main | about us | download | faq | readings | blogs | support | tutorial | *

As the world’s problems become too complex to solve, we give in to fantasies and then chaos reigns, can we do something about this?Donald Steward10/2/2011


Hyper-connectivity has made the problems of our society more complex while our abilities to deal with complexity have not matured commensurately. This has led people to try to solve these problems with oversimplifications that have created fantasies rather than solutions. Without the necessary reasoning capabilities, these fantasies have been difficult to unmask. They have led to inappropriate actions that are causing chaos to reign in many regimes. A method is presented here that uses a computer program to help people solve a large and important class of problems that involve cause-and-effects that may otherwise be too complex to be within human reasoning capabilities.


As communication technologies have vastly improved, our world has become hyper-connected. Solving our social problems now must satisfy more constraints, making them too complex for humans to understand and solve. Our abilities to manage complexity have not risen commensurately. So in the absence of reason, we often try to solve our problems using oversimplifications. This has led to failure to find valid solutions, causing frustration, fallacious thinking, and the growing rage we are seeing around the world.

People say whatever they believe will have the desired effect on their listener. It may or may not be true; it may or may not be what they believe; and it may be manipulative. It may be based on what the speaker thinks the listener believes. People herd with groups of people who share some of their beliefs and tend to accept the other beliefs held by that group. This can result in society being very unstable. It can and does contribute to chaos. What is lacking is the ability of everyday people to reason about these very complex situations. Although they may easily describe the dots that make up the problem, if there are too many dots, they cannot put the dots together logically.


Here we will consider a class of problems where the situation can be described by cause-and-effect relations and the problem is to use that knowledge of the situation to explain the causes of behaviors that can arise from that situation. As we will see, this covers a very broad class of problems that can now be solved by the method presented here.

An example of a fact is that ‘Joblessness is high’. An example of the cause-and-effect relations to describe a situation is: ‘Joblessness’ is caused by ‘Businesses unable to sell their products and services’, which is caused by ‘Customers don’t have the money to buy their goods and services’. Sometimes these cause-and-effect statements can form circuits such as if we add that ‘Customers don’t have the money to buy their goods and services’ is caused by ‘Joblessness’. Such circuits can be vicious or virtuous cycles. To prevent what the cycle portends, we must usually find an intervention. In Appendix B we show a more comprehensive description of how the economy works and ask for an explanation for what caused the recent economic crisis and how it may be prevented from happening again. Unfortunately, this suggests that if we continue what we are doing now, we can expect to repeat this crisis. The example in the appendix may be naïve. It requires further work by people who know econometrics better then I to put together the cause-and-effect statements that describe the economic situation.

Many problems can be stated in this form. Primarily it can be used to diagnose the causes of system failures, as in machines, processes, or medical diagnoses. But there are many other important applications.


We will consider that cause-and-effect can be represented by logical implication. But the logic required to solve these problems does not fit within the domain of classical propositional logic, which was developed to deal with open-infinite systems. For example, if A implies B, B is considered to be true even when A is not true. Supposedly if A does not imply B, then something else is assumed to imply B. But this something else may be outside the system that we know or care nothing about.

We had to use a different logic that will deal with closed-finite systems. If A is false, for B to be true we must be able to find an effect elsewhere within the system that implies that B is true. If we cannot find an implication that would make B true, B is considered to be false. And we must assure that this new logic satisfies all the consistency conditions that would be expected of a classical logic. This has not been trivial.

Furthermore, this new logic must be able to deal with circuits in the cause-and-effects. That has required the use of graph theory where the nodes of the graph represent the effects and the arcs represent cause-and-effect relations. Graph theory is used to find blocks containing the circuits and solving for the natural solutions for each block, i.e. solutions that would occur if there were no input causes from outside the block. These natural solutions are used to find simple cause-and effect relations between the inputs and outputs of the block that replace the circuits.

Once there are no circuits, the sets of assumptions that explain the behavior can be found directly as a function of the behavior to be explained by using substitution to eliminate the intervening effects, much as we solve systems of simultaneous linear algebraic equations.


The following illustrates how the method involving people and the computer program works:

  1. People with different perspectives on a problem collaborate to collect and discuss the pertinent cause-and-effect statements that describe the knowledge of the situation from which the behavior to be explained arises. These individual statements are usually quite simple and easy to understand and discuss. (We call this Collection.)
  2. The people present a behavior to be explained or produced. The computer uses this cause-and-effect knowledge of the situation to find all the sets of assumptions that would explain that behavior, or would produce that behavior if the assumptions were turned into actions. (This is Abduction.) Since the behavior is usually inadequately defined initially, many false explanations may be proposed.
  3. The computer takes each of these proposed explanations and uses the cause-and-effect knowledge to find all the other behaviors that that explanation would also predict. (Logical Deduction.)
  4. The computer asks the people to determine whether these other behaviors actually occur. So people then examine each of these other predicted behaviors to see whether they actually do occur. (This we call Examination.)
  5. Then people must rule-out all those false explanations leaving only those explanations that predict the desired behavior and do not also predict behaviors that do not occur. (This we call Selection.)
  6. For each explanation that remains, the computer produces a cause-and-effect scenario summarizing how the conclusions were reached. These scenarios can be used to either convince other people of the validity of the reasoning, or give them the information they need to refute the conclusions, providing the basis for improving the cause-and-effect knowledge. (This is called Scenario.) The is the CADESS (Collection, Abduction, Deduction, Examination, Selection, Scenario.)) method for solving problems.


Initially the cause-and-effect statements are likely to be wrong and incomplete. The results of the analysis might not be adequate to explain what is to be explained. The reasoning with those cause-and-effect statements shown in the scenarios can be analyzed to point out the weaknesses in these cause-and-effect statements or whether further such statements need to be added. The changes are made and the process is repeated until an acceptable explanation can be found.

So this is an iterative process in which the cause-and-effect statements are continually revised until a satisfactory understanding and solution emerges. Once this knowledge is established, it might be used, perhaps by adding more cause-and-effect statements, to solve other problems in similar domains. It could also be used by other people. Thus, a library of cause-and-effect knowledge about various domains might be developed.


The Explainer may be used to:

  1. Conduct the scientific method for the analysis of new observations and development of new theories to explain them.
  2. Form a social network where people can discuss their ideas about the current social problems of the day and use the computer to extend their reasoning capabilities in order to enhance their discussion. They may possibly communicate their reasoning to those who would have the responsibility and capability of implementing their solutions. They may be able to shoot down the many fallacies that are floating around so as to reduce some of the chaos we observe today. Money could be made by charging for the time the system is used and by charging for the use of their knowledge base by others. Those using the knowledge may choose to receive advertisements showing where they can purchase or find literature that relates to their conversation. Clicks on those advertisements may also generate revenue.
  3. Diagnose the cause of medical symptoms. A panel that keeps up with the latest literature may develop and maintain an up-to-date medical knowledge base that would be provided as a subscription to physicians. If this knowledge base is respected by the courts, it could protect physicians against malpractice suits and reduce the number of costly and unnecessary tests. If the physicians provide feedback of the symptoms, treatments, and results, it could provide for the collection of Experienced Based Medicine.
  4. Diagnose the causes of failures in many other types of devices or systems such as machinery or even production processes.
  5. Determine what is most likely to have happened in the past based on forensic evidence as in solving crimes, or as in analyzing historic or archeological evidence to develop an historical explanation.
  6. Teach students how to solve problems. The students could use the Internet to find problems to solve and find the information they need to generate the cause-and-effect statements to develop a knowledge base that describes the situation from which the problem arises. Then they would use this method to attempt to develop solutions to these problems. Their solutions might be submitted to those who have the responsibility and means to implement the solutions they propose.
  7. Develop plans for how to achieve desired objectives where the cause-and-effect statements of the scenario derived by explaining the goal become the steps in the plan.
  8. Handle the management of emergencies when time is inadequate for people to think the problem through.

The Explainer has also been used in tests to illustrate how it might be applied to understanding a number of different types of problems such as understanding how terrorists might design a plan to do us harm and how all those plots might be thwarted with the most economical interventions, how we might consider the various factors involved in plans to leave Iraq (now it should be applied to leaving Afghanistan), how it might be used to provide a medical diagnostic service to physicians who do not have the time to read all the latest literature, and so on.

But none of these examples was intended to do any more than to illustrate how these problems might be formulated and resolved. Better understanding of these problems would require that the method be put in the hands of more knowledgeable people to develop better cause-and-effect statements.


The following are some preliminary ideas of how it might be used. But they will require more investigation.

  1. Design systems by representing the behavior of each likely component of the design by cause-and-effect statements, then requesting the method to explain how these components can be used to explain the requirement specifications of the design.
  2. Develop computer programs by using previously programmed components to satisfy the required program specification. The input and initial data state of a component would be its cause, and the output and final state would be its effect. The assumptions would be the input to the program that will produce the desired result. Changes to the program could be made by having the method automatically regenerate the program by just changing the pertinent pieces of the requirements without the danger of introducing errors while trying to change the program manually.
  3. An agent that can be used as a device such as a robot that can use its sensors to detect and collect likely cause-and-effects occurring in its environment. Then use these to develop and extend its existing set of cause-and-effect statements. And then use its cause-and-effect knowledge with actuators to experiment on its environment to rule out false cause-and-effects and learn more cause-and-effect statements, and also produce effects it desires to occur in its environment. (This might also be the basis for a philosophical theory to explain how we learn from our environment.)

In this method, probabilities may be added to the cause-and-effect statements to indicate the prior belief in the validity of the individual cause-and-effect statements to produce solutions with a probability measure of belief in its outcomes.

The same method used here to work with circuits may be applied in other contexts where Bayes theorem is used.



The following illustrates how a collaborative conversation can collect the cause-and-effect statements that describe the knowledge about the situation from which the behaviors to be explained can arise. This illustration is very short and naïve, just enough to indicate the concept. The problem shown here is a very simple model of how to deal with the economic crisis. A more comprehensive model is shown in Appendix B. But even that is only a beginning. A more complete explanation would require the use of the method by more knowledgeable econometricians. Rather than use cause-and-effect statements, this conversation may use WHY-BECAUSE statements that are equivalent to cause-and-effect statements, but may be more natural for many to use.

WHY: Many foreclosures.
BECAUSE: Borrowers have low income and equity
AND Borrowers could buy Adjustable Rate Mortgages at low initial interest rates
AND Lenders were willing to sell low rate ARM mortgages
AND Rates on ARM mortgages later rose
AND Borrowers with low income and low equity could not pay the higher rates.
WHY: Lenders could get back the houses and principal already paid.
BECAUSE: Many foreclosures.
WHY: Borrowers with low income and low equity could buy homes with Adjustable Rate Mortgages? BECAUSE: ARMs initially had low interest rates.
WHY: Lenders were willing to sell low rate ARM loans?
BECAUSE: Lenders had low risks in selling low interest rate loans.
WHY: Lenders had low risks in selling low interest rate loans?
BECAUSE: Lenders could bundle the loans and sell them at low risks to them AND Investors believed that bundled loans had low risks.
WHY: Investors believed that bundled loans had low risks? BECAUSE: Bundled loans were rated as low risk by the rating agencies.
WHY: Bundled loans were rated as low risk by the rating agencies?
BECAUSE: Rating agencies were financed by the lenders AND Lending agencies financed by lenders was not against regulations.
WHY: Lending agencies financed by lenders was not against regulations?
BECAUSE: Lenders contributed financial support to congressional campaigns.
WHY: Lenders contributed financial support to congressional campaigns?
BECAUSE: Unattributed business contributing to congressional campaigns was not regulated.



Cause-and-effects can be input to the Explainer program in the form of an outline as follows:


Meaning that Effect 1 is caused either by (A and not B) or by C.


Now we ask the computer for the assumptions that would explain the behavior: ‘Many Foreclosures’. The computer responds with:

EFFECT   1 Many Foreclosures
       2 Borrowers have low income and equity
AND    4 Rates on ARM mortgages later rose
AND    5 Borrowers with low income and equity could not pay higher rates
AND   16 Unattributed businesses contributing to Congressional campaigns is not regulated

This shows that there are four causes that when taken together, i.e. joined by logical ANDs, are necessary to produce the behavior that was to be explained. But note that the first three causes are not within our ability to change. The forth cause, ‘16 Unattributed businesses contributing to Congressional campaigns is not regulated’ is something that might be changed through legislation. Since all these causes are required to produce the effect, eliminating any one or more causes will prevent the effect from occurring.

This is a trivial example involving only 16 effects. Appendix B shows a more comprehensive analysis involving 34 effects. But this is still only a start. It needs to be worked on further by econometricians more knowledgeable about economics then I.


If people could be convinced that this method works, the implications would be substantial. We could dispel some of the chaos in government we are currently facing.

When presented to an expert in logic, the response is supportive of the logic used and enthusiastic about the implications of using this method.

But unfortunately when presented to someone who does not have an in depth understanding of logic, it is generally greeted with suspicion. ‘It won’t work. It could never be done’. When they look at examples of problems that have been solved using this method, if it confirms their prior beliefs, they respond that the conclusion is obvious and such a method is not needed. If it does not correspond to their prior belief, they consider that the method and its conclusions must be wrong. ‘How dare you disabuse me of my fantasy?’ In this paper we do dare. But the potential of this method is being neglected.

If this method could be turned into a social network to discuss, understand, and resolve complex problems, people would use it and it would just be taken for granted that the method works. It could then be used as a political BS detector.


The Explainer shows a method that might be used to help people collaborate in order to understand complex problems and see how they might be resolved. Unlike expert systems, this method can handle circuits, which are not uncommon, and there is no ambiguity in the sequencing of the application of rules.

If equivalent methods are available to study complex problems such those that this method appears capable of understanding, there does not appear to be any evidence that such other methods exist and are being used to deal with the many significant problems that we are facing today.

We have a running program that so far handles problems without cause-and-effect circuits, and are finishing the program so it will handle circuits. In the meanwhile we have been able to handle some problems with circuits by supplementing the program with hand manipulations to deal with the circuits.


  • [1] Brachman, Ronald & Hector Levesque (2004) Knowledge Representation and Reasoning (Morgan Kaufmann)
  • [2] Garey, Michael R. & David S. Johnson, (1979) Computers and Intractability: A Guide to the Theory of NP-Completeness (W. H. Freeman and Company)
  • [3] Glymour, Clark (2001) The Mind’s Arrows: Bayes Nets and Graphical Causal Models in Psychology (The MIT Press)
  • [4] Gopnik, Alison, (2009) The Philosophical Baby, (Farrar, Straus and Giroux)
  • [5] Gore, Al (2007) The Assault on Reason (The Penguin Press, New York)
  • [6] Groopman, Jerome, (2007) How Doctors Think (Houghton Mifflin)
  • [7] Josephson, John R. & Susan G. Josephson (1996) Abductive Inference: Computation, Philosophy, Technology (Cambridge University Press)
  • [8] King, Ross D., Oliver Ray, Amanda Clare, Maria Liakata, Larisa Soldatova, Ken Whelan “Towards the automation of scientific method” 21st International Joint Conference on Artificial Intelligence IJCAI'09
  • [9] King, Ross D. “Rise of the Robo Scientists” Scientific American January 17, 2011
  • [10] Mueller, Eric T. (2006) Commonsense Reasoning (Morgan Kaufmann Publishers)
  • [11] Pearl, Judea, (1988) Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference Revised 2nd Printing (Morgan Kaufmann, Publishers, Inc.), particularly pg. 195 footnote,
  • [12] Pearl, Judea, (2000) Causality: Models, Reasoning, and Inference (Cambridge University Press)
  • [13] Peirce, Charles Sanders, (1998) “Pragmatism as the Logic of Abduction” in The Essential Peirce, Volume 2: Selected Philosophical Writings, 1893-1913 (Indiana University Press)
  • [14] Peng, Yun & James A. Reggia, Abductive Inference Models for Diagnostic Problem-Solving (1990) (Springer; 1st edition)
  • [15] Russell, Stuart. & Peter Norvig, (2003) 2nd ed., Artificial Intelligence: A Modern Approach, (Prentice-Hall of India)
  • [16] Steward, Donald V, (1981) Systems Analysis and Management: Structure, Strategy and Design, (Petrocelli Books, Inc., now McGraw-Hill)
  • [17] Walton, Douglas (2004) Abductive Reasoning (The University of Alabama Press)
  • [18] Zhang, Du & Doan Nguyen (1994) “PREPARE: A Tool for Knowledge Base Verification” IEEE Trans. on Knowledge and Data Engineering, vol. 6, No. 6, Dec 1994)
  • [19] Go to then Click on APPLICATIONS-E.doc to see demonstrations of how the Explainer has been used to deal with various types of problems.

Note: Some of this work was done in conjunction with Joshua Knight, graduate student, and David Singer, professor, of the University of Michigan, and Janel Nixon of Integrative Engineering under a grant from the Office of Naval Research. This grant has been terminated.  

Look for more blogs |