I've often seen fallibilism discussed with reference to the Münchhausen trilemma and the problem of scientific induction. There is a simpler and more encompassing model of fallibilism that I might call "mechanistic fallibility."

According to mechanistic fallibility, the human brain is a machine made of neurons and so forth, and all machines are at least slightly fallible due to a variety of hardware defects, software defects, and environmental interference. Therefore the human brain is fallible. Therefore, the conclusions of a human brain cannot be trusted with perfect certainty. This holds even if the brain in question is your own, and even on a priori propositions.

The argument also holds up even if we allow the possibility that the human brain is not a machine or that the mind is somehow independent of it. Because, as long as this is merely one possibility and not an absolute certainty, the chance the brain is a machine is still nonzero. Thus the chance that the brain is a machine and the machine has just made an error is also always nonzero.

In practice we observe that humans do often make mistakes on a priori propositions; arithmetic errors, for example, are such mistakes. We make errors because of inattention, the inherent randomness of neuron operation, micro-strokes, and so on. Computers sometimes make arithmetic errors too, for a variety of hardware and software reasons. So, it is not rational to perfectly trust the conclusions of computers or humans, including yourself, on any proposition.

This holds no matter how thoroughly the human or computer checks their work. There could always be a mechanistic failure anywhere in the error-checking process - in all stages of checking, or just the last and most critical stage.

What philosopher is best associated with this type of mechanistic fallibilism?

Is any philosopher strongly associated with it? I would now accept as an answer a philosopher who emphasized the importance of accidental mistakes in calculation, with the conclusion that we cannot be certain of anything, even if they did not use a direct comparison to machines.

  • 1
    There is no need to assume that brain or mind is a machine, or anything else in particular, for this type of argument. Machines are not the only things prone to glitches, so one can remain metaphysically neutral. Peirce, the founder of fallibilism, for example, says of 6x7=42 that "it is conceivable that this proposition should be a mistake due to some peculiar insanity affecting the whole of human race" CP 5.125 (although he did not think it very likely). Münchhausen trilemma route is more associated with Popper.
    – Conifold
    Aug 30, 2021 at 3:37
  • @Conifold It's a little harder to accept that some peculiar insanity has affected everyone, than to accept that the brain is a machine with parts (such as neurons) that sometimes break down and don't operate properly for various reasons, ranging from inattention to the stochastic nature of neurons to micro-strokes. The latter gives a perfectly mundane reason why the brain could and does sometimes break down, that is not only compatible with the worldview of modern science but a necessary consequence of it.
    – causative
    Aug 30, 2021 at 3:39
  • Neurons breaking down is just one potential mechanism for the "peculiar insanity". You get a stronger argument by not tying yourself to any particular mechanism, human fallibility is commonly acknowledged by idealists and materialists alike, for whatever reasons. And you'll still need it to affect everyone in the same place, to explain why they all made the same mistake .
    – Conifold
    Aug 30, 2021 at 3:49
  • @Conifold it's a potential mechanism that science says is the actual mechanism. It's relatively easy to just wave your hands and dismiss some vague "peculiar insanity" as impossible - it's less easy to say that your neurons could never suffer faults in operation, which science says they do. No, the faults do not need to affect everyone; if I think "X is true and everyone says X is true", when X is false, then perhaps the fault is only in my own head. Only my perception of consensus need be wrong.
    – causative
    Aug 30, 2021 at 4:07
  • I think you are right in this sense. People exhibit so-called conjunction bias: when presented with a specific and a general scenario (that covers the specific one) they often find the specific one "more probable". So presenting the neuron mechanism may indeed make it easier to believe psychologically. But it does not make it more probable. And when it comes to justification, the strength of the argument is more salient than its believability.
    – Conifold
    Aug 30, 2021 at 4:24

1 Answer 1


There's a clash of principles. The kind of problems you talk about here are ones of degree, like in science. We can never be completely certain in a scientific result, all results are tentative, dependent on what we know so far. That doesn't stop us doing science, and getting to better and more certain knowledge. Tools like consilience, convergence of evidence, the verification by more than one process or approach to the same thing, like verification of the experienced by more than one sense, help.

All of science says "not to perfectly trust the conclusions of computers or humans, including yourself, on any proposition". See Cartwright's 'How The Laws Of Physics Lie'.

What is more interesting than easily checkable errors you talk about, are systemic errors, that might relate to human cognition. We have no access to events completely unfiltered by human minds. This could impact how we understand time for instance, with memory only working in one direction, but time (or something) moving forwards and backwards. The really radical idea, is we may be Boltzmann brains, just popped into existence with illusory memories.

The Munchausen trilemma is about the logical basis, of methodologies not being able to justify themselves, ie in this case, Hume's priblem of induction. The trilemma covers the basis of math as well induction.

What it says is no surprise, we can see why from Godel's work: a system can't evaluate the truth of all the statements that can be made within it. And the solution in practice, is Hofstadter strange loops and tangled hierarchies - basically coherentism. We can see that described by David Deutsch's Fabric of Reality, with four approaches to knowledge, information theory, quantum mechanics, evolution & epistemology, all leaning on each other and providing a basis for what we know in the other three domains by distinct methodology of the domain.

Donald Hoffman writes on The Evolutionary Argument Against Reality. That is, why we can't rely on evolution to give us the most accurate picture of reality, or even to converge towards it. But, we can learn to correct for cognitive biases, and develop other tools to improve on evolution, which is only concerned with replication of replicators (we have the noosphere). See: Is the idea of a causal chain physical (or even scientific)?

  • "Easily checkable errors"? But there could also be a mechanistic mistake in their checking! No error correcting mechanism can ever attain 100% protection against errors, as there could be an error in any or all stages of the error correction.
    – causative
    Aug 30, 2021 at 17:16
  • @causative: That wasn't my point. You list incorrectly carried out things, which checking the process or repeating observations will reduce. There are bigger problems with inferences: en.wikipedia.org/wiki/Statistical_hypothesis_testing These extend into issues of not knowing what we don't know, & systematic errors.
    – CriglCragl
    Aug 30, 2021 at 17:51
  • I mentioned scientific induction in the OP to distinguish that kind of problem from mechanistic fallibility (i.e. problems of scientific induction are not on topic). Scientific induction is less encompassing: it applies only to problems of observing and formulating models of the physical world, where mechanistic fallibility can apply to anything. The Munchausen trilemma also is narrow: it applies only to the justification of the axioms, and says nothing about the validity of deductions, assuming the axioms are correct. Mechanistic fallibility says any of that could be wrong.
    – causative
    Aug 30, 2021 at 18:03
  • 1
    @causative If greater plausibility is the selling point the problem you have is that mechanistic failures you describe do not plausibly produce persistent errors. Yes, there can be glitches in checking, rechecking and checking of checking, but the probability quickly dwindles down with each step. This sort of fallibility is pragmatically irrelevant. We need a reason for glitches to systematically conspire to produce recurring errors. For contrast, cognitive biases that do that have much more complex genesis than accumulation of mechanistic glitches.
    – Conifold
    Aug 31, 2021 at 9:22
  • @Conifold It is not necessary for the errors to be persistent; I may think, "7+9=15 and I have always known this." But the error may only be in the present; I may be mistaken both in thinking 7+9=15, and in thinking i have always known it. However, the more salient point is that we are speaking of the plausibility of a chance of error, and we only care whether this chance is zero or nonzero. If the chance of error is nonzero, even if it is very small, then we cannot rationally have perfect 100% certainty. We may rationally have very high certainty, but that is not the same thing.
    – causative
    Aug 31, 2021 at 16:06

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .