Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

RE-REPRESENTATION AND EMERGENT INFORMATION IN THREE

CASES OF PROBLEM SOLVING

DONALD PETERSON
University o/Birmingham

1. Introduction
An interesting case of creative thought is to be found in problem solving through
re-representation. We are presented with a problem in its original representation,
in which condition it is laborious if not impossible to find a solution, and we then
discover a new way of formulating the problem which renders it dramatically easier
to solve.
The phenomenon is well recognised. As Holyoak (1990) puts it "People do not
always simply establish a representation of a problem and then perform search;
rather, they sometimes change their representation in major ways". As Boden
(1990: 94) puts it "In general, problem solving is critically affected by the rep-
resentation of the problem that is used by the problem-solver". Re-representation
has been advocated as a practical technique in mathematical problem solving (polya,
1948), and the Gestalt school of psychology has given particular attention to the
're-structuring' of problems prior to their solution (Duncker, 1945; Ohlsson, 1984;
see also Boden, 1987: 333-344).
It might seem natural to characterise such re-representations as 're-description
of problems under useful new concepts'. The purpose of the present chapter is to
investigate the question whether this characterisation is adequate. To this end we
examine three problems and possible solutions through re-representation, and we
answer our question in the negative. The re-representations considered do crucially
involve the introduction of useful new concepts, but they also crucially involve
extra 'emergent information' which can be calculated once the new concepts are
introduced (and which in some cases is then built into a diagram). It is the presence
of this emergent information, and the calculation which generates it, which make
the cases below more than re-descriptions under new concepts.

2. The Four Knights problem


2.1. INITIAL REPRESENTATION

The Four Knights Problem is initially presented as follows. Given the chess con-
figuration in Figure I, the task is to find the minimum number of standard knight's
moves necessary to interchange the positions of the black pair and the white pair
of knights, within the constraints of the nine square board.
81
T. Dartnall (ed.), Artificial Intelligence and Creativity, 81-92.
© 1994 Kluwer Academic Publishers.
82 DONALD PEI'ERSON

G H ·1

Fig. 1. initial representation of the Four Knights problem

Clearly, we may succeed in interchanging the positions of the two pairs of


knights by experiment, but it is not easy to be certain that we have actually found
the minimum number of moves required to do so. We try therefore to devise
an alternative formulation of the problem which simplifies the calculation of its
solution.

2.2. INTRODUCTION OF NEW CONCEPTS

We look first at the situation of a single knight. We define a state to be the positioning
'of the knight on a square of the board. Thus the set of states is: {A, B, C, D, E,
F, G, H, I}. Where a single knight's move allows transition from a state S to a
state T, we say that S has access to T, and notate this as 'a(S, T)'. The access
relation is symmetrical, that is, a( S, T) ¢} aCT, S). We define a route as a series
of states such that each state has access to the next.

2.3. CALCULATION OF EMERGENT INFORMATION

We can now, by hand or by computer program, calculate extra information using


these concepts. The set of (symmetrical) access relations as given by the problem is
{a(A, F), a(F, G), a(G, B), a(B, /), a(I, D), a(D, C), a(C, H), a(H, An. We
note that state E cannot be accessed. We pick a starting state, say A, and calculate
the possible routes from A. What we find is that only one (symmetrical) route
is possible, and that it eventually leads back to A. This is the 'circular' route
(A, F, G, B, I, D, C, H, A). This incorporates all states, except E, and all access
relations, and so whatever state we start with the same will apply. The 'transition
space' of the problem is thus circular, and we have discovered a structure to the
problem which was not initially apparent. This discovery promises to simplify the
process of solution, and so we go on to incorporate it in a re-representation of the
problem.
RE-REPRESENTATION AND EMERGENT INFORMATION 83
It is significant that this information is additional to that given in the initial
representation of the problem, which makes no mention of a circular structure.
Also, the information is emergent in the sense that it employs concepts not present
in the initial representation-the initial representation is given in terms of a spatial
diagram of a board, and a spatially defined knight's move, not in terms of states
and access relations.

2.4. RE-REPRESENTATION AND SOLUTION PROCEDURE

The circular structure will constrain the movements of four knights, just as it
constrains the movements of one. Thus, using our new information, we can re-
represent the Four Knights Problem diagrammatically in the manner of Lauriere
(1990: 47) as in Figure 2.

Fig. 2. re-representation of the Four Knights problem

In the diagram, a knight can move to one or other of its adjacent positions, and
our task is to find the length of the shortest sequence(s) of moves required to change
the present configuration into one in which points G and I are occupied by white
knights, and points A and C are occupied by black knights. All that is required is
to move each knight in sequence one position in a clockwise direction, continuing
until the goal state is reached. This will involve sixteen individual knight's moves.
Variations on this procedure are possible, but it is clear that they will not reduce the
number of moves required. The answer to the problem, therefore, is •sixteen' .

3. The Magic Square


3.1. INITIAL REPRESENTATION

The problem of the 3 by 3 Magic Square is to arrange the numbers 1 to 9 in a grid


such as the one below, so that each row, column and diagonal of the grid adds up
to 15.
84 DONAlD PETERSON

A B C

D E F

G H I

uncompleted magic square

Although easy to state, the problem in this form has a large search-space of 9!
(362,880) and is by no means easy to solve. However, by using a method originally
devised by Sloman (1989: xxiii-xxiv) we can re-represent the problem so as greatly
to reduce its search space.

3.2. INTRODUcrrON OF NEW CONCEPTS

In looking for useful new concepts we abstract away from the physical represen-
tation of the grid. We notice that it constitutes a set of 8 triples of slots (the rows,
columns and diagonals of the grid), with an intersection pattern. And we notice that
each slot belongs to a given number of triples. For example, the slot A belongs to
three triples: (A, B, C), (A, E, I) and (A, D, G). The number of triples to which a
slot belongs we call its level.
When a number is assigned to a slot in a triple, the other slots in the triple
will have to be filled in accordance with the '15 constraint', and so we define a
completion pair of a number n to be a pair of numbers (different to each other and
different to n) between 1 and 9 which together with n add up to 15. And the number
of completion pairs a number has we again call its level.
We notice that if a number is to be allocated to a slot, then it must have enough
completion pairs to fill up all the triples in which that slot occurs. So we have
the levels constraint: in allocating a number to a slot, the number's level must be
greater than or equal to the slot's level.

3.3. CALCULATION OF EMERGENT INFORMATION

On the basis of the above we now calculate the levels of the slots, and the completion
pairs and levels of the numbers 1 to 9.
Again, this information concerning levels and completion pairs is additional and
emergent relative to the initial representation.
RE-REPRESENTATION AND EMERGENT INFORMATION 85

A13 BI2 C/3

D12 EI4 Fl2

GI3 H/2 113

slots with levels

I numbers I levels I completion pairs


1 2 (5, 9), (6, 8)
2 3 (4, 9), (5, 8), (6,7)
3 2 (4, 8), (5,7)
4 3 (2, 9), (3, 8), (5,6)
5 4 0,9), (2,8), (3,7), (4,6)
6 3 (1,8), (2,7), (4,5)
7 2 (2, 6), (3, 5)
8 3 0, 6), (2, 5), (3, 4)
9 2 (I, 5), (2,4)

numbers with levels and completion pairs

3.4_ RE-REPRESENTATION AND SOLUTION PROCEDURE

The re-representation of the problem comprises the two tables above, together with
the levels constraint. We pick a slot and allocate a number to it, consistently with
the levels constraint. We then fill up whatever rows, columns and diagonals the
slot belongs to by using the number's completion pairs, consistently with the levels
constraint. We carry on, observing the levels constraint and the '15' constraint, and
backtracking if necessary, until the grid is complete, and the result will be a magic
square. If we start by filling the middle slot E, there is only one choice, namely
5, and the remainder of the task is simply to allocate the four relevant completion
pairs appropriately.

4. Missionaries and Cannibals


4.1. INITIAL REPRESENTATION

The initial representation of the problem is as follows. Three missionaries and three
cannibals have to cross a river, from the left to the right bank. They have a boat
which can take one or two people (any person or pair of people from among the
86 DONAlD PETERSON

six). However, if at any time, on either bank, the cannibals outnumber the mis-
sionaries, the latter will be eaten. So the problem is to find the shortest sequence(s)
of crossings which will get them allover the river safely.
Again, we may find solutions to the problem by experiment with such things
as cutlery on a table. However it is not easy to see whether the sequences of
river crossings we find are of minimum length. We look therefore for a helpful
re-representation of the problem.

4.2. INTRODUCTION OF NEW CONCEPTS

Looking initially to eliminate irrelevant factors, we notice that it doesn't matter


which particular missionaries and cannibals are involved in any crossing. Also,
the time taken by a crossing is irrelevant, and we can therefore treat crossings as
instantaneous.
Looking for useful new concepts, we consider distributions of missionaries and
cannibals. Since the population of the right bank can be inferred from the population
of the left bank, together with the total numbers of missionaries and cannibals, we
need consider only the population of the left bank, and so we define a (left bank)
state as the numbers of missionaries and cannibals on the left bank. If there are m
missionaries and c cannibals on the left bank, we notate this state as s( m, c). Given
the constraints of the problem, some states will be legal, and others will be illegal.
We define a transition as an operation which converts one state into another. If a
transition moves m missionaries and c cannibals with the boat starting on bank b,
we notate this as 't( m, c, b)'. Those transitions which start on the left bank we call
'L-transitions', and those which start on the right bank we call' R-transitions'.

4.3. CALCULATION OF EMERGENT INFORMATION

We can now calculate some useful information, by hand or by computer program.


The legal states correspond to distributions of missionaries and cannibals in which
nobody gets eaten. These form the set {s(O, 0), s(O, 1), s(O, 2), 8(0,3), s(l, 1),
s(2,2),s(3,0),s(3,1),s(3,2),s(3,3)}.
The boat moves one or two people (one missionary, two missionaries, one
cannibal, two cannibals or one of each), and moves in one of two directions.
Thus the set of possible transitions is {t(l, 0, L), t(1, 0, R), t(O, 1, L), t(O, 1, R),
t(2, 0, L), t(2, 0, R), t(O, 2, L), t(O, 2, R), t(l, 1, L), t(l, 1, R)}.
By working methodically we can now calculate the pattern of access relations
involved in the problem. For each legal state, we try each transition, and see whether
the resulting state is legal. In this way we build up the information in the table below.
RE-REPRESENTATION AND EMERGENT INFORMATION 87
I initial state I by L-transitions I by R-transitions
s(O,O) s(O, 1), s(O, 2), s(1, 1)
s(O, 1) s(O,O) s(0,2),s(0,3~s(1,1)
s(0,2) s(O, 0), s(O, 1) 8(0,3),s(2,2)
s(0,3) s(O, 1), s(O, 2)
s(l,l) s(O, 0), s(O, 1) s(2,2), s(3, 1)
s(2,2) s(O, 2), s(l, 1) 8(3,2),s(3,3)
s(3,0) s(3, 1), s(3, 2)
s(3,1) s(l, 1), s(3,0) s(3, 2), s(3, 3)
s(3,2) 8(2,2), s(3, 0), s(3, 1) .s(3,3)
s(3,3) 8(2,2~s(3,1),s(3,2)

access between legal states

The table indicates that a given state in the left-hand column has access by single
L-transitions to those states in the middle column, and by single R-transitions to
those states in the right-hand column.
Again, to state the obvious, this new information is calculated and presented
using new concepts. The initial representation of the problem concerns missionaries,
cannibals, a river and a boat, while the above table concerns states and transitions.

4.4. FlRST RE-REPRESENTATION AND SOLUTION PROCEDURE

As a first re-representation of the problem, we can simply use our extra information
as stated in the table itself.
At each stage in constructing a route from our start state s( 3, 3) to our goal state
s(O, 0), we look up the states to which the current state has access. We follow the
constraint that we alternate between the L-transition and the R-transition columns
of the table (just as the boat alternates in direction of crossing). We record the states
visited, and we use this as a record of the route, and as data for checking for loops
in it. A loop occurs if transformations of the same type (L or R) are applied to the
same state twice, and in this case we backtrack to find alternative transitions. In
this way we discover that there are just four loop-free routes, each involving twel ve
states and eleven transitions, and so the answer to the problem is 'eleven'.

4.5. SECOND RE-REPRESENTATION AND SOLUTION PROCEDURE

Alternatively, we can represent this information in a diagram. Figure 3 is an adapta-


tion of the diagram given by Amarel (1968: 154).1 The diagram is read as follows:
the vertical and horizontal axes give the left bank's populations of missionaries and
cannibals respectively, hollow bullets indicate legal states, solid bullets indicate

1 The original diagram gives all states, but nol all transitions.
88 DONALD PETERSON

illegal states, hollow arrow heads indicate R-transitions, and filled arrow heads
indicate L-transitions (two arrow heads being placed on the same shaft).

Fig. 3. Diagrammatic re-representation of the Missionaries and Cannibals problem

The task is now to find the shortest sequence(s) of arrows of alternating type
from 3,3 to 0,0 in the diagram. This can be done by 'pivoting' on arrow heads of
the same type-that is, the type of arrow head used to enter a node should be the
same as that used to leave it.
The diagram expresses the same information as the preceding table, but it has
the advantage of utilising our capacities for spatial perception and reasoning. In
particular, using the diagram, it is very easy at each stage to perceive what moves
are possible.

4.6. ALTERNATIVE CONCEPTS

We now try alternative definitions of states and transitions such that the boat's
position is incorporated into the definition of a state rather than that of a transition.
When there are m missionaries and c cannibals on the left bank, and the boat is in
position b we notate this state as s( m, c, b). When a transition moves m missionaries
and c cannibals (in either direction), we notate this as 't( m, c)'. When there is a
transition from state S to state T, we say that S has access to T, and we notate this
as a(S, T). The access relation is symmetrical.

4.7. CALCULATION OF ALTERNATIVE EMERGENT INFORMATION

The set of possible transitions is {t( 1, 0), teO, 1), t(2, 0), teO, 2), t( 1,1), t( 1,1)}.
The set of legal states is {s(O, 0, L), s(O, 0, R), s(O, 1, L), s(O, 1, R), s(O, 2, L),
RE-REPRESENTATION AND EMERGENT INFORMATION 89
s(O, 2, R),s(O, 3, L), s(O, 3, R),s(I, 1, L), sCI, 1, R),s(2, 2, L), s(2, 2, R),s(3, 0,
L), s(3, 0, R), s(3, 1, L), s(3, 1, R), s(3, 2, L), s(3, 2, R), s(3, 3, L), s(3, 3, RH.
Again, by hand or computer we can calculate the pattern of access relations as
given in the table below. We note that four of the legal states do not appear in the
table, since they are not accessible from any other legal state.

a( s(3, 3, L), s(3, 2, R))


a(s(3,3,L),s(3,I,R))
a( s(3, 3, L), s(2, 2, R))
a(s(3, 2, L), s(3, 1, R))
a( s(3, 2, L), s(3, 0, R))
a( s(3, 2, L), s(2, 2, R))
a(s(3,I,L),s(3,0,R))
a(s(3, I,L),s(l, I,R))
a( s(2, 2, L), s(1, 1, R))
a( s(2, 2, L), s(O, 2, R))
a(s(I, 1, L),s(O, I,R))
a( sCI, 1, L), s(O, 0, R))
a( s(O, 3, L), s(O, 2, R))
a( s(O, 3, L), s(O, 1, R))
a( s(O, 2, L), s(O, 1, R))
a(s(0,2, L),s(O,O,R))
a(s(O,I,L),s(O,O,R))
access relations between legal states

4.8. THIRD RE-REPRESENTATION AND SOLUTION PROCEDURE


As a third re-representation of the problem, we can use the preceding table itself.
The task is to construct a route from state s(3, 3, L) to state s(O, 0, R). At each
stage we record the current state, find a state to which it has access, check that the
new state is different from the previously visited states, and continue. Since access
is symmetrical, the access relations in the table can be read in either 'direction'.

4.9. FOURTH RE-REPRESENTATION AND SOLUTION PROCEDURE

Alternatively, in the manner of Thomas (1974: 259),2 we can represent the same
information in a diagram as in Figure 4.
The task now is simply to find a route through the diagram from s(3, 3, L) to
s(O, 0, R). This is very easy, since what has been revealed is a surprisingly simple
structure, and in finding a route through it, only two choices are involved. The
diagram should be read topologically, and its 'u' shape is incidental.
2 The original diagram omits the state s(O, 1, L).
90 DONALD PETERSON

Fig. 4. alternative diagrammatic re-representation of the Missionaries and Cannibals prob-


lem

Again, the diagram expresses the same infonnation as does the preceding table,
but it allows our capacities for spatial perception and reasoning greater opportunity
to assist us. In particular, for each state, the accessible states correspond to the
spatially adjacent circles in the diagram. Spatial neighbourhood has been made to
reflect conceptual neighbourhood, and spatial neighbourhood is relatively easy to
perceive (see Freksa, 1991).

S. Discussion
S.l. USEFUL NEW CONCEPTS

It is of course true that our re-representations involve 'useful new concepts', though
a few qualifications to this are required.
The relevant concepts are useful in that they allow the generation of re-represent-
ations, which in turn render the problems much less complex to solve. To be fair,
the benefits of reduced searCh-space and reduced load on working memory have
to be balanced against the effort of producing a re-representation, and this effort
may seem especially worthwhile when the task has to be repeated often, or when
generalisation to a class of problems is achieved.
The relevant concepts are new in that they are not used in the initial problem
statements. However, they need not be altogether new, and certainly need not be
new to the person who devises the re-representation. In Boden's tenns, they need be
RE-REPRESENTATION AND EMERGENT INFORMATION 91

neither H-creative nor P-creative (Boden, 1990: 32). The application of the concepts
of states and transitions, for example, would come most easily to someone with a
knowledge of search techniques in artificial intelligence. What is 'new' about the
concepts in question, then, is their application to the problem at hand.

5.2. EMERGENT INFORMATION


The main point of the present chapter is that, at least for the cases considered, there
is more to re-representation than the application of useful new concepts. Our cases
reveal a common four-element pattern:
1. Initial representation.
2. Introduction of new concepts.
3. Calculation of emergent information.
4. Re-representation and solution procedure.
Once the new concepts have been introduced, the crucial emergent information
can be generated through calculation. The calculations involved in our cases are
straightforward, but the point is that they are calculations, not that they should be
complex calculations. And to make this point, simple Prolog programs were written
to perform them.
The result is information which is additional in the sense that it is not given
in the initial problem representation, and which is emergent in the sense that it is
formulated using concepts not present in the initial representation.
In the Four Knights case, we derive information about a circular structure in the
problem domain. In the Magic Square case, we derive information about levels,
completions, and possible allocations of numbers to squares which reveals a search
space much smaller than is initially evident. In the Missionaries and Cannibals
case, we derive information about accessibility between legal states, and in its
fourth re-representation we see that the problem has a surprisingly simple structure
(see Thomas, 1974: 260).
In each case what we have is information, rather than just concepts, since it is
propositional (hence it is right or wrong, and if we changed our re-representations
we could render them erroneous). This information constitutes a discovery which
increases our understanding of the problem_ The new concepts allow us to see
the problem differently, but calculation also allows us to see something, namely
emergent information about the structure of the problem.
It might be suggested that this information is implicit in the original represen-
tations of the problems, and that subsequent calculation merely draws out what is
already there. But this is to attribute too much power to calculation: the relevant
calculations require the new concepts, and they take as their input and output in-
formation formulated using those concepts. Thus their results are emergent rather
than implicit, relative to the Original representations.
Two final points are worth mentioning. First, the set of re-representations con-
sidered is not comprehensive, and alternative re-representations may be devised for
92 DONAlD PEI'ERSON

the same problems. Rouse Ball and Coxeter (1974: 193-221), for example, provide
several alternatives for Magic Squares. Second, what is given here is an abstract
(or 'conceptual' or 'logical') analysis which asserts that in the cases considered
the given elements are present. This is neutral regarding the manner of implemen-
tation of these elements, and to derive a concrete cognitive theory of the process
of re-representation or an algorithm for automated re-representation, it would be
necessary to decide, for example, what to assign to discrete modules, how these are
to interact, and how the elements of calculation and conceptual innovation are to
be achieved.

6. Conclusion
Problem solving through re-representation is a particular case of creative thought,
and we have addressed the question whether it is adequately characterised as 're-
description of problems under useful new concepts'. We answer the question in the
negative, since we find that in the cases considered the 'inspiration' of conceptual
change is partnered by the 'perspiration' of calculation, and the result is not just
re-description but emergent information.

Acknowledgements
For many useful comments and suggestions, I am very grateful to Ben du Boulay,
Terry Dartnall, Aaron Sloman, and in particular to Antoni Diller.

References
Amarel, S.: 1968, On representations of problems of reasoning about actions, in Michie, D. (ed.),
Machi1U! InteUigence 3, Edinburgh University Press, Edinburgh.
Boden, M.: 1987, ArtijiciallnteUigence and Natural Man, 2nd edn, MIT Press, London.
Boden, M.: 1990, The Creative Mind: Myths and Mechanisms, Weidenfeld and Nicolson, London.
Duncker, K.: 1945, On problem solving, Psychological Monographs, 58: 1-113.
Freksa, C.: 1991, Qualitative spatial reasoning, in Mark, D. M. and Frank, A. U. (eds), Cognitive and
Linguistic Aspects of Geographic Space, Kluwer Academic Publishers, Dordrecht.
Holyoak, K.: 1990, PIDblem solving, in OSherson, D. N. and Smith, E. E. (eds), Thinking, in the
series An Invitation to Cognitive Science, Vol. 3, MIT Press, Cambridge, Mass.
Laurrere, J-L.: 1990, Problem Solving and Artijiciallntelligence, trans. Howlett, J., Prentice Hall,
Hemel Hempstead.
Ohlsson S.: 1984, Restructuring Revisited [: Summary and critique of the Gestalt theory of problem
solving, Scandinavian Journal ofPsychology, 25: 65-78.
Polya, G.: 1948, How To Solve It: A New Aspect of Mathematical Method, Doubleday, New York.
Rouse Ball, W. W. and Coxeter, H. S. M.: 1974, Mathematical Recreations and Essays, 12th edition,
University of Toronto Press, Toronto.
Sloman, A.: 1989, A personal view of artificial intelligence, in Sharples, M. et. al. (eds), Computers
and Thought: a Practical Introduction to Artijicial IntelUgence, MIT/Bradford, Cambridge, Mass.
Thomas J .C.: 1974, An analysis of behaviour in the Hobbits-Orcs problem, Cognitive Psychology,
6: 257-269.

You might also like