The different artificial intelligence approaches can be classified into three distinct groups: brain simulation, symbolic and sub-symbolic, and statistical. The symbolic and sub-symbolic approaches can be further classified into their own groups: cognitive simulation, logic-based intelligence and knowledge-based intelligence fall under the symbolic approach, while bottom-up and computational intelligence theories are identified as sub-symbolic artificial intelligence approaches. Years of advancement in the research and application of these theories has led to the formation of integrated approaches, combining principles from multiple schools of thought to generate more sophisticated artificial intelligence (AI) systems.
AI development first hit major developmental strides during the 1940s. By utilizing principles from neurology, cybernetics, and basic cognitive processing theories, researchers were able to build robots with primitive levels of intelligence based on brain simulation, allowing the avoidance of certain obstacles through sensory detection. Limited advancement between the 1940s and the 1960s, however, led to the abandonment of this paradigm, with researchers opting to develop other, more promising artificial intelligence approaches.
In the mid-1950s to early 1960s, AI researchers attempted to simplify human intelligence into symbol manipulation, believing that the ability of humans to learn about and adapt to objects in their environment revolves around the interpretation and reinterpretation of objects as basic symbols. A chair, for example, could be simplified into a symbol defining it as an object to sit on. This symbol could then be manipulated and projected onto other objects. Researchers were able to create a number of flexible and dynamic artificial intelligence approaches by incorporating this symbolic approach into AI development.
The ability to simulate the different cognitive approaches to symbolic thought allowed AI developers to create logic-based and knowledge-based intelligence. The logic-based approach worked on the underlying principles of logical thinking, focused almost entirely on solving problems rather than replicating human-like thinking ability. Logic was eventually balanced by "scruffy" logic, which took into account the fact that solutions can be found outside of a given logical algorithm. Knowledge-based intelligence, on the other hand, took advantage of a computer's ability to store, process, and recall vast amounts of data to provide solutions to problems.
Interest in brain simulation was revived in the 1980s after advancement in symbolic intelligence slowed. This led to the creation of sub-symbolic systems, artificial intelligence approaches that revolved around combining thinking with the more basic intelligence needed for movement and self-preservation. This allowed models to relate the environment around them to data in their memory stores. The statistical approach developed in the 1990s helped polish both the symbolic and sub-symbolic artificial intelligence approaches by using sophisticated mathematical algorithms to determine the course of action most likely to result in the machine's success. Research often tackles AI development using principles from all approaches.