Evidence and absurdity

In the previous posts I recalled five categories of design explanation. I used the innocuous example of deciding to insert a large window into a living room. I explored how a client/reader might regard various explanations as either sensible or absurd.

I wanted to avoid the language of truth and falsity, arguing that decisions and their explanations are often disconnected, or at least explanations and justifications are contingent on the context in which they are presented.

I was reminded of this disconnect when Donald Trump presented his explanation for imposing tariffs on goods imported to the USA. At times it seemed the decision was to restore fairness to the trade system; at other times tariffs were to serve as a bargaining tool to induce neighbouring countries to restrict the flow of migrants and illegal drugs; at other times it was as retaliation for unfavourable attitudes and actions from other countries.

The fact that Trump and his advisors were unable to settle on a singular explanation signalled the illusive nature of “real reasons” — if not the absurdity of both this particular decision and its explanations.

Evidence-based rationale

Any explanation may also call on evidence, that tariffs have in the past had the desirable effects claimed of them, that the flow of migrants and drugs are related and are of a scale that warrants such measures, and that countries have behaved as claimed: “they have treated us very badly!”

To return to design, let’s incorporate evidence into each of the five categories of explanation outlined in the previous post. NotebookLM helped me with these examples.

Formal Logic

Sensible: In a logical explanation, evidence serves to support the premises from which the conclusion is deduced. For example, if we logically argue that large windows are desirable for natural light based on the premise that “sufficient natural light improves mood,” the evidence for this premise could come from empirical studies in environmental psychology. The validity of the logical argument itself is separate from the truth of the premises, but the persuasiveness and applicability of the explanation depend on the evidence supporting those premises.

Absurd: An absurd logical explanation might rely on premises that lack any supporting evidence or are contradicted by available evidence. In our furniture discomfort example, the premise that “all furniture dislikes sunlight intensely” has no empirical basis and is easily refuted by evidence of materials used in furniture design and the general appreciation of natural light. (That’s even accepting that it is meaningful to ascribe likes and dislikes to furniture.) Even if the logic is sound, the lack of evidential grounding makes the explanation absurd.

Rule-based system

Sensible: When explaining large windows by citing a rule (e.g., building regulations), the evidence lies in the documented existence and authority of that rule. The explanation is strengthened by providing the source and context of the regulation, demonstrating its legitimacy and applicability to the situation.

Absurd: An absurd explanation using a “rule” might cite a non-existent rule or misrepresent an actual rule. Our migratory bird rule example (“living rooms must be designed to facilitate the unobstructed passage of migratory birds”) is absurd precisely because there is no evidence to support its existence as a legitimate building regulation. The explanation is baseless and easily shown to be false through a lack of evidence.

Precedents

Sensible: Using precedents effectively relies on evidence of the precedent’s success and relevance to the current situation. This evidence could include documented user satisfaction, performance data, or expert reviews of the precedent design. The explanation is stronger if there is clear evidence linking the features of the precedent (large windows) to positive outcomes in similar contexts.

Absurd: An absurd precedent-based explanation might cite a precedent whose context is entirely dissimilar or whose alleged success is not supported by evidence. Our medieval dungeon example (that exaggerating the size of openings is a logical continuation of the dungeon window precedent) is absurd because the context and purpose of a dungeon are different from a living room, and there’s no evidence suggesting that the window design in dungeons is a desirable precedent for domestic spaces. The lack of relevant evidence and the misapplication of the precedent contribute to the absurdity.

Metaphor

Sensible: While metaphor is not typically based on empirical evidence in the same way as logic or rules, a sensible metaphor often draws on shared cultural understandings and experiences, which could be considered a form of collective “evidence” for its resonance. A successful metaphor provides an insightful and relatable way to understand a concept.

Absurd: An absurd metaphorical explanation often lacks any grounding in shared experience or creates a nonsensical and unhelpful comparison. Our “living room as a giant internal nostril” metaphor is “absurd” because there’s no common understanding of buildings having olfactory organs, and the comparison doesn’t illuminate the function or purpose of large windows in a meaningful way. The lack of relatable “evidence” for the comparison makes it absurd. Metaphors that thwart logical interpretation can be seen as nonsensical from a purely logical standpoint.

Neural network

Sensible: A sensible explanation generated by an LLM should ideally be traceable back to patterns and evidence within its training data. While the LLM doesn’t process semantic information in a human way, the basis of its explanation lies in the statistical co-occurrence of features (such as large windows) with certain contexts or outcomes in the vast amount of text and code it has processed. Evidence, in this case, is the training data.

Absurd: An absurd LLM “explanation” can arise from spurious correlations in the data that have no “real-world” causal relationship, as in our cats-in-trees example — i.e. if it happens coincidentally that the training data includes a lot of text from discussion forums where people who mention living room windows also happen to talk about cats in trees. The “evidence” (the statistical correlation) is misleading and doesn’t provide a meaningful justification for the design decision. This highlights the danger of relying solely on data without considering underlying relationships or common sense. The model’s inability to discern meaningful connections from mere co-occurrence leads to absurd “explanations” based on flawed or irrelevant “evidence.”

NotebookLM summarised: evidence plays a crucial role in grounding explanations, whether they are based on logic, rules, precedents, metaphors, or the outputs of LLMs. Sensible explanations are supported by relevant and valid evidence, while absurd explanations often rely on a lack of evidence, missapplied evidence, or nonsensical associations that lack any experiential basis. Understanding the role of evidence within each explanatory framework helps us to critically evaluate the validity and meaningfulness of the justifications provided.

The process of interpretation, as highlighted in hermeneutics, also comes into play when we evaluate evidence and determine its relevance and significance within a particular context. Such processes are critical when we consider how influential is the choice of data, and the frameworks through which we gather, assess and understand it.

That said, unlike in trade wars, I do not want to abandon a role for absurdity in design development and decision making. See post Marx on nonsense.