Model of human intelligence/Component model

Traditional AI research focuses either on narrow intelligence (e.g. face detection or voice recognition) or general intelligence (e.g. hill climbing algorithm or nearest neighbor classification). Both these approaches have wide range of applications, but none can describe human intelligence. This is because human intelligence is not a single algorithm. In order to fully define human intelligence, one must perceive it as a collection of cooperating components. None of these components is dominant. In other words, there is no "essence" of human intelligence.

Component nature of human intelligence is visible in several ways. Brain imaging shows that certain activities tend to induce activity in specific parts of the brain. People with brain lesions often lose only very specific mental abilities. Newborns show abilities (e.g. object and face recognition, sensitivity to tone of voice) that they didn't have time to learn on their own. Certain very specific behavioral traits are constant among cultures.

Open issues

• Links.
• Feedback loop. Mutual support between components. No unidirectional layering.
• Note that there are a few big general-purpose components which make major contributions to intelligence. This needs to be mentioned to avoid creating illusion that human intelligence is just a collection of small narrow intelligences.
• Additive/suppressive signal combination. Local layering of activating/suppressing signals to achieve abstraction.
• How many components are there?