Meta-rational practices: Overview 2

Opportunities and methods for meta-rational improvement

This post is a draft of a chunk of my meta-rationality book. You can see the outline of where this piece belongs in the Table of Contents, with links to earlier draft pieces.

This chunk is much more specific than the last post’s. Following helpful reader feedback on the previous post (thanks!), I’ve also included many brief examples in this one. I intend the next draft chunk to include an in-depth analysis of a single example, perhaps the discovery of penicillin. Later chapters will get more specific and detailed still.

This post also includes more extensive exercises. I’d love to hear how those go for you!

My thanks to the fourteen readers who signed on as paying subscribers after reading my last post. I really appreciate it! This one is part-paid: there’s a paywall about half way through.

The post uses several terms without definitions. These are discussed extensively in Part Three of the book, which belongs before this one. It’s unwritten, although a long summary is public.

If you run into any unfamiliar terms, consult the Glossary. I am adding to it as I write draft sections, and it should cover all those used in this post.

If you notice any problems, or opportunities for improvement in this draft, please leave a comment!

Opportunities for meta-rational improvement

We can categorize opportunities for meta-rational improvements in terms of sites in the “Anatomy of Rationality” diagram from Part Three:

• Reengineer the outside world to better fit the system

• Improve circumrational practice on human side of the interface

• Improve circumrational practice by modifying the system internally

• Improve the ontology

• Improve the internal operation of the rational system

• Improve the material cognitive supports available for Problem Solving

• Apply meta-rational operations to the system as a whole

Like all taxonomies, this should not be taken rigidly. It’s a suggestive guide to a nebulous field of opportunities. Each of these categories can be expanded into subcategories, as we’ll do in just a minute. Those subcategories could be expanded into subsubcategories. Later chapters will do some of that; but we could continue taxonomizing unboundedly, and it could become an encyclopedia.

Numerous intermediate methods, of varying degrees of generality, are applicable at each site. Each method could be illustrated with many interesting examples. A serious catalog could be useful, but gigantic. Instead, this section points out a few, arbitrarily selected, methods for most subcategories of opportunities. The intent here is to give you the flavor of what multi-rational activity looks like, rather than to provide a comprehensive manual. Later chapters do explore many of the categories and methods in greater depth, illustrated with case studies.

In the Exercises at the end of this section, I’ll ask you to provide your own case studies, from your own experience or reading!

In practice, you learn the intermediate methods most relevant for you through experience in a particular domain. Some you learn explicitly, through reading, or more often because someone demonstrates them or tells you verbally. Others you figure out for yourself in the course of your work, by noticing a pattern of unusually good or bad system functioning.

Gradually, you get to be able to apply intermediate methods across domains. This is a key to fluid competence, or “seemingly omniscient sorcery,” discussed later in this chapter.

Reengineer the outside world to better fit the system

Rationalism’s mythological prototype is the discovery of universal natural laws: a fitting of a rational system to the world. The converse is far more common: rationality is made to work by fitting the world to the system.

• More stringent standardization of inputs

  • By internal policy: “from now on, we will purchase only the 99.99% pure version of this chemical, instead of the 99.44% product”

  • By technological means: “we have developed an improved purification method for the chemical”

  • By external policy enforcement: “we now mandate the improved purification method for our suppliers”

The development of scientific agriculture, including the mid–20th-century Green Revolution which saved a billion people from starvation, has depended heavily on standardizing inputs. That includes the crops grown, through selective breeding and then increasingly sophisticated genetic technologies. It also includes, for example, increasingly precise control of the quantity and quality of irrigation water.

Generally, this has been an enormous contribution to human well-being. James C. Scott’s meta-rational Seeing Like A State examines several cases in which, due to oblivious rationalism that ignored context, purpose, and nebulosity, scientific agriculture failed, sometimes causing mass famines.

• More effective shielding

  • Better vacuum seals, more precise temperature control, fences to keep tumbleweeds from bumping the gravitational wave detector’s external walls…

  • Strengthening social structures and norms that keep outsiders from interfering (physically or through social action)

  • Computer security measures to keep hackers out of the database

Improve circumrational practice on human side of the interface

• Better ways of describing external phenomena in the system’s terms (its ontology and notation)

  • Replace a drop-down selection list in a user interface with a free text field, or vice versa

  • Use visual comparison of textile samples with standardized color swatches to more accurately classify their colors before entering them in the inventory database

• Better ways of interpreting the system’s formal demands as concrete activities

  • More detailed instructions on how to determine whether a crate of bananas arriving at the warehouse are overripe and should be rejected

  • Provide video demonstrations of laboratory protocols, not just textual instructions

• Upskill participants on known circumrational methods

• Find ways to avoid, repair, or reduce the cost of circumrational troubles

  • Mechanical or electronic interlocks that prevent common user errors

  • Institute quality checks earlier in the production line, so they’re detected before they get expensive

  • Metaphorically, “production line” can include any sequential process, such as planning a research or development project; if you can detect or predict likely circumrational failures early, it saves a lot of work

  • Modify practices that can result in painting yourself in a corner by changing the order of operations. “Drat, I should have checked the pH before adding the third chemical to the beaker; now it is too late!”

• Better material supports for circumrational work

  • measuring devices

  • cognitive supports (calculator, whiteboard, software)

  • recording technologies (notebook, computer data entry)

  • media for circumrational workers to communicate with each other; a major case study later in this Part concerns the development of one such, the Xerox Eureka system for sharing knowledge among copier technicians

• Better institutions, management, or other social organization for circumrational work

• Proceduralization of routine circumrational activities so they can be performed mindlessly, without needing nebulous reasonable judgement

  • prescriptive documentation of standard operating procedures

  • fully rationalizing them, thereby bringing them inside the formal system itself

  • automating them altogether

Improve circumrational practice by modifying the system internally

• More accurate, specific, informative, or interpretable system outputs

  • Provide data visualizations such as scatterplots instead of just summary statistics

  • Software for telephone customer service reps provides exact words to say instead of general directions

• Better system input affordances

  • Speech recognition for doctor’s clinical record inputs instead of keyboard text entry

  • Standardized worksheet for recording a particular type of experimental result, instead of using your free-form laboratory notebook

• More coherence and reality checking of inputs and outputs

  • Routinized or automated quality checks on externally-provided, supposedly standardized materials, such as laboratory reagents or third-party manufactured parts

  • Statistical outlier detection, and removal or correction, at any stage in a rational process

  • Quality control on final products

The category “Better system input affordances” is not clearly distinct from “Better material supports for circumrational work,” because circumrationality operates at and across the boundary of the rational system. Similarly, “More coherence checking of inputs and outputs” overlaps with the upcoming category “Improve the internal operation of the rational system.”

Improve the ontology

• Split categories based on a distinction newly understood to be relevant, or newly measurable

A critical step in the development of chemistry was the realization that the old category “substance” included two radically different sorts: stoichiometric compounds, meaning roughly “molecules made from a fixed number of atoms of particular elements” and mixtures, in which the constituents could combine in arbitrary proportions.

• Merge categories that appeared superficially different, but that should be treated the same way for the purposes the ontology serves

Increasingly it appears that long covid, chronic Lyme disease, and ME/CFS have similar causes and processes, in which case knowledge about each may illuminate the others, and effective treatments for any one of them—if they are discovered—may work across the merged category. 

• Replace a binary distinction with a continuous scale

Previously, it was thought that psychopathy was an inherent, binary condition: either you are a psychopath or you aren’t. Recent progress in understanding, with hope for treatment, comes from recognition that the frequency and intensity of psychopathic thinking, feeling, and acting are continuously varying, and a much larger fraction of the population is “somewhat psychopathic” than are clear-cut “psychopaths.”

• Extend the ontology to cover previously-marginal cases that couldn’t be classified coherently

Before the cause of AIDS was definitely established, it was classified together with many disparate conditions, some involving viruses and some not. A major advance was adding it to the category of human T-cell lymphotropic virus diseases. When HIV was first isolated, it was named HTLV-III, the third known virus in that category. This turned out to be wrong, but significantly helpful on the way to understanding and then treating AIDS.

• Reassign anomalous cases that got chucked arbitrarily into some category into a newly-created wastebasket category (“miscellaneous”)

An improved ontology is not necessarily more precise or accurate. Merging categories that get treated identically in practice reduces unnecessary complexity. Adding a wastebasket category can be an improvement, as can adding “not applicable” or “unknown” property values. (It’s better to record that the source of a chemical is “unknown” than to arbitrarily assign it one of the twelve designated suppliers.) 

• Recategorize wastebasket contents based on better understanding, and eliminate the now-empty wastebasket category

“Neurasthenia” was a common, debilitating medical condition prevalent until quite recently. In retrospect, it was a wastebasket category, used as the diagnosis for many unrelated conditions (from inflammatory bowel disease to high blood pressure to sexual frustration) that might be summarized as “general yucch.” It got gradually narrowed to mean roughly “neurotic fatigue,” and was officially eliminated only in 2019.

• Improve explanations of the ontology and why it is good

Ontological remodeling is the topic of a whole chapter later in Part Four. We’ll return to some of the examples here, and explore them in greater depth.

Ontological remodeling inherently cannot be done rationally, because rationality takes ontology as fixed. Ontology imposes a definite pattern on nebulous phenomena, so creating or improving one requires prior non-rational understanding.