[ CogSci Summaries home | UP | email ]

Simon H.A. (1995). Explaining the Ineffable: AI on the Topics of Intuition, Insight and Inspiration. Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, 1, 939-948.

 author =       {Herbert A. Simon},
 title =        {Explaining the Ineffable: AI on the topics of
 intuition, insight, and inspiration},
 year =                  {1995},
 booktitle = "Proceedings of the Fourteenth International Joint
 Conference on Artificial Intelligence" 
volume =    {1},
 pages =     {939--948},

Author of the summary: Bragoszewska A., 2006 , 3ab5@qlink.queensu.ca

Cite this paper for:

By carefully considering the cognitive phenomena to be tested, it is possible to agree on key defining characteristics. When the phenomenon can be recognized accurately, it can be tested and studied with equally accurate experiments. With this strategy, insight, intuition, and inspiration were classified,and based on their classification it was shown that there exist AI programs that display these phenomena. In turn, modeling these phenomena resulted in a deepened understanding of the phenomena themselves, and therefore of the human mind.



  1. solution is reached rapidly after problem is presented
  2. problem solver is unable to give clear account of the steps taken to arrive at the solution

STUDIED IN:[p.942]

EPAM program, designed originally to stimulate human rote verbal learning. Consists of tree-like discrimination net. Has short term memory where it retains a "few" new symbols for no longer than 2 seconds, unless it has time to "rehearse them". Nets grow in response to presented stimuli. EPAM learns appropriate discriminations by experience, using the feedback ("right"/"wrong") offered to its responses. EPAM recognizes patterns once learned.


  1. EPAM stimulated with a collection of medical symptoms(something it has dealt with before), recognizes this pattern and can very quickly access the information it contains about the disease. Result of recognition (name of disease) remains in short-term memory, will be reported by EPAM.
  2. The program is however not designed to report the results of each of the individual tests it performed on the input data. An account of how the solution is found is not available.
OTHER NOTES: [p.943]

Experiments designed to test intuition should be carried out on subject matter that allows easy verification of the fact / correctness, to avoid "false intuition". Intuition is actually recognition.



  1. (1) period of unsuccessful work and frustration preceeds the occurence of insight
  2. (2) the sudden idea that follows (1) is not always the solution, sometimes it is "the conviction of its imminence"
  3. (3) new understanding / feeling for the problem , new way of representing it occurs sometimes:
  4. (4) period of "incubation" preceeds occurence of insight, during this time the problem is not consciously attended to
STUDIED IN: [p.944]

A program that combines capabilities of EPAM and the General Problem Solver, found capable of expressing the following integral properties, which support the four characteristics of insight.

Assume the program is in mid-search:

  1. it is capable of conducting a selective search for a problem solution, which is unsuccessful for the first while
  2. it holds information about the problem and knowledge about how to attack it in long-term memory
  3. at the moment, is following a path that will not lead to a solution and is not aware of this
  4. it is assumed the search is serial, and the flow of control in the program controls direction of the search, via attentional mechanisms
  5. information about the local situation is held in short-term memory, it is continuously changing
  6. simultaneously, the program is noticing, learning and storing in long-tem memory the more permanent features of the problem -- change, improvement in the information available for solving the problem
  7. interrupt mechanism programmed into the control structure -- will pause search after some period of unsuccessful efforts, and change attention to a part of search whish pertains to a different representation of the problem or a different search control structure
  8. when interruption occurs, control information held in short-term memory is purged -- if search is restarted later, this information will not be available and the search might explore new direction
  9. this new search direction is likely to be more productive than the previous one, since information stored in long-term memory will help determine the new search direction

The model of insight summarised with the above nine properties is supported by the research of Craig Kaplan and H.A. Simon (Kaplan and Simon, 1990). Human subjects were asked to solve a puzzle (cover chessboard with specific number of dominoes). The process of solving the problem, and specifically the phenomenon of insight was studied. The observations serve to support the nine properties- model of insight.

OTHER NOTES: [p.944]

The nine integral properties of the EPAM-GPS - like program were identified by evidence obtained while many different kinds of tasks were being performed.

[p.945] It was shown in both the human and computer program studies, that the key mechanisms for finding a solution were the 'emptying' of the short term memory , and allowing for a new representation of the original problem.



The production of novelty, using "combinatorics".

STUDIED IN:[p.946]

BACON program, designed to take uninterpreted numerical data as input and outputting scientific laws that fit the data. Not always successful. Starts with simple functions, uses "combinatorial means" to create more complex ones. Data guides BACON as to which function to try next -- program tests ratios of the variables to interpret data. If the problem involves data concerning more than two variables, BACON changes one independent variable at a time to find conditional dependencies among small sets of variables. It then uses this information to explore the effects of altering other variables. BACON's ability to discover new concepts is attributed to its final heuristic: "When [BACON] discovers that there is an invariant relation in the interaction between two or more elements in a situatuion, it assigns a new property to the elements, measuring its magnitude by the relative strength of each element's action (one of the elements is assigned a unit value, becoming the standard.)"


When provided with data on the periods of revolution and distances from the sun for different planets, BACON first arrived at Kepler's erroneous square law but it quickly rejects it as an insufficient fit to the data, and goes on to find the correct law.

OTHER NOTES: [p.946]

In searching for the relationship between a planet's period of revolution and its distance from the sun, Kepler initially made a mistake in his evaluation of the data; years later he found the correct law.

Summary author's notes:
Ineffable (from Oxford Concise): "too great for description in words"
H.A. Simon is careful not to suggest fantastic implications of this research, instead he points out that these studies should serve to show how AI can serve to further other fields of study such as human psychology and cognition, and that they can inturn serve to further it.
Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies (jim@jimdavies.org)