Sunday, April 16, 2017

fudge-topped double fudge dipped in fudge

Two truisms to start with. First, finite creatures like us don't have the ability to swiftly uncover all the details of each large/complex thing or situation. However, we still need to work with such things in order to do all sorts of necessary tasks. Second, we may combat our lack of exhaustive knowledge about single cases by extracting and documenting patterns from many cases and carefully fashioning the patterns into reusable models.

I'm using "model" in a philosophically broad sense: it's an abstracted symbolic representation which is suitable for specifying or explaining or predicting. It's a sturdy set of concepts to aid the work of manipulating the information of an object. This work is done by things such as human brains or computing machines. The model feeds into them, or it guides them, or it's generated by them. It reflects a particular approach to envisioning, summarizing, and interpreting its object. Endless forms of models show up in numerous fields. Sometimes a thoroughly intangible model can nonetheless be more tangible and understandable than its "object" (a raw scatter plot?). 

Some models are sketchy or preliminary, and some are refined. Some seem almost indistinguishable from the represented object, and some seem surprising and obscure. A model might include mathematical expressions. It might include clear-cut declarations about the object's characteristic trends, causes, factors, etc. The most prestigious theories of settled science are the models that merit the most credence. But many models are adequate without reaching that rare tier. Whether a model is informal or not, its construction is often slow and painstaking; it comes together by logically analyzing all the information that can be gathered. It's unambiguous, though it might be overturned or superseded eventually. It's fit to be communicated and taught. Chances are, other models of high quality were the bedrock for it; if not, then at minimum these others aren't contradictions of it. It's double-checked and cross-checked. Its sources are identifiable.

Despite the toil that goes into it, the fact is that a typical model is probably incomplete. Comprehensive, decisive data and brilliant insights could be in short supply. Or portions of the model could be intentionally left incomplete/approximate on behalf of practical simplicity. When relevant features vary widely and chaotically from case to case, inflexible models would be expected to perfectly match no case besides the "average". Perhaps it's even possible to model the model's shortcomings, e.g. take a produced number and increase it by 15%.

For whatever reason, the model and some cases will differ to some extent...but improving the model to completely eliminate the individual differences would be infeasible. So the disparity from the model is fudged. I'm using this verb as broadly as I'm using "model". It's whenever the person applying the model decides on an ad hoc adjustment through their preferred vague mix of miscellaneous estimations, hunches, heuristics, and bendable guidelines. Hopefully they've acquired an effective mix beforehand through trial and error. (The result of fudging may be referred to as "a fudge".)

If a realm of understanding is said to be an art as much as a science, then the model is the science and the fudging is the art. In the supposed contrast of theory with practice, the model is the theory and the fudging is one part of the practice. The model is a purer ideal and the fudging is a messier manifestation of the collision with the complexity of circumstance. The model is generally transparent and the fudging is generally opaque. Importantly, a person's model is significantly easier for a second person to recreate than their fudging.

The crux is that a bit of a fudge, especially if everyone is candid about it, is neither unusual nor a serious problem. But overindulgence rapidly becomes a cause for concern. The styles of thinking that characterize normal fudging are acceptable as humble supplements to models but not as substitutes. One possible issue is reliance on implicit, unchecked assumptions. Models embody assumptions too, but model assumptions undergo frank sifting as the model is ruthlessly juxtaposed with real cases of its object.

Another issue is the temptation to cheat: work backwards every time and keep quiet about the appalling inconsistencies among the explanations offered. Someone's slippery claim of fudging their way to the exact answer in case A through simple (in retrospect) steps 1, 2, and 3 should lose credibility after they proceed to claim that they fudged their way to the altogether different exact answer in case B through simple (again, in retrospect) alternative steps 1', 2', and 3'. A model would impose a requirement of self-coherency—or impose the inescapable confession that today's model clashes with yesterday's, and the models' underlying "principles" have been whims.

Yet another issue is the natural predisposition to forget and gloss over the cases whenever fudging was mistaken. The challenge is that new memories attach through, then later are reactivated through, interconnections with prior ideas. But a mismatch implies that the fudging and the case don't have the interconnections. The outcome is that a lasting memory of the mismatch won't form naturally. The primal urge to efficiently seek out relations between ideas comes with the side effect of not expending memory to note the dull occurrence of ideas not having relations. After being picked up, unrecorded misses automatically drift out of memory, like the background noise of an average day. A rigid model counteracts this hole because it enforces conscious extra effort.

The issues with replacing modeling with too many fudges don't stop there. A sizable additional subcategory stems from the customary risk of fudging: judging between hypothetical realities by the nebulous "taste" of each. That method would be allowable...as long as a person's sense of taste for reality is both finely calibrated and subjected to ongoing correction as needed. Sadly, these conditions are frequently not met. Some who imagine that their taste meets the conditions may in actuality be complacent egotists who're incapable of recognizing their taste's flaws.

A few comparisons will confirm that overrated gut feelings and "common" sense originate from personal contexts of experience and culture. Divergent experiences and cultures produce divergent gut feelings and common sense. An infallible ability to sniff out reality wouldn't emerge unless the sniffer were blessed with an infallible context. That's...extremely improbable. Someone in an earlier era might have confidently said, "My long-nurtured impression is that it's quite proper to reserve the privilege and responsibility of voting to the kind of men who own land. Everybody with common sense is acutely aware of the blatant reality that the rest of the population cannot be trusted to make tough political decisions. Opening the vote to them strikes me as foolhardy in the innermost parts of my being."

But context is far from the sole way to affect taste. Propagandists (and marketers) know that repetition is an underestimated strategy. Monotonous associations lose the taste of strangeness. Once someone has "heard a lot" about an assertion, they're more likely to recall it the next time they're figuring out what to think just by fudging. Its influence is boosted further if it's echoed by multiple channels of information. For people who sort reality by taste, its status doesn't need to achieve airtight certainty to be a worthwhile success. Success is achieving the equivocal status that there "must be something to it" because it's been reiterated. A fog of unproven yet pigheaded assertions would be too insubstantial to meaningfully revise a model, but with enough longevity it can evidently spoil or sweeten a reality's taste by a few notches.

Repetition is clumsy, though. Without a model to narrow its focus, the taste for reality is susceptible to cleverer attacks. Taste is embodied in a brain packed with attractions and aversions. The ramification is that emotional means can greatly exaggerate scarce support. A scrap in a passionately moving frame manages to alter taste very well. In the midst of weighing perspectives by impromptu fudging, the stirring one receives disproportionate attention. If the scale has been tilted masterfully, someone will virtually recoil from the competing perspectives. Gradually establishing the plausibility of a model is a burden compared to merely tugging on the psychological reins.

If distorting taste by exploiting the taster's wants seems brazen, the subtler variation is exploiting what the taster wants to believe. It may be said that someone is already half-convinced of notions that would fit snugly into their existing thoughts. The desire for a tidy outlook can be a formidable ally. It's not peculiar to favor the taste of a reality with fewer independent pieces and pieces that aren't in dischord. The more effortlessly the pieces fall into place, the better. Purposefully crafting the experience that a proposal gratifies this component of taste is like crafting the experience that it's as sound as a mathematical theorem. It will appear not only right but indisputable. Searching will immediately stop, because what other possibility could be more satisfying? ...Then again, some unexpected yet well-verified models have been valued all the more as thrilling antidotes to small-mindedness.

This extended list of weaknesses suggests that compulsive fudgers are the total opposite of model thinkers. However, the universe is complicated, so boundaries blur. To repeat from earlier, model thinkers regularly have the need to fudge the admitted limits of the models. And the reverse has been prevalent at various times and locations in human history: fudge after fudge leads to the eventual fabrication of a quasi-model. The quasi-model might contain serviceable fragments laid side by side with drivel. It might contain a combination of valid advice and invalid rationales for the advice. The quasi-model is partially tested, but the records of its testing tend to be patchy and expressed in all-or-nothing terms. It could be passed down from generation to generation in one unit, but there's uncertainty about which parts have passed or failed testing and to what degree.

Once someone does something more than fudge in regard to the quasi-model, it might develop into a legitimate model. Or, its former respectability might be torn down. The dividing line between quasi-model and model is a matter of judgment. If it's resting on fudges on top of fudges, then signs point to quasi-model.