The Fabric of Reality David Deutch

particular relationship to it. Anything or everything that we perceive might be

---

an illusion or a dream. Illusions and dreams are, after all, common.
Solipsism, the theory that only one mind exists and that what appears to be
external reality is only a dream taking place in that mind, cannot be logically
disproved. Reality 
might consist of one person, presumably you, dreaming a
lifetime’s experiences. Or it might consist of just you and me. Or just the
planet Earth and its inhabitants. And if we dreamed evidence — any
evidence — of the existence of other people, or other planets, or other
universes, that would prove nothing about how many of those things there
really are.
Since solipsism, and an infinity of related theories, are logically consistent
with your perceiving any possible observational evidence, it follows that you
can logically deduce nothing about reality from observational evidence. How,
then, could I say that the observed behaviour of shadows ‘rules out’ the
theory that there is only one universe, or that eclipse observations make the
Newtonian world-view ‘rationally untenable’? How can that be so? If ‘ruling
out’ does not mean ‘disproving’, what does it mean? Why should we feel
compelled to change our world-view, or indeed any opinion at all, on account
of something being ‘ruled out’ in that sense? This critique seems to cast
doubt on the whole of science — on any reasoning about external reality that
appeals to observational evidence. If scientific reasoning does not amount to
sequences of logical deductions from the evidence, what does it amount to?
Why should we accept its conclusions?
This is known as the ‘problem of induction’. The name derives from what
was, for most of the history of science, the prevailing theory of how science
works. The theory was that there exists, short of mathematical proof, a
lesser but still worthy form of justification called 
induction. Induction was
contrasted, on the one hand, with the supposedly perfect justification
provided by deduction, and on the other hand with supposedly weaker
philosophical or intuitive forms of reasoning that do not even have
observational evidence to back them up. In the inductivist theory of scientific
knowledge, observations play two roles: first, in the discovery of scientific
theories, and second, in their justification. A theory is supposed to be
discovered by ‘extrapolating’ or ‘generalizing’ the results of observations.
Then, if large numbers of observations conform to the theory, and none
deviates from it, the theory is supposed to be justified — made more
believable, probable or reliable. The scheme is illustrated in Figure 3.1.
The inductivist analysis of my discussion of shadows would therefore go
something like this: ‘We make a series of observations of shadows, and see
interference phenomena (stage 1). The results conform to what would be
expected if there existed parallel universes which affect one another in
certain ways. But at first no one notices this. Eventually (stage 2) someone
forms the generalization that interference will 
always be observed under the
given circumstances, and thereby induces the theory that parallel universes
are responsible. With every further observation of interference (stage 3) we
become a little more convinced of that theory. After a sufficiently long
sequence of such observations, and provided that none of them ever
contradicts the theory, we conclude (stage 4) that the theory is true.
Although we can never be absolutely sure, we are for practical purposes
convinced.’

---

It is hard to know where to begin in criticizing the inductivist conception of
science — it is so profoundly false in so many different ways. Perhaps the
worst flaw, from my point of view, is the sheer 
non sequitur that a
generalized prediction is tantamount to a new theory. Like all scientific
theories of any depth, the theory that there are parallel universes simply
does not have the form of a generalization from the observations. Did we
observe first one universe, then a second and a third, and then induce that
there are trillions of them? Was the generalization that planets will ‘wander’
round the sky in one pattern rather than another, equivalent to the theory
that planets are worlds, in orbit round the Sun, and that the Earth is one of
them? It is also not true that repeating our observations is the way in which
we become convinced of scientific theories. As I have said, theories are
explanations, not merely predictions. If one does not accept a proposed
explanation of a set of observations, making the observations over and over
again is seldom the remedy. Still less can it help us to create a satisfactory
explanation when we cannot think of one at all.
FIGURE 3.1 
The inductivist scheme.
Furthermore, even mere predictions can never be justified by observational
evidence, as Bertrand Russell illustrated in his story of the chicken. (To
avoid any possible misunderstanding, let me stress that this was a
metaphorical, anthropomorphic chicken, representing a human being trying
to understand the regularities of the universe.) The chicken noticed that the
farmer came every day to feed it. It predicted that the farmer would continue
to bring food every day. Inductivists think that the chicken had ‘extrapolated’
its observations into a theory, and that each feeding time added justification
to that theory. Then one day the farmer came and wrung the chicken’s neck.
The disappointment experienced by Russell’s chicken has also been
experienced by trillions of other chickens. This inductively justifies the
conclusion that induction cannot justify any conclusions!
However, this line of criticism lets inductivism off far too lightly. It does
illustrate the fact that repeated observations cannot 
justify theories, but in
doing so it entirely misses (or rather, accepts) a more basic misconception:
namely, that the inductive extrapolation of observations to 
form new theories
is even possible. In fact, it is impossible to extrapolate observations unless
one has already placed them within an explanatory framework. For example,
in order to ‘induce’ its false prediction, Russell’s chicken must first have had
in mind a false explanation of the farmer’s behaviour. Perhaps it guessed
that the farmer harboured benevolent feelings towards chickens. Had it
guessed a different explanation — that the farmer was trying to fatten the
chickens up for slaughter, for instance — it would have ‘extrapolated’ the
behaviour differently. Suppose that one day the farmer starts bringing the
chickens more food than usual. How one extrapolates this new set of
observations to predict the farmer’s future behaviour depends entirely on
how one explains it. According to the benevolent-farmer theory, it is

---

evidence that the farmer’s benevolence towards chickens has increased,
and that therefore the chickens have even less to worry about than before.
But according to the fattening-up theory, the behaviour is ominous — it is
evidence that slaughter is imminent.
The fact that the same observational evidence can be ‘extrapolated’ to give
two diametrically opposite predictions according to which explanation one
adopts, and cannot justify either of them, is not some accidental limitation of
the farmyard environment: it is true of all observational evidence under all
circumstances. Observations could not possibly play either of the roles
assigned to them in the inductivist scheme, even in respect of mere
predictions, let alone genuine explanatory theories. Admittedly, inductivism is
based on the common-sense theory of the growth of knowledge — that we
learn from experience — and historically it was associated with the liberation
of science from dogma and tyranny. But if we want to understand the true
nature of knowledge, and its place in the fabric of reality, we must face up to
the fact that inductivism is false, root and branch. No scientific reasoning,
and indeed no successful reasoning of any kind, has ever fitted the
inductivist description.
What, then, 
is the pattern of scientific reasoning and discovery? We have
seen that inductivism and all other prediction-centred theories of knowledge
are based on a misconception. What we need is an explanation-centred
theory of knowledge: a theory of how explanations come into being and how
they are justified; a theory of how, why and when we should allow our
perceptions to change our world-view. Once we have such a theory, we
need no separate theory of predictions. For, given an explanation of some
observable phenomenon, it is no mystery how one obtains predictions. And
if one has justified an explanation, then any predictions derived from that
explanation are automatically justified too.
Fortunately, the prevailing theory of scientific knowledge, which in its modern
form is due largely to the philosopher Karl Popper (and which is one of my
four ‘main strands’ of explanation of the fabric of reality), can indeed be
regarded as a theory of explanations in this sense. It regards science as a
problem-solving process. Inductivism regards the catalogue of our past
observations as a sort of skeletal theory, supposing that science is all about
filling in the gaps in that theory by interpolation and extrapolation. Problem-
solving does begin with an inadequate theory — but not with the notional
‘theory’ consisting of past observations. It begins with our best existing
theories. When some of those theories seem inadequate to us, and we want
new ones, that is what constitutes a 
problem. Thus, contrary to the
inductivist scheme shown in Figure 3.1, scientific discovery need not begin
with observational evidence. But it does always begin with a problem. By a
‘problem’ I do not necessarily mean a practical emergency, or a source of
anxiety. I just mean a set of ideas that seems inadequate and worth trying to
improve. The existing explanation may seem too glib, or too laboured; it may
seem unnecessarily narrow, or unrealistically ambitious. One may glimpse a
possible unification with other ideas. Or a satisfactory explanation in one
field may appear to be irreconcilable with an equally satisfactory explanation
in another. Or it 
may be that there have been some surprising observations
— such as the wandering of planets — which existing theories did not
predict and cannot explain.

---

This last type of problem resembles stage 1 of the inductivist scheme, but
only superficially. For an unexpected observation never initiates a scientific
discovery unless the pre-existing theories already contain the seeds of the
problem. For example, clouds wander even more than planets do. This
unpredictable wandering was presumably familiar long before planets were
discovered. Moreover, predicting the weather would always have been
valuable to farmers, seafarers and soldiers, so there would always have
been an incentive to theorize about how clouds move. Yet it was not
meteorology that blazed the trail for modern science, but astronomy.
Observational evidence about meteorology was far more readily available
than in astronomy, but no one paid much attention to it, and no one induced
any theories from it about cold fronts or anticyclones. The history of science
was not crowded with disputes, dogmas, heresies, speculations and
elaborate theories about the nature of clouds and their motion. Why?
Because under the established explanatory structure for weather, it was
perfectly comprehensible that cloud motion should be unpredictable.
Common sense suggests that clouds move with the wind. When they drift in
other directions, it is reasonable to surmise that the wind can be different at
different altitudes, and is rather unpredictable, and so it is easy to conclude
that there is no more to be explained. Some people, no doubt, took this view
about planets, and assumed that they were just glowing objects on the
celestial sphere, blown about by high-altitude winds, or perhaps moved by
angels, and that there was no more to be explained. But others were not
satisfied with that, and guessed that there were deeper explanations behind
the wandering of planets. So they searched for such explanations, and found
them. At various times in the history of astronomy there appeared to be a
mass of unexplained observational evidence; at other times only a scintilla,
or none at all. But always, if people had chosen what to theorize about
according to the cumulative number of observations of particular
phenomena, they would have chosen clouds rather than planets. Yet they
chose planets, and for diverse reasons. Some reasons depended on
preconceptions about how cosmology ought to be, or on arguments
advanced by ancient philosophers, or on mystical numerology. Some were
based on the physics of the day, others on mathematics or geometry. Some
have turned out to have objective merit, others not. But every one of them
amounted to this: it seemed to someone that the existing explanations could
and should be improved upon.
One solves a problem by finding new or amended theories, containing
explanations which do not have the deficiencies, but do retain the merits, of
existing explanations (Figure 3.2). Thus, after a problem presents itself
(stage 1), the next stage always involves 
conjecture: proposing new
theories, or modifying or reinterpreting old ones, in the hope of solving the
problem (stage 2). The conjectures are then 
criticized which, if the criticism
is rational, entails examining and comparing them to see which offers the
best explanations, according to the criteria inherent in the problem (stage 3).
When a conjectured theory fails to survive criticism — that is, when it
appears to offer worse explanations than other theories do — it is
abandoned. If we find ourselves abandoning one of our originally held
theories in favour of one of the newly proposed ones (stage 4), we
tentatively deem our problem-solving enterprise to have made progress. I
say ‘tentatively’, because subsequent problem-solving will probably involve

---

altering or replacing even these new, apparently satisfactory theories, and
sometimes even resurrecting some of the apparently unsatisfactory ones.
Thus the solution, however good, is not the end of the story: it is a starting-
point for the next problem-solving process (stage 5). This illustrates another
of the misconceptions behind inductivism. In science the object of the
exercise is not to find a theory that will, or is likely to, be deemed true for
ever; it is to find the best theory available now, and if possible to improve on
all available theories. A scientific argument is intended to persuade us that a
given explanation is the best one available. It does not and could not say
anything about how that explanation will fare when, in the future, it is
subjected to new types of criticism and compared with explanations that
have yet to be invented. A good explanation may make good predictions
about the future, but the one thing that no explanation can even begin to
predict is the content or quality of its own future rivals.
FIGURE 3.2 
The problem-solving process.
What I have described so far applies to all problem-solving, whatever the
subject-matter or techniques of rational criticism that are involved. 
Scientific
problem-solving always includes a particular method of rational criticism,
namely 
experimental testing. Where two or more rival theories make
conflicting predictions about the outcome of an experiment, the experiment
is performed and the theory or theories that made false predictions are
abandoned. The very construction of scientific conjectures is focused on
finding explanations that have experimentally testable predictions. Ideally we
are always seeking 
crucial experimental tests — experiments whose
outcomes, whatever they are, will falsify one or more of the contending
theories. This process is illustrated in Figure 3.3. Whether or not
observations were involved in the instigating problem (stage 1), and whether
or not (in stage 2) the contending theories were specifically designed to be
tested experimentally, it is in this critical phase of scientific discovery (stage
3) that experimental tests play their decisive and characteristic role. That role
is to render some of the contending theories unsatisfactory by revealing that
their explanations lead to false predictions. Here I must mention an
asymmetry which is important in the philosophy and methodology of science:
the asymmetry between experimental refutation and experimental
confirmation. Whereas an incorrect prediction automatically renders the
underlying explanation unsatisfactory, a correct prediction says nothing at all
about the underlying explanation. Shoddy explanations that yield correct
predictions are two a penny, as UFO enthusiasts, conspiracy-theorists and
pseudo-scientists of every variety should (but never do) bear in mind.
If a theory about observable events is untestable — that is, if no possible
observation would rule it out — then it cannot by itself explain why those
events happen in the way they are observed to and not in some other way.
For example, the ‘angel’ theory of planetary motion is untestable because no
matter how planets moved, that motion could be attributed to angels;

---

therefore the angel theory cannot explain the particular motions that we see,
unless it is supplemented by an independent theory of how angels move.
That is why there is a methodological rule in science which says that once
an experimentally testable theory has passed the appropriate tests, any 
less
testable rival theories about the same phenomena are summarily rejected,
for their explanations are bound to be inferior. This rule is often cited as
distinguishing science from other types of knowledge-creation. But if we take
the view that science is about explanations, we see that this rule is really a
special case of something that applies naturally to all problem-solving:
theories that are capable of giving more detailed explanations are
automatically preferred. They are preferred for two reasons. One is that a
theory that ‘sticks its neck out’ by being more specific about more
phenomena opens up itself and its rivals to more forms of criticism, and
therefore has more chance of taking the problem-solving process forward.
The second is simply that, if such a theory survives the criticism, it leaves
less unexplained — which is the object of the exercise.
FIGURE 3.3 
The course of scientific discovery.
I have already remarked that even in science most criticism does not consist
of experimental testing. That is because most scientific criticism is directed
not at a theory’s predictions but directly at the underlying explanations.
Testing the predictions is just an indirect way (albeit an exceptionally
powerful one, when available) of testing the explanations. In Chapter 1, I
gave the example of the ‘grass cure’ — the theory that eating a kilogram of
grass is a cure for the common cold. That theory and an infinity of others of
the same ilk are readily testable. But we can criticize and reject them without
bothering to do any experiments, purely on the grounds that they explain no
more than the prevailing theories which they contradict, yet make new,
unexplained assertions.
The stages of a scientific discovery shown in Figure 3.3 are seldom
completed in sequence at the first attempt. There is usually repeated
backtracking before each stage is completed — or rather, 
solved, for each
stage may present a problem whose solution itself requires all five stages of
a subsidiary problem-solving process. This applies even to stage 1, for the
initiating problem itself is not immutable. If we cannot think of good
candidate solutions we may return to stage 1 and try to reformulate the
problem, or even choose a different problem. Indeed, apparent insolubility is
only one of many reasons why we often find it desirable to modify problems
we are solving. Some variants of a problem are inevitably more interesting,
or more relevant to other problems; some are better formulated; some seem
to be potentially more fruitful, or more urgent — or whatever. In many cases
the issue of what precisely the problem is, and what the attributes of a ‘good’
explanation would be, receive as much criticism and conjecture as do trial
solutions.

---

Similarly, if our criticisms at stage 3 fail to distinguish between rival theories,
we try to invent new methods of criticism. If that does not seem to work we
may backtrack to stage 2 and try to sharpen our proposed solutions (and
existing theories) so as to get more explanations and predictions out of them
and make it easier to find fault with them. Or we may again backtrack to
stage 1 and try to find better criteria for the explanations to meet. And so on.
Not only is there constant backtracking, but the many sub-problems all
remain simultaneously active and are addressed opportunistically. It is only
when the discovery is complete that a fairly sequential argument, in a pattern
something like Figure 3.3, can be presented. It can begin with the latest and
best version of the problem; then it can show how some of the rejected
theories fail criticism; then it can set out the winning theory, and say why it
survives criticism; then it can explain how one copes without the superseded
theory; and finally it can point out some of the new problems that this
discovery creates or allows for.
While a problem is still in the process of being solved we are dealing with a
large, heterogeneous set of ideas, theories, and criteria, with many variants
of each, all competing for survival. There is a continual turnover of theories
as they are altered or replaced by new ones. So all the theories are being
subjected to 
variation and selection, according to criteria which are
themselves subject to variation and selection. The whole process resembles
biological evolution. A problem is like an ecological niche, and a theory is like
a gene or a species which is being tested for viability in that niche. Variants
of theories, like genetic mutations, are continually being created, and less
successful variants become extinct when more successful variants take
over. ‘Success’ is the ability to survive repeatedly under the selective
pressures — criticism — brought to bear in that niche, and the criteria for
that criticism depend partly on the physical characteristics of the niche and