannot always determine whether the statements they generate are true.
Computer vision
allows computers to recognize patterns and objects in images. Vision systems first extract image features, such as edges and textures, then use rules or neural networks to identify objects in the scene. The complexity of developing computer vision is often underestimated because humans appear to recognize patterns and objects relatively easily. Like natural language processing, computer vision has been transformed by generative models. Such models can use computer vision techniques to produce realistic images of faces and other objects that are difficult for humans to determine as fake.
Robotics
studies the control of mechanical robots. Robots must control their motion to manipulate objects and avoid obstacles. AI enables robots to perform complicated tasks by combining motion planning, interpretation of visual and sound information, and generation of artificial speech. Robots with advanced AI can operate without human supervision.
