Heuristics Are Not That Simple: Review of Simple Heuristics That Make Us Smart by Gerd Gigerenzer, Peter M. Todd, and the ABC Research Group
|Simple Heuristics That Make Us Smart, by Gerd Gigerenzer, Peter M. Todd, and the ABC Research Group, Oxford University Press, 1999, 416 pp. $35.00|
Business rules come in various shapes, sizes, and types. A term type business rule is always a well-defined word or phrase, such as Product (a physical thing our company sells right now) or Customer (a person or organization we do, or have done, business with). A fact type business rule is a relationship between terms, such as a Customer can place many Orders. Heuristic business rules are a mystery to many people, and a heuristic business rule takes more explanation.
A recently-published book goes into great detail about what heuristics are and how they can be used to make good decisions. This book may help business rules analysts understand just what a heuristic is and why it is a significant part of human decision-making. It also demonstrates why heuristic business rules may be hard to formulate. It does not give lists of sample heuristic business rules, and it doesn't teach you how to write one.
This book comes out of an academic background, from the Center for Adaptive Behavior and Cognition (ABC), an interdisciplinary research group based at the Max Planck Institute for Human Development in Berlin. It is edited by Gigerenzer and Todd and consists of 16 chapters written by various combinations of 18 members of the ABC Group, who work in Germany, the USA, and the UK.
Many of the chapters are write ups of research conclusions from fields as diverse as psychology, sociology, biology, statistics, computing science, and economics. References are also made to cognitive psychology, economic game theory, and animal learning. This may seem a little far afield from the realm of business, but there is much to learn about heuristics in the book, which I'll call "Simple" for short.
So, what are heuristics? A quick look in a fat dictionary (Webster's Third, 1993) shows three possibly useful entries -- the first two for heuristic as an adjective.
aid or direction in the solution of a problem but otherwise unjustified or
incapable of justification." In other words, something that helps
you find an answer, but you can't prove that it helped, except that it did.
Heuristically speaking, this definition isn't too useful and, in a book
consisting of statistically backed-up research finds, is definitely not what
these authors have in mind.
- The next
definition of heuristic, also as an adjective, mentions exploratory
problem-solving techniques, self-educating techniques, and feedback to
improve performance. The example given is a heuristic computer
program. This one is a little closer to our ballpark. The
fields of artificial intelligence (AI), neural nets, and robotics come to
mind here. An interesting chapter section is on the game of chess and
the heuristics used in the Deep Blue computer defeat of Garry Kasparov in
- The third definition, although a noun, is not satisfying. "The science or art of heuristic procedure," which I interpret circularly to be procedure that is heuristic.
Surprisingly, "Simple" does not have a statement that says directly what a heuristic is. The book has a section in the introductory chapter that traces the word's Greek origin -- "serving to find out or discover" also is used in an adjectival sense. The authors use it as a noun and in sentences that contain phrases such as "a strategy for making a decision" and "inference mechanisms," but they never define it. They appear to mean decision-making mechanisms used by humans or computers.
A phrase the authors use throughout the book is "fast and frugal heuristics." This is what they mean by simple heuristics. The authors make clear that their purpose is to test various kinds of heuristics and see whether or when people really use them. They conclude that people do.
The authors use computational models to investigate simple heuristics. They look at environments where heuristics might perform well, and try to test them in those environments. Many of these are worth looking at by business analysts.
The authors lay out three areas about heuristics that concern business rules analysts: the principles used to (1) guide a search, (2) stop the search, and (3) make a decision using the search results. They explain when and why people use search, stop, and decision with simple heuristics to perform well, without the aid of a computer and under great time pressure. The authors use the example of a lion chasing prey to demonstrate time pressure. Deer clearly don't use computers nor do they have the luxury of leisurely contemplation of the situation.
In a similar (if not quite so fatal) vein, "Simple" has a chapter on something they call the recognition heuristic. They invested actual funds in the stock market and made a modest profit on stocks people recognized over ones they didn't. (They used German stocks for Americans and American stocks for Germans.) They proved statistically that this heuristic works better for people who are more ignorant about particular stocks than market experts.
The recognition heuristic is a one-step decision and therefore fast and frugal. It doesn't require years of following stocks and markets, thick compendiums of stock evaluations, or a giant computer database. This kind of heuristic (1) uses recognition to make a simple search, (2) stops after the one decision, and (3) invests based on the search results. This may be the way some people think, but what about computers?
Computer searches generally don't need to have a stop mechanism. Order is irrelevant -- if time consuming. They usually look at everything, and declare what the computer finds is correct, since there is no other alternative except 'Not Found.' With all the data indexes and clever automated search mechanisms, what we need is a faster computer.
Not so, claim the authors. They mention several very large numbers of alternatives turned up in their research statistics. The chess example is a good one. The number of alternative positions after just eight chess moves is 35 to the 8th power, or more than 2.2 trillion. To calculate all possible positions is impossible, even for Deep Blue -- and that's on a strictly limited 8x8 field where all possible answers are theoretically known in advance. In most real world situations neither people nor computers know all the possibilities.
What the authors do suggest, although indirectly, is that analysts and programmers setting up search trees need good heuristics to pick the search criteria and to order them well. In decision-making situations where all outcomes cannot be predicted, good search criteria and search order are essential, even with the fastest supercomputers available. Sorting and searching through charge and payment transactions for a monthly billing can be optimized to perfection. Heading off an incoming nuclear warhead cannot, and the heuristics better be at least as good as for the deer pursued by the lion.
Statistical and mathematical models are ubiquitous in research results and the studies in "Simple" are no exception. Occam's razor, which has come to be understood lately as "use the simplest model that explains the data," heads Chapter 6, one of several with mind numbing statistical formulas. You can probably skip most of the statistical information in the book, unless reading about Markov blankets, Landau's criterion, and the Mann-Whitney U-test is your idea of a good time. If you nod in a chapter, go on to the next one.
For a respite, look at Chapter 13, the one on mate search. It has funny starting quotes, understandable graphs, and a really good demonstration of how searching less can be useful. It might even help you find a mate.
To have a complete business model requires you to include business rules. Sooner or later, heuristic business rules will come up, and "Simple" will help give you an idea of what to look for. The order of how business rules apply is a significant part of decision tree models. A good programmer does those models before coding. If your business needs very quick decisions, such as to approve credit instantly or to head off a missile attack, you may not have time to search large amounts of data, even if you have extensive data warehouses. A good analyst can anticipate such requirements using some kinds of heuristic business rules.
Where do heuristic business rules fit in the Zachman Framework, the most useful mechanism for categorizing models? Current thinking is the Motivation (or Why) Column. Rows 1 and 2 are places for planning and business analysis, and heuristics need fairly early and high-level input in planning and development.
Like many academic studies, this book has a huge bibliography. Little in it applies to business rules, business modeling, or data processing. There are a few computer science and AI references, but most are not relevant. The index and table of contents may be more useful. The type is too small, but the book is well edited and proofread. Several studies contain humor, to get readers through them.
Copyright, 2000. September 2000
(c) Michael Eulenberg, 2000