|
Simple Heuristics
That Make Us Smart by Gigerenzer and Todd
This book is recommended for those involved in the field of computer chess who want to better understand
how humans behave when attempting to solve problems.
This book is about fast and frugal heuristics. The theory
of fast and frugal heuristics involves two principles for rational decision making: 1. decision rules are bounded
in their rationality (rules are frugal in what they take into account, and therefore fast in their operation) 2. the rules
are ecologically adapted to the environment, which means that they "fit to reality."
[From wikipedia, Ecology, or ecological science, is the study of the distribution and abundance of living organisms
and how these properties are affected by interactions between the organisms and their environment. The environment of an organism
includes both the physical properties, which can be described as the sum of local abiotic factors like climate and geology,
as well as the other organisms that share its habitat]
Perhaps we should look at how real decisions are made. Humans have limited time, information and
knowledge, yet somehow they make decisions that are adequate, good enough, or acceptable. Gigerenzer suggests that
"fast and frugal heuristics" not only explain human decision making, but can be useful for building intelligent systems.
p.5"This book is about fast and frugal heuristics for making decisions - how they work, and when
and why they succeed. These heuristics can be seen as models of the behavior of both living organisms and artificial systems.
From a descriptive standpoint, they are intended to capture how real minds make decisions under constraints of limited time
and knowledge. From an engineering standpoint, these heuristics suggest ways to build artificially intelligent systems - artificial
decision makers that are not paralyzed by the need for vast amounts of knowledge or extensive computational power. These two
applications of fast and frugal heuristics do not exclude one another - indeed, the decision tree in figure 1.1 [not reproduced
here - classifies incoming heart attack victims as low risk or high risk patients] could be used to describe
the behavior of an unaided human mind or could be built into an emergency-room machine... We propose replacing the image of
an omniscient mind computing intricate probabilities and utilities with that of a bounded mind reaching into an adaptive toolbox
filled with fast and frugal heuristics."
The environment plays a critical role for decision-making heuristics that are expected to be successful. In
fact, one cannot even begin to construct such a heuristic unless one understands the environment and the hidden cues present
in the environment that lead to discovery.
p.13"The second component of [Herbert] Simon's view of bounded rationality, environmental structure,
is of crucial importance because it can explain when and why simple heuristics perform well: if the structure of the heuristic
is adapted to that of the environment. Simon's (1956a) classic example concerns foraging organisms that have a single need:
food. One organism lives in an environment in which little heaps of food are randomly distributed; it can get away with a
simple heuristic, that is, run around randomly until a heap of food is found. For this, the organism needs some capacity for
vision and movement, but it does not need a capacity for learning. . A second organism lives in an environment where food
is not distributed randomly but comes in hidden patches whose locations can be inferred from cues. This organism can use more
sophisticated strategies, such as learning the association between cues and food, and a memory for storing the information.
The general point is that to understand which heuristic an organism employs, and when and why the heuristic works well, one
needs to look at the structure of the information in the environment. Simon (1956a) was not the only one to make this important
point; it was made both before his work.. and at various times since... including the extreme statement that only the environment
need be studied, not the mechanisms of the mind... We use the term 'ecological rationality' to bring environmental structure
back into bounded rationality. A heuristic is ecologically rational to the degree that it is adapted to the structure of an
environment... Thus, simple heuristics and environmental structure can both work hand in hand to provide a realistic alternative
to the ideal of optimization, whether unbounded or constrained."
A decision should also be robust - that is, it should be able to withstand a limited number of unforeseen
circumstances. The decision should be made after looking at the most important factors and should not be influenced by "noise".
p.20"Robustness goes hand in hand with speed, accuracy, and especially information frugality. Fast
and frugal heuristics can reduce overfitting by ignoring the noise inherent in many cues and looking instead for the "swamping
forces" reflected in the most important cues. Thus, simply using only one or a few of the most important cues can automatically
yield robustness."
Heuristics should owe their success or failure to measured performance in the real world. In fact,
an iterative cycle of testing the heuristic and modifying the heuristic should be part of any heuristic development. The development
cycle should include testing that simulates, as close as possible, the performance conditions that are likely to be seen in
real-world conditions:
p.22"To conclude: Heuristics are not simply hobbled versions of optimal strategies. There are no
optimal strategies in many real-world environments in the first place. This does not mean, though, that there are no performance
criteria in the real world. As a measure of success of a heuristic, we compare its performance with the actual requirements
of its environment, which can include making accurate decisions, in a minimal amount of time, and using a minimal amount of
information."
Heuristics use the information present in the environment, in an adaptive way in order to become
useful and successful. Think how many times you used "Google" to get information on a subject of interest, or used Mapquest
or Mapblast to find directions to where you want to go. Years ago, people went to the library and used an Encyclopedia
or an almanac as a quick reference. Since decision time is often limited, a quick check of Google might be the only step many
people take before making a decision.
p.24"Bounded rationality. Decision-making agents in the real world must arrive at their inferences
using realistic amounts of time, information, and computational resources... Ecological rationality. Decision-making agents
can exploit the structure of information in the environment to arrive at more adaptively useful outcomes. To understand how
different heuristics can be ecologically rational, we characterize the ways information can be structured in different decision
environments and how heuristics can tap that structure to be fast, frugal, accurate, and otherwise adaptive at the same time."
Sometimes a heuristic is useful for just thinking about a (possibly complex) problem:
p.26"As used by [Albert] Einstein, then, a heuristic is an approach to a problem that is necessarily
incomplete given the knowledge available, and hence unavoidably false, but which is useful nonetheless for guiding thinking
in appropriate directions."
A heuristic is like an expert reaching into a toolbox for a custom-made tool, carefully
shaped to fix a difficult mechanical problem:
p.30"Just as a mechanic will pull out specific wrenches, pliers, and spark-plug gap gauges for each
task in maintaining a car's engine rather than merely hitting everything with a large hammer, different domains of thought
require different specialized tools. This is the basic idea of the adaptive toolbox: the collection of specialized cognitive
mechanisms that evolution has built into the human mind for specific domains of inference and reasoning, including fast and
frugal heuristics... But we believe that simple heuristics can be used singly and in combination to account for a great variety
of higher order mental processes that may at first glance seem to require more complex explanation..."
Knowledge of the problem can help us decide which heuristic to use for which problem:
p.32"How does the mind choose which heuristic in the adaptive toolbox to apply to a specific problem?...
In cases where there is more than one applicable heuristic, the knowledge that the decision maker has can be used to select
the heuristic... Other external factors, such as time pressure and success, may further help to select heuristics... The tools
in the adaptive toolbox are made from more primitive components, including the heuristic principles of information search,
stopping, and decision discussed earlier."
It seems that one critical difference between an expert and a novice is the type of "cues" that are used for decision
making. The quantity of information gathered does not appear to be significantly different:
p.166"Shanteau (1992) ["How Much Information Does an Expert Use", Acta Psychologica 81]
argued that the amount of information used is not connected with expertise. The assumption that experts make better judgments
because they use more information does not appear to be true. Instead Shanteau assumed that the difference between novices
and experts is the ability to discriminate between relevant and irrelevant cues. Experts seem to use the same number
of cues but are more likely to use cues that are more useful for making appropriate decisions. Klayman (1985) examined the
influence of cognitive capacity. He showed that children with high cognitive capacity acquire more information for important
decisions where several alternatives are available."
Most knowledge is dynamically updated from our present store of knowledge. We do not need to have
instant recall of everything we learn. We do, however, need to dynamically update our knowledge base as new information is
presented to us:
p.208"The adaptive process of knowledge updating relieves us of the need to store everything we
have thought, said, or experienced in the past. Updating makes us smart by preventing us from using information that may be
outdated due to changes in the environment. As [Sir Frederic] Bartlett put it: 'In a world of constantly changing environment,
literal recall is extraordinarily unimportant'... Adaptive updating has an uninvited byproduct: hindsight bias. But this by-product
may be a relatively low price to pay for a memory that works fast and frugally."
Gigerenzer and Todd (along with co-authors Goodie, Ortman, Davis, Bullock and Werner) now turn
their attention to the game of chess, and offer their insight. Clever heuristics seem to be the answer for reducing the
staggering amount of possible moves in the search tree. The authors speculate that it is diagnostic information present in
the location of pieces on the chess board that should be tabulated and used to focus the search efforts. The evaluation
function should be used to answer the question, "what moves are more promising for further inspection". The authors discuss
a heuristic algorithm similar to the one being proposed in this paper. The quotation below is among the most powerful
support I can find in published literature for the heuristic ideas proposed, and in fact it is something that I wished myself
to have written:
p.329"Getting Along in a Deceptively Simple Environment: Chess
In 1992, Herbert Simon wrote: 'It is difficult to predict when computers will defeat the best human
player....It can only be a matter of a few years before technological advances end the human supremacy at chess' (Simon &
Schaeffer, 1992, pp. 14-15). Simon's prediction was right, and five years later, in 1997, [May 11] a computer called Deep
Blue finally defeated Garry Kasparov, the world's best human chess player, in a tournament. This watershed event in the development
of artificial intelligence was also a significant moment in the development of satisficing mechanisms. The game of chess occurs
in a small field (an area of 64 discrete squares), with a limited set of alternate events at each point in the game (16 pieces
per player, each with a defined set of legal moves), and a single goal (the capture of the opponent's king). What could be
simpler? But as anyone knows who has attempted to play the game, or who has tried to program a computer to play it, finding
the one best solution to the problem of putting an opponent in checkmate is unimaginably difficult. There are simply too many
possible lines of play to simulate them all, even with the most powerful computers... At the most elevated level, the strategy
of chess programmers is like that of expert players, namely to try to find lines of play that seem likely to lead to good
outcomes short of checkmate. So, like humans, computers must try to evaluate the relative value of various positions that
are available after some smaller number of moves... The attempt to limit the scope of the search in chess dates at least to
the work of Herbert Simon and George Baylor (Baylor & Simon, 1966)... They reasoned that one major goal of the game is
to reduce the opponent's mobility... and so paths were sought solely on the basis of how much mobility the opponent had following
each move... Clever design of the evaluation function is the difference between an artificial world champion and a machine
that races toward bad outcomes. Indeed, the creators of Deep Blue credited their recent victory over the world champion to
their much improved evaluation function... The key to improved chess is thus recognized to be clever algorithms - in other
words, those that are fast, frugal, and accurate - rather than more powerful attempts... These algorithms must find
the cues that are subtle but available on the board and provide the most diagnostic information about the outcome of various
moves; in other words, they must capitalize on the structure of the environment... Since the cleverness of the evaluation
function, rather than the power of the computer, has proved to be a better road to improved chess, perhaps chess research in
the future will return to the satisficing notions with which it began."
So what is the deal with fast and frugal heuristics? Why use them at all? Why are they better than
any other approach and what makes us think they can be useful for building a machine that plays chess?
p.360"There are two reasons for the surprising performance of fast and frugal heuristics: their
exploitation of environment structure and their robustness (generalizing appropriately to new situations as opposed to overfitting
- see chapter 1). Ecological rationality is not a feature of a heuristic, but a consequence of the match between heuristic
and environment... By matching these structures of information in the environment with the structure implicit in their building
blocks, heuristics can be accurate without being too complex. In addition, by being simple, these heuristics can avoid
being too closely matched to any particular environment - that is, they can escape the curse of overfitting, which often strikes
more complex, parameter-laden models. This marriage of structure with simplicity produces the counterintuitive situations
in which there is little trade-off between being fast and frugal and being accurate."
One could argue that the heuristic proposed in this paper is neither fast nor frugal, when compared to the heuristic
presently being used by traditional chess programs. In response, what ultimately matters is performance, not speed or quantity
of information processed - "Furious activity is no substitute for understanding (H. H. Williams)". Performance can be measured
by wins and losses in tournament games. It can be measured by the number of test problems that are correctly solved. It can
also be measured by the number of chess programs sold on the commercial market.
Perhaps the proposed heuristic has a greater usefulness for those who are performing analysis, such as those involved
in correspondence chess games or those who analyze the completed games of top players.
|
