Patterns, Similarity and Relevance

Elliot Temple
,
1

One of the really hard problems in epistemology is about patterns. Which patterns can be extrapolated more widely? What situations do they apply to? What situations are too different so they don't apply? Usually, there are lots of other situations or scenarios where a pattern is a pretty good or decent fit, but not a perfect fit. So then you have to consider how good of a fit is good enough. And that's hard.

This is a companion discussion topic for the original entry at https://criticalfallibilism.com/patterns-similarity-and-relevance/
1 Like
2
My outline
  • Difficult questions about patterns:
    • Which patterns can be extrapolated more widely?
    • What situations do they apply to?
    • What situations are too different so they don’t apply?
    • how good is the fit?
      • a single pattern can fit different situations to varying degrees
      • “no evidence” means the evidentiary patterns they know of don’t fit well enough
      • other ways of judging the fit of a pattern is to judge the similarity and relevance of a pattern to situations
  • induction relies on answers to these questions
    • because induction says the future probably resembles the past, i.e. that the future is similar to the past
    • induction must also know how to know which patterns will continue, because some breaks
    • for any occurrence we can pick a pattern from an infinite pool of perfectly fitting patterns
      • due to mathematical and logical properties of patterns
      • so why bother with good fit when you can have perfect fit?

Exact Matching

  • exact matching means we don’t need to judge how well things fit
  • it’s too strict to be useful, we wouldn’t be able to find useful patterns that match real data especially over time
  • we can add fuzz into the pattern to make it fit more easily
    • we can use ranges or say a certain characteristic can be one of many categories
    • but that wouldn’t match reality because the patterns are exact but our measurements have margins of error or we leave out details in our calculation
  • in reality scientist do fuzzy matching with the assumption that the data is fuzzy
    • which means they are trying to figure out how good the fit is
  • we can add fuzz to the data like adding margins of error which we assume is due to the measurement imprecision
    • then we have to ask how much fuzz is OK?
    • we will choose patterns that don’t quite match while there are exactly matching patterns as well because those patterns seem useful while the exact ones don’t
      • so now we have to think about which other characteristics that make patterns good and how much that allows us to fuzzily match the data
      • induction is supposed to create knowledge from observed patterns, but with our preferences some of, if not all, the knowledge creation is coming from our preferences for patterns

Don’t Focus on Patterns

  • solution? don’t base the epistemology on patterns, similarity, etc.
  • deduction works but is limited
  • evolution
  • instead of focusing on patterns use critical arguments to reject errors
  • instead of degrees of fitting a pattern we have non-refuted vs refuted as statuses for ideas
    • secondarily we can look at how much criticism an idea has faced and survived and also it’s a new revolutionary idea or an old traditional idea
      • we can critique ideas by saying they haven’t been critically investigated enough
        • we don’t have infinite time to critique every idea
        • how much to critique an idea varies by importance and context
        • we have to use judgment to decide how much time and energy to spend on an idea
        • how much to critique an idea is an idea itself which we can criticize too
          • your idea of much to critique usually changes throughout the process of critiquing and learning
  • conclusion: focus on criticism to avoid problems with focusing on patterns, similarity and relevance
  • literature and debaters haven’t yet solved these problems
3

Induction also runs into the huge problem of figuring out which patterns will continue over time. The premise about patterns continuing is kind of missing the point. No matter what happens, some patterns will continue and others will break.

Is it missing the point because it doesn’t help much? That patterns continue is trivial, but some won’t and the difficult problem is figuring out which will. So the premise doesn’t help when looking at patterns and reasoning/inducing about them.

Fuzzing the data itself, especially when you start ignoring some contradictory data points or outliers, leads to questions about how much it should be fuzzed.

Can’t we reason about how much precision we expect from our measurement instruments and base the fuzz on that? Although I can’t really see how the standards for precision could be truly objective. But do we need perfect standards for precision, isn’t some somewhat arbitrary standard good enough?

Because scientists do rationally use imperfect standards for precision in order to refute or pass theories, right?

For the purpose of refuting theories an imperfect standard for precision is good enough. But induction is based on pattern matching so it would need better standards for what fits. Otherwise it would be using a stolen-concept of non-inductive reasoning.

Also, some patterns perfectly fit the data with no fuzzing, so why prefer patterns that require fuzzing the data?

Because we assume the measurements are imprecise. I think that means we actually wouldn’t expect the exact pattern to be the correct one since we think it’s more likely that the data isn’t exactly correct.

Here was a thought I had: “A pattern that fits many situations fuzzily is preferred over ones that only fit a few situations exactly”. But a pattern that fits many situations doesn’t actually say much on the quality of the pattern.

Patterns that fit many situations could be like the ones Popper said was likely true according to the calculus of probability. Those would be more unconstrained or fuzzy patterns. Such patterns fit easily to more situations because they’re less specific and contain less content. They just doesn’t tell us that much. Rather we should look for explanations/patterns that say a lot and say more exactly what things are like. Such explanations/patterns are less likely to be true in the sense of the calculus of probability.

4

Hmm. One difference in my head seems to be that biological evolution seems to be based on what survives versus evolutionary epistemology is a pick and choose kind of thing. Though I guess if you prove stuff wrong then it’s a survival kind of thing. Hmm.

5

Biological evolution adapts genes to a purpose. Evolution requires selection criteria. The criterion in nature is, roughly, having grandchildren (more is better). […]

Intellectual evolution adapts ideas to a purpose. But humans can choose their own purposes. We aren’t born with a single, fixed purpose. On one day I can adapt ideas to one purpose, and on another day I can change my mind and adapt them in a different way. I can use different selection criteria at different times or for different topics.

The genes also needs to reproduce. Not just survive.

Yeah they need to survive whatever selection criteria based on the purpose we have for them. An interesting thing that occurred to me: when animals goes extinct the genes can disappear with them, but for ideas we can keep rejected ones in memory and when our purposes and selection criteria changes they might be useful again.

6

Ideas have to survive people picking and choosing what ideas they think are good or bad. People get rid of some ideas.

7

Hmm. So they can also pick wrong. Right?:Hmm. Biological evolution can do that too? After all the things that don’t survive picked wrong.