Tinkering and uncertainty at different levels — email thread

• “Nestedness” is key to evolution — which is another trial and error process
• Nestedness allows one part of the system to fail without affecting the whole thing
• Nestedness is related to but not the same as modularity (modularity doesn’t have the property that a module can fail, only that it can be replaced)
• Non-teleological trial-and-error process
  • No fully articulated end state
• Four ways of knowing
  • Observation
  • Deduction
  • Induction
  • Trial-and-error
• Properties of T&E
  • Related to observation but not passive
  • Usually a process
• The point is then to outline the basics of T&E
  • (Simon called it generate and test, but this really elides a lot of important stuff like
  • how do you generate?
  • how do you rank which things to test first?
  • how parallel should you go?
  • when should you iterate vs stop the process?
  • what is the feedback from the testing to the generation?
  • does a test change the landscape such that future tests will have different results? etc.)
  • And then, finally, to talk about adapting this to innovation.
• Mechanisms of science

Responses

• Trial and error is actually a combination of induction, observation, and (manipulation? Affordance finding? Actuation?)
• Trial and error process = manipulation observation induction repeat
• Taxonomized the mechanisms of science?
• I think you’re correct that there are four ways of knowing, but I don’t think trial-and-error is the right fourth way, because it actually includes some of the others. I feel like personal experience is actually legitimate evidence on what T&E entails and in (a lot of? too much?) experience with trial and error in physical, digital, and human systems, it goes like:
• Use induction or deduction to suggest a way to act on the system (I’m going to ignore the base case — maybe it’s just true randomness)
• Act on the system
• Observe the results of acting on the system
• Either incorporate the observation into your deductive model or build on it with a deductive process to suggest a new way to act on the system
  This is the T&E “process” you allude to below and I would argue that T&E is only a process. What we need is a word for knowing by having acted on a system — the “finger knowledge” of a craftsperson. I tried out “actuation” “testing” “feels” “affordance-understanding” but I think the correct word is “intuition.” A master woodworker intuits which tool to pick up for the task not because he thinks “ah yes, creating this shape in this wood needs this tool” he just intuits it, an experienced physicist intuits that a theory is BS not because she thinks “ah yes this violates theorems x, y, z” (though she can quickly think of those) but because her gut screams against it, having worked with those theorems for decades.

Human system differ significantly from evolution. One hot take is that it has a lot to do with how “next steps” in a closed loop are generated. In evolution it is purely stochastic, in human systems it’s not (though this starts to touch on philosophy — free will and all). But I think that actually matters because in extremely high dimensional spaces, if you make a huge random jump your chance of hitting a viable spot is effectively zero given the lifetime of the universe.

This may also have to do with the distinction that Brian Arthur makes between Combinatorial evolution and Darwinian evolution are different

Published

Authors

• Jason Crawford
• Jerry Neumann

  Link

  Related

• Jerry Neumann posits that there are four ways of knowing
• There are model-based and model free ways of knowing
• There are several ways to taxonomize ways of knowing
• Group-based and individual-based knowledge seems to be different
• Trial and error requires a feedback loop
• Tinkering and uncertainty at different levels — email thread

Web URL for this note

Comment on this note