Specialty Digest

Global English editionLearners’ Study Dashboard — in development
The Knowledge Library for All

Computational Intelligence

nounArtificial Intelligencealso CI, soft computing

In one line

Computational intelligence is a branch of AI that uses nature-inspired methods such as neural networks, fuzzy logic and evolutionary computation to solve complex problems.

Computational intelligence is a branch of artificial intelligence that solves complex, poorly defined problems with methods inspired by nature, mainly neural networks, fuzzy logic and evolutionary computation. These methods tolerate noise, uncertainty and incomplete information, which makes them useful where exact mathematical solutions are impractical. The field is sometimes called soft computing, in contrast to hard computing methods that demand precise inputs and fixed logic.

History of computational intelligence

The label took shape in the 1990s. The IEEE Computational Intelligence Society, formed from earlier neural network groups, helped establish it as a recognized research area, and its large conferences brought together three communities that had developed separately. Neural network research dated to the 1940s and 1950s. Fuzzy set theory was introduced by Lotfi Zadeh in 1965. Evolutionary computation grew from work in the 1960s and 1970s by researchers such as John Holland, who developed genetic algorithms. The common thread was learning from experience and adapting to a changing environment, in contrast to the logic-based symbolic AI that dominated early artificial intelligence.

Main methods of computational intelligence

Three pillars define the field, and others are often added.

  • Neural networks: layered networks of simple units that learn patterns from data. They are the ancestors of modern deep learning.
  • Fuzzy systems: methods that reason with degrees of truth, such as fairly warm or very fast, instead of strict true and false values.
  • Evolutionary computation: search methods that evolve a population of candidate solutions through selection, crossover and mutation.
  • Swarm intelligence: algorithms inspired by the collective behavior of ants, bees or birds, such as ant colony optimization and particle swarm optimization.
  • Probabilistic methods: approaches that reason under uncertainty, such as Bayesian networks.

How computational intelligence works

Most computational intelligence methods learn by trial and adjustment instead of following a fixed program. A neural network tunes the strengths of its connections to reduce error. An evolutionary algorithm generates many candidate designs, keeps those that score best on a fitness function and combines and mutates them to produce a new generation. A fuzzy controller turns sensor readings into membership grades, applies if-then rules such as if the room is quite hot then cool strongly, and blends the results into one smooth action. The methods are often combined. A neuro-fuzzy system, for example, uses a neural network to tune the rules of a fuzzy controller.

A worked example: fuzzy temperature control

Imagine an air conditioner controlled by fuzzy logic. A conventional thermostat switches on at one exact temperature and off at another, which can cause abrupt cycling. A fuzzy controller instead treats the temperature as partly cool, partly warm and partly hot at once, each with a degree between zero and one. Rules such as if the room is warm then run at medium power combine these degrees, and the output is a smooth power level, not an on-off switch. The result is steadier comfort and less wear on the equipment. Similar ideas have been used in washing machines, camera focus systems and train control.

Applications of computational intelligence

Fuzzy controllers have been used in household appliances, train braking systems and industrial process control. Evolutionary algorithms have designed antennas, optimized delivery routes, scheduled factory production and tuned the structure of neural networks. NASA’s Space Technology 5 mission carried antennas whose shapes were found by an evolutionary algorithm. Swarm methods help route network traffic and plan logistics. Neural networks power image recognition and speech processing. Because these methods handle messy data well, they are common in engineering, finance and bioinformatics.

Computational intelligence vs. artificial intelligence

The two overlap heavily, and different authors draw the boundary differently. Artificial intelligence is the broader field and includes symbolic reasoning, knowledge representation and planning. Computational intelligence is usually presented as the subset that relies on numerical, bio-inspired learning methods. In practice, the rise of machine learning and deep learning has absorbed much of what used to be called computational intelligence, and the term is now more common in academic journals than in product marketing.

Advantages and disadvantages

The methods are flexible, robust to noise and able to find good solutions to problems with no known formula. They need little domain theory and can adapt as conditions change. On the other hand, they typically offer no guarantee of finding the best possible solution, can be slow because they evaluate many candidates, and some, particularly neural networks, are hard to interpret. Results can also vary from run to run because several methods involve randomness.

Key terms in computational intelligence

A few words recur across the field. A population is the set of candidate solutions an evolutionary algorithm keeps at one time, and each round of selection and variation produces a new generation. A fitness function scores each candidate. In fuzzy logic, a membership function gives the degree, between zero and one, to which a value belongs to a fuzzy set such as warm. In neural networks, weights are the adjustable connection strengths, and training is the process of tuning them. In swarm methods, a particle or agent is a simple unit whose movement follows local rules and the best positions found by its neighbors. Knowing these terms makes it easier to read papers and documentation across the three pillars.

Future directions

Computational intelligence is increasingly blended with mainstream machine learning. Neuroevolution uses evolutionary algorithms to design or train neural networks, and automated machine learning systems use search methods to choose model structures and settings. Fuzzy logic is being combined with deep learning to add interpretable rules to systems that would otherwise be opaque. Swarm and evolutionary methods are also applied to robotics, where teams of simple machines coordinate without central control. The central appeal remains the same: methods that adapt, tolerate uncertainty and find workable solutions when exact answers are out of reach.

FAQs

Is computational intelligence the same as soft computing? The terms are closely related and often used together. Soft computing stresses tolerance of imprecision, while computational intelligence stresses learning and adaptation.

What is a fitness function? It is a score that tells an evolutionary algorithm how good a candidate solution is, so the best ones are kept.

Does computational intelligence use biology? It borrows ideas from biology, such as neurons, evolution and swarming, but does not simulate living organisms in detail.

Key facts

  • Computational intelligence is usually described as three core pillars: neural networks, fuzzy systems and evolutionary computation.1
  • The IEEE Computational Intelligence Society organizes research and conferences in the field.2
  • Fuzzy set theory was introduced by Lotfi Zadeh in 1965.1

Related concepts

Quick checkWhat are the three core pillars of computational intelligence?Show answer

Neural networks, fuzzy systems and evolutionary computation.

Sources

  1. Wikipedia, Computational intelligence, https://en.wikipedia.org/wiki/Computational_intelligence
  2. IEEE Computational Intelligence Society, About CIS, https://cis.ieee.org/
  3. Britannica, Artificial intelligence, https://www.britannica.com/technology/artificial-intelligence

Editorially reviewed by Specialty Digest Editorial TeamLast reviewed November 17, 2026Researched and drafted with AI assistanceReport an issue