0.1The question and the claim
A question about what to do can be put in two ways. It can ask what matters in a situation, or it can name two courses and ask which is better. The second way is the ordinary shape of a decision, and it carries a quiet assumption: that the two courses can be set on one scale, so that one comes out as fairer, safer, wiser or more reasonable than the other. This paper studies a set of records that put their questions the second way and then, in the same record, say what that way of putting it leaves out.
The records come from Cosmic Press, a studio in the RISCON family whose public catalog holds AI-generated questions. Each record pairs a question in a person's voice with a set of notes on how that question is framed. No record answers its question, and none carries a judgment. All of them are made by AI, and the situations they describe are invented. The records are therefore not evidence about the households, lifts, gyms and flower beds they describe. They are evidence about one thing only: how this record set frames a choice.
The material here is the closing file of the catalog, sixteen records numbered q753 to q768. In them a person working from home wonders whether to spend a lunch break tidying a shared storeroom, a residents' meeting votes on raising a repair fund to replace an old lift, a family argues over a lock app on a child's phone against the child's autonomy, a couple's washing-up waits late at night, and a meeting decides whether to tear down old alleys and keep only photographs. Its question is this: when a record weighs two goods on one scale, how does it account for what the weighing costs, and what does its labelled ladder of six levels do with the choice?
The claim is in two parts. First, in this study's reading, the records fault their own weighing, and they do so with remarkable regularity, but the fault they name is almost always a split: the question set two poles against each other, or reduced the matter to one side, and so missed a middle course or a consideration that belongs on the scale. The counterquestion then proposes that middle course. The records do not say that the two goods cannot be measured on one scale; they say that the scale was used too narrowly. Second, the six scale notes no longer ask anything. They restate the act at six labelled levels and, in doing so, add measures of their own: at the level of the Earth the act becomes power used or saved, and at the level of the Universe it becomes energy, order, entropy or signal. The last note restates one of the two courses in physical words and says nothing about the choice between them.
The contribution is a description of a generated form in which commensuration, the weighing of different goods on one scale, and a critique of that weighing are built into every record, together with an account of where the critique stops. The form criticises the split and keeps the scale, and it criticises the questioner's reduction while carrying out a reduction of its own in its ladder. Both findings concern how the record set frames choices. Neither is a finding about the choices.
0.2Two debates
The first debate is over commensuration. Espeland and Stevens, Commensuration as a Social Process (1998), define commensuration as the comparison of different entities according to a common metric. They treat it as a general social process, evident in routine decision-making, and they analyse it as a mode of power; they also discuss the cognitive and political stakes in calling something incommensurable. On the other side of the debate stand claims that goods do not share a metric. Sunstein, Incommensurability and Valuation in Law (1994), separates two claims: that human values are plural, valued in ways not reducible to some larger value, and that human goods are not commensurable, not assessed along a single metric. The two claims, though related, are importantly different. Anderson, Value in Ethics and Economics (1993), rejects the market as a common measure of value and cost-benefit analysis as a guide in social life and ethics. Valuing, in Anderson's account, is done not more or less but in different ways, and asking how a thing is valued, instead of how much, shows how goods differ in kind: objects of love, respect and admiration differ from objects of mere use.
A further strand of the debate concerns the reaction to certain weighings. Fiske and Tetlock, Taboo Trade-offs: Reactions to Transactions That Transgress the Spheres of Justice (1997), describe trade-offs that violate deeply held intuitions about the integrity of certain relationships: asked to put a money value on their children, their loyalty or acts of friendship, people find the question itself morally offensive. Tetlock, Kristel, Elson, Green and Lerner, The psychology of the unthinkable: Taboo trade-offs, forbidden base rates, and heretical counterfactuals (2000), report studies in which people met taboo trade-offs that monetized sacred values with moral outrage and moral cleansing.
The records place themselves in this debate in an unusual way. Nearly every question performs a commensuration, and every record then names a cost of it. The paper asks which side of the debate that self-criticism falls on: whether it says that the scale itself fails, as Sunstein and Anderson would have it, or only that the scale was used too narrowly.
The second debate is over levels and reduction. Cosmic Press's scale notes, in the stretch read here, carry six labels in a fixed order from the person to the universe. A ladder of labelled levels is an old figure in science and philosophy. Potochnik and McGill, The Limitations of Hierarchical Organization (2012), note that the concept of levels of organization is prominent in science and central to debates in philosophy of science, and argue that many difficulties plague the concept of universal and discrete hierarchical levels. They propose scale instead: investigating causal processes at different scales gives quasi levels that avoid those difficulties. Wimsatt, Aggregativity: Reductive Heuristics for Finding Emergence (1997), observes that most scientists regard a property of a system as emergent when it depends on how the system's parts are organized, a view consistent with reduction. Emergence can be analysed as a failure of aggregativity, the state in which the whole is nothing more than the sum of its parts. Aggregativity requires four conditions, which give tools for analysing modes of organization, criteria for choosing decompositions of a system, and heuristics for detecting the biases of vulgar reductionisms. Kim, Supervenience, Emergence, Realization, Reduction (2005), treats concepts that have played prominent roles in metaphysics in recent decades, in the debates over the mind-body problem and the status of the special sciences, chiefly in characterizing how mental properties are related to physical properties and processes. Among them are the widely accepted view that fixing a creature's physical nature fixes its mental nature, and the suggestion that consciousness and rationality are emergent, novel systemic properties that in some sense transcend the simpler properties of a system's parts.
The records do not argue about levels. They use them. The paper asks what the use does: whether the six levels add to the question, restate it, or substitute a description at one level for a choice that was posed at another.
0.3Where the records bear and where they cannot
The records can show how one record set frames a choice, what it says its framing costs and what its ladder does with the result. They cannot show whether any course in them is better, whether the goods they weigh are commensurable, or whether the physical and biological sentences in their notes are true. A record is not a person deliberating; it is a generated text in a fixed form. When this paper reads a record against Espeland and Stevens or against Potochnik and McGill, the reading is this study's, and it tests a lens on a text, not a theory on a population. The method, the selection rule and the counts are set out in Section 1.1; the two findings follow in Sections 1.2 and 1.3; the conclusion says what the findings add, where they stop and what could test them further.