why cld isn't enough for programmatic theory

[hyunjimoon]

tl;dr

• i lean towards statistical reasoning not causal reasoning. the latter is idealistic but humans have no tool to verify as long as there exists aleatoric uncertainty.

---

• definition of programmatic theory, see CroninStoutenvanKnippenbergAMR.2021.0517.Final.pdf

• from 7. Of the idea of necessary connection, David Hume argues:

Having dispensed with these alternative explanations, he identifies the source of our knowledge of necessary connections as arising out of observation of constant conjunction of certain impressions across many instances. In this way, people know of necessity through rigorous custom or habit, and not from any immediate knowledge of the powers of the will.

Causal loop diagrams (CLDs) are problematic as a theoretical framework because they assume necessary connections between variables that, according to Hume, humans can never truly verify through empirical observation alone 🔍🤔. While CLDs draw deterministic arrows indicating direct causation, Hume shows that we only observe constant conjunctions between events, learning through statistical patterns and associations rather than explicit causal logic - this points to why we should favor probabilistic programs that can better capture uncertainty and the subtle patterns through which humans and animals actually process information and make decisions 🎲🧠.

Based on Hume's insight that knowledge emerges through "constant conjunction" - the repeated observation of one event reliably following another - I aim to develop an algorithm for mining test quantities with exchangeability. Rather than assuming direct causal links, exchangeability emerges from documenting these constant conjunctions systematically across different contexts. This approach guides model development at both individual and societal levels by focusing on what patterns consistently conjoin in practice (like Tesla's observation of how certain battery configurations reliably lead to better performance) rather than attempting to deduce optimal solutions from assumed causal relationships.

using Hume's Critique of Causal Determinism cld

[hyunjimoon]

additional search on "spaghetti criticism" on vensim, i found two interesting links:

• Bill cutler prescribes hierarchy for spaghetti 💊antidote

• John Sterman, George Richardson et al. had interesting discussion on increasing complexity of loops (and regulations thereof) in [feedback forward](Feedback -forward?)

joke about systems dynamicists. The joke was that system
dynamicists just add more and more feedback loops whenever the model
didnt match historical data such that everything was nothing more than
a spaghetti!! This spaghetti-looking diagram may be one factor that
impedes the progress of system dynamics as a tool or methodology.

spaghetti seems to be SD-derogatory term?

I dont think the "spaghetti" criticism of System Dynamics is reasonable. In fact, I think that people are prone to be simplistic and get locked into a single governing phenomena that is in vogue, like the path dependence phenomena that has been discussed previously.

understanding cognitive upper bound of human shouldn't be the complexity ceiling as human cognition evolve and without an aligned evaluation metric for measuring cognitive capacity (due to low understanding and too diverse context of use of cognition to evaluate with unified metric) - there's a high chance your estimate is bad (either under/over estimate). also is everyone prone to be simplistic? there's always a matcher that preys on simple models with unrealistically heroic assumptions.

---

more on

• https://metasd.com/2010/04/hypnotizing-chickens/
• https://www.ventanasystems.co.uk/forum/search.php?keywords=Spaghetti
• left comment in https://metasd.com/2010/09/interactive-diagrams-obesity-dynamics/#comment-343649 asking Tom for further key reference

[hyunjimoon]

this reminds me of Josh's class:

• comparison between causal graph and probabilistic program
• statistical vs causal reasoning

extracted from #250 (comment)