More on this book
Community
Kindle Notes & Highlights
Read between
September 18 - October 29, 2025
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).
...more
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.
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.
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.
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.
The new world-view that may be implicit in a theory that solves a problem, and the distinctive features of a new species that takes over a niche, are emergent properties of the problem or niche. In other words, obtaining solutions is inherently complex.
Evolution, or trial and error – especially the focused, purposeful form of trial and error called scientific discovery – are the only ways.
in biology variations (mutations) are random, blind and purposeless, while in human problem-solving the creation of new conjectures is itself a complex, knowledge-laden process driven by the intentions of the people concerned. Perhaps an even more important difference is that there is no biological equivalent of argument. All conjectures have to be tested experimentally, which is one reason why biological evolution is slower and less efficient by an astronomically large factor.
Both in science and in biological evolution, evolutionary success depends on the creation and survival of objective knowledge, which in biology is called adaptation. That is, the ability of a theory or gene to survive in a niche is not a haphazard function of its structure but depends on whether enough true and useful information about the niche is implicitly or explicitly encoded there.
We never draw inferences from observations alone, but observations can become significant in the course of an argument when they reveal deficiencies in some of the contending explanations. We choose a scientific theory because arguments, only a few of which depend on observations, have satisfied us (for the moment) that the explanations offered by all known rival theories are less true, less broad or less deep.
Inductivism supposes that theories are somehow extracted or distilled from observations, or are justified by them, whereas in fact theories begin as unjustified conjectures in someone’s mind, which typically precede the observations that rule out rival theories.
Theories arrive from the presence of problems, which is an absence of understanding and a desire for it. This may come in the form of an observation, but the theory is not derived from the observation then. It is from the problem that comes from lack of understanding about the observed phenomenon, which leads to conjecture and then criticism of the theory, which might include further observations or a critical experiment as part of that criticism.
When we succeed in solving a problem, scientific or otherwise, we end up with a set of theories which, though they are not problem-free, we find preferable to the theories we started with. What new attributes the new theories will have therefore depends on what we saw as the deficiencies in our original theories – that is, on what the problem was.
The better theories are connected to the last best theory in that they are solutions to the problems present in that last best theory. They’re then defined by the last best theory. Additionally, every best theory we have contains some errors, always.
A problem exists when it seems that some of our theories, especially the explanations they contain, seem inadequate and worth trying to improve.
The purpose of science is to understand reality through explanations. The characteristic (though not the only) method of criticism used in science is experimental testing.
We seek explanations when we encounter a problem with existing ones. We then embark on a problem-solving process. New explanatory theories begin as unjustified conjectures, which are criticized and compared according to the criteria inherent in the problem. Those that fail to survive this criticism are abandoned. The survivors become the new prevailing theories, some of which are themselves problematic and so lead us to seek even better explanations. The whole process resembles biological evolution.
Noticing problems required first a world view—having a collection of explanations for how things are in the world. Without them, you won’t notice how things could be better, or rather where things are lacking. When your best explanations fail, that’s the opportunity. How you define the problem in that instance determines the shape of the explanations that you can come up with.
Ad hoc modifications had been proposed to improve the accuracy of the geocentric theory, and it was hard to quantify the predictive powers of the two rival theories.
Where there are ad hoc modifications to a theory, there is a gap in knowledge—a better explanation waiting in plain sight.
if making the right predictions were our only constraint, we could invent theories which say that anything we please is going on
It may be an absurd theory, but the point is that it cannot be ruled out by experiment. Nor is it valid to rule out any theory solely on the grounds that it is ‘absurd’:
If the Inquisition had seriously tried to understand the world in terms of the theory they tried to force on Galileo, they would also have understood its fatal weakness, namely that it fails to solve the problem it purports to solve. It does not explain planetary motions ‘without having to introduce the complication of the heliocentric system’.
Perhaps the ad hoc modifications to an explanation are closer to the truth; may try removing the underlying explanation and working from the ad hoc modifications alone to see what problems they address in the prior, underlying explanation and then build a new explanation with them.
Explanations are not justified by the means by which they were derived; they are justified by their superior ability, relative to rival explanations, to solve the problems they address.
A prediction, or any assertion, that cannot be defended might still be true, but an explanation that cannot be defended is not an explanation.
Not only do explanations change, but our criteria and ideas about what should count as an explanation are gradually changing (improving) too. So the list of acceptable modes of explanation will always be open-ended, and consequently the list of acceptable criteria for reality must be open-ended too.
the criterion for reality that is used in science, namely, if something can kick back, it exists.
It is enough that when we ‘kick’ something, the object affects us in ways that require independent explanation.
We must adopt a methodological rule that if something behaves as if it existed, by kicking back, then one regards that as evidence that it does exist. Shadow photons kick back by interfering with the photons that we see, and therefore shadow photons exist.
Similarly, to postulate that angels come through the other slits and deflect our photons would be better than nothing. But we can do better than that. We know exactly how those angels would have to behave: very much like photons. So we have a choice between an explanation in terms of invisible angels pretending to be photons, and one in terms of invisible photons. In the absence of an independent explanation for why angels should pretend to be photons, that latter explanation is superior.
Though you cannot disprove the presence of angels as an explanation of phenomenon, the lack of explanation for the angels behavior leaves a gap. A better explanation is then one that doesn’t involve angels whose behavior cannot he explained and does involve only those things whose behavior can be explained. It’s that old saying, “looks like a duck; quacks like a duck” where must be a duck is the simplest explanation that is wholly explained. It could also be a robot disguised as a duck with an explanation of how looks and sounds like a duck. We couldn’t know and both explanations would be sufficient for answering the problem, “what is it?”
Whenever I have used Dr Johnson’s criterion to argue for the reality of something, one attribute in particular has always been relevant, namely complexity. We prefer simpler explanations to more complex ones. And we prefer explanations that are capable of accounting for detail and complexity to explanations that can account only for simple aspects of phenomena. Dr Johnson’s criterion tells us to regard as real those complex entities which, if we did not regard them as real, would complicate our explanations.
Thus the observed complexity in the structure or behaviour of an entity is part of the evidence that that entity is real. But it is not sufficient evidence. We do not, for example, deem our reflections in a mirror to be real people. Of course, illusions themselves are real physical processes. But the illusory entities they show us need not be considered real, because they derive their complexity from somewhere else. They are not autonomously complex.
If, according to the simplest explanation, an entity is complex and autonomous, then that entity is real.
If the explanation explains the complexity of the system and in terms of the system as the source of that complexity, then it is a real thing.
a complex process is one that in effect presents us with the results of a substantial computation.
Complex because it requires more information to represent the thing that is being computed—that the computation whose output is a representation of that thing.
If a substantial amount of computation would be required to give us the illusion that a certain entity is real, then that entity is real.
Defining real by how much information is required to simulate the thing that exists—to compute it. The harder to compute, the more complex, the more real.
There are degrees of real then?
a story that is in effect about shadow photons necessarily appears in any explanation of the observed effects. The irreducible complexity of that story makes it philosophically untenable to deny that the objects exist.
Without the story of the shadow photons, there is no explanation sufficient so as to compute—to simulate—the result. The complex story of shadow photons necessitates that they exist—that they are as real as anything else sufficiently complex—because their complexity is required to compute the observed outcome.
But what is genuinely out there is evidence, or, more precisely, a reality that will respond with evidence if we interact appropriately with it. Given a shred of a theory, or rather, shreds of several rival theories, the evidence is available out there to enable us to distinguish between them. Anyone can search for it, find it and improve upon it if they take the trouble. They do not need authorization, or initiation, or holy texts. They need only be looking in the right way — with fertile problems and promising theories in mind. This open accessibility, not only of evidence but of the whole
...more
The more fundamental a theory is, the more readily available is the evidence that bears upon it (to those who know how to look), not just on Earth but throughout the multiverse.
The very existence of general, explanatory theories implies that disparate objects and events are physically alike in some ways.
Thus reality contains not only evidence, but also the means (such as our minds, and our artefacts) of understanding it.
To the extent that these symbols, images and theories are true — that is, they resemble in appropriate respects the concrete or abstract things they refer to — their existence gives reality a new sort of self-similarity, the self-similarity we call knowledge.
Occam’s razor (My formulation) Do not complicate explanations beyond necessity, because if you do, the unnecessary complications themselves will remain unexplained.
self-similarity Some parts of physical reality (such as symbols, pictures or human thoughts) resemble other parts. The resemblance may be concrete, as when the images in a planetarium resemble the night sky; more importantly, it may be abstract, as when a statement in quantum theory printed in a book correctly explains an aspect of the structure of the multiverse.
Real entities behave in a complex and autonomous way, which can be taken as the criterion for reality: if something ‘kicks back’, it exists.
Scientific reasoning, which uses observation not as a basis for extrapolation but to distinguish between otherwise equally good explanations, can give us genuine knowledge about reality.
Computers are physical objects, and computations are physical processes. What computers can or cannot compute is determined by the laws of physics alone, and not by pure mathematics.
One of the most important concepts of the theory of computation is universality. A universal computer is usually defined as an abstract machine that can mimic the computations of any other abstract machine in a certain well-defined class. However, the significance of universality lies in the fact that universal computers, or at least good approximations to them, can actually be built, and can be used to compute not just each other’s behaviour but the behaviour of interesting physical and abstract entities. The fact that this is possible is part of the self-similarity of physical reality
Thus the laws of physics impose no limit on the range and accuracy of image generators. There is no possible sensation, or sequence of sensations, that human beings are capable of experiencing that could not in principle be rendered artificially.
In Deutsch’s conception, by direct stimulation of the brain—since all sense organs lead back to synapses firing and signal interpretation by the brain
At present, progress in such technology is all about inventing more diverse and more accurate ways of stimulating sense organs. But that class of problems will disappear once we have cracked the codes used by our sense organs, and developed a sufficiently delicate technique for stimulating nerves. Once we can artificially generate nerve signals accurately enough for the brain not to be able to perceive the difference between those signals and the ones that our sense organs would send, increasing the accuracy of this technique will no longer be relevant.
From the foregoing discussion it seems that any virtual-reality generator must have at least three principal components: a set of sensors (which may be nerve-impulse detectors) to detect what the user is doing, a set of image generators (which may be nerve-stimulation devices), and a computer in control.
Thus virtual-reality generators of the future would be better described as having only one principal component, a computer, together with some trivial peripheral devices.
Specifying a virtual-reality environment does not mean specifying what the user will experience, but rather specifying how the environment would respond to each of the user’s possible actions.
How do we recognize unpredictable environments, and how do we confirm that purportedly random numbers are distributed fairly? We check whether a rendering of a roulette wheel meets its specifications in the same way that we check whether the real thing does: by kicking (spinning) it, and seeing whether it responds as advertised.
A random number generator as the only reliable yoo for falsifiable tests of reality, of an inaccurately rendered object or environment. It doesn’t kick back as expected. You can never confirm that an environment is rendered accurately; you can only falsify the expected behavior.
a short-sighted view of science is that it is all about predicting our sense-impressions. The correct view is that, while sense-impressions always play a role, what science is about is understanding the whole of reality, of which only an infinitesimal proportion is ever experienced.
