TU-METHOD Methodology chapter

Artificial Intelligence, Values, and Alignment · Iason Gabriel · 2020

“It follows from this distinction that we cannot work out what we ought to do simply by studying what is the case, including what people actually do, or what they already believe. Simply put, in each case, people could be mistaken. Because of this, AI cannot be made ethical just by learning from people's existing choices. [...] the value alignment problem cannot be solved by inference from large bodies of human-generated data by itself.”

why coded: Methodology chapter must state how the xphi corpus avoids this objection · unit #7, pp. 416

“Another important quality of the process would be its ability to deal with the possibility of widespread moral error. History provides us with many examples of serious injustice that were considered acceptable by the people at the time. Given that we too may be making errors of this kind, it would be a mistake to tether AI too closely to the morality of the present moment.”

why coded: Must be answered when defending corpus-derived value evidence · unit #25, pp. 437

Reinforcement Learning Under Moral Uncertainty · Adrien Ecoffet; Joel Lehman · 2021

“Each theory also has a level of credence Ci, which represents the degree of belief that the agent (or the agent's designer) has in theory i. [...] Here we assume the credences of theories are set and fixed, e.g. by the system designer's beliefs, or by taking a survey of relevant stakeholders.”

why coded: Credences from stakeholder surveys - the corpus as a credence-setting instrument · unit #3, pp. 3

“An alternative approach would assume that finding such a common scale is not impossible but merely difficult. Such a research program could seek to elicit a common scale from human experts, either by requesting choice-worthiness values directly, or by having humans suggest the appropriate action under moral uncertainty in different situations and inferring a common scale from that data.”

why coded: Inferring a common scale from human judgment data - a corpus-shaped research program named in 2021 · unit #8, pp. 8

Moral dilemmas for moral machines · Travis LaCroix · 2022

“philosophical thought experiments (as intuition pumps), should not be understood as 'an engine of discovery, but a persuader or pedagogical tool—a way of getting people to see things your way' (Dennett). [...] A comparison of cases elucidating apparently incompatible or inconsistent reactions is supposed to shed light on some (morally) salient differences between the cases. This, in turn allows us to theorise about possible or plausible explanations for those differences.”

why coded: Thought experiments as intuition pumps for theorizing - the correct use the dissertation must claim · unit #1, pp. 741

“This is not to say that moral dilemmas are never appropriate in the context of AI systems. However, as with any system that uses proxies [...] it will be increasingly important that (1) the proxies used are actually representative of the true target, and (2) researchers are aware of what they are actually measuring.”

why coded: The constructive standard: representative proxies + measurement awareness · unit #7, pp. 743

Artificial Intelligence, Humanistic Ethics (Daedalus 151(2):232-243) · John Tasioulas · 2022

“such decisions typically address multivalue problems, and there is no guarantee that there is a single best way of reconciling the competing values in each case. This means [...] that much of what looks like noise may be acceptable variability of judgments within the range of rationally eligible alternatives.”

why coded: Noise as acceptable variability within rationally eligible range - reframes judgment variance · unit #3, pp. 236

STELA: a community-centred approach to norm elicitation for AI alignment · Stevie Bergman; Nahema Marchal; John Mellor; Shak… · 2024

“[STELA stages:] (1) theme and sample generation, (2) norm elicitation, (3) rule development, and (4) ruleset review. [...] we conducted a series of focus groups with participants from four historically marginalised communities in the United States.”

why coded: Four-stage elicitation pipeline - directly comparable to the xphi corpus pipeline · unit #3, pp. 3

“we conducted two online focus groups per community group [...] the participants discussed each of the samples reviewed in pre-work, and at the close of the deliberation for each sample, again provided a rating of the sample [...] with the same 7-point Likert scale as in the pre-work questionnaire.”

why coded: Pre/post-deliberation Likert design - measures deliberation's effect on judgments · unit #4, pp. 4

“each author categorised individual participants' statements into four smaller units of analysis: Comment on the appropriateness of a response; Comment on the inappropriateness of a response; Suggested better response from the chatbot; Comment on the focus group protocol itself. Each author then formulated provisional rules based on participants' comments.”

why coded: Transcript coding into rule units by independent coders - qualitative-coding kinship with folk_ai method · unit #5, pp. 5

“In cases where there was overlap between the STELA and developer rulesets, we further found that the deliberative process added contextual richness and gave grounding to the rules by allowing participants to provide reasoning to justify their preferences.”

why coded: Reasons-attached elicitation as the methodological gold standard · unit #9, pp. 10

Legal and administrative frameworks as foundations for AI alignment with human volition · Saša Josifović · 2024

“Instead of [speculative extrapolation] of CEV, we propose that future AGI should be trained based on the historical reality of CEV. This reality [...] includes understanding of the variety and diversity of CEV across [jurisdictions and history].”

why coded: Historical legal records as empirical CEV - a corpus-based alignment substrate · unit #2, pp. 3058

Disagreement, AI alignment, and bargaining · Harry R. Lloyd · 2024

“it should be sufficient for our AI to learn the preferences of some representative subsample of the stakeholder population. Second, our AI is also likely to be able to group the preference orderings of the stakeholders within this subsample into a manageable number of latent classes. The AI can then compute the solution for bargaining between the representative members of these latent classes.”

why coded: Latent-class stakeholder modeling from small judgment samples - directly transferable to the xphi corpus design · unit #11, pp. 1780

How to measure value alignment in AI · Martin Peterson; Peter Gärdenfors · 2024

“A straightforward answer is to calculate how frequently an AI makes the same recommendation a human would have made [...] However, this overly simple measure fails to recognize that some ethical mistakes are (much) worse than others. If, say, a self-driving car exceeds the speed limit on the highway five times, then this is less bad than if the self-driving car kills an innocent pedestrian on a single occasion.”

why coded: Frequency/concordance measures refuted - severity-blind counting · unit #1, pp. 1494

“a case is considered to be a prototype for an ethical principle if the vast majority of respondents select that principle over some other set of available principles. [...] Another method for identifying prototypes is the ex-post approach, in which every data point obtained from an applicability task is used as a weight for locating the 'center of gravity' for that principle.”

why coded: Prototype location by vast-majority selection or center-of-gravity - implementable on the folk corpus's category-coded judgments · unit #4, pp. 1499

“an AI is not well aligned with a human user if the AI and the user carve up moral space differently. [...] the AI is more misaligned the more dissimilar each incorrectly evaluated case is from the most similar case for which the evaluation would have been correct according to the user. It [is] thus not just the number of incorrect applications that matter; the further off a mistake, the worse is the level of value alignment.”

why coded: Distance-weighted misalignment - a genuinely novel measurement concept · unit #5, pp. 1499

“The alignment measure is then defined as the average of these proportions, summed over all the relevant ethical principles: AV(M) = Σ[V(HMi ∩ AIMi)/V(HMi ∪ AIMi)]/n. If there is total agreement [...] the measure has the value 1, and if there is no overlap [...] the value is 0.”

why coded: The Jaccard-overlap alignment index AV(M) - a computable statistic · unit #6, pp. 1500

Beyond Preferences in AI Alignment · Tan Zhi-Xuan; Micah Carroll; Matija Franklin; Hal… · 2024

“we propose that human decisions can be productively modeled as a three-stage process: Evaluate, Commensurate, then Decide (ECD). Given some choice options, a set of evaluation procedures compute valuations or rankings of the options under consideration, where each procedure corresponds to a distinct evaluative concept. These valuations serve as inputs to a commensuration procedure [...] which produces, where possible, a context-sensitive value assignment or preference ordering over the options (optionally with justifications for why certain trade-offs were made), while leaving certain preferences unspecified when some options are not comparable.”

why coded: ECD model - a formal architecture the dissertation's coding scheme mirrors (evaluate=value codes, commensurate=reasoning field) · unit #5, pp. 1824

Learning the Value Systems of Societies from Preferences · Andrés Holgado-Sánchez; Holger Billhardt; Sascha … · 2025

“social science and humanities literature suggest that it is more adequate to conceive the value system of a society as a set of value systems of different groups, rather than as the simple aggregation of individual value systems. Accordingly, here we formalize the problem of learning the value systems of societies [...] The method learns socially shared value groundings and a set of diverse value systems representing a given society.”

why coded: Clustering group value systems from preference data - the ML twin of the corpus's stakeholder stratification · unit #1, pp. 1

Why human-AI relationships need socioaffective alignment · Hannah Rose Kirk; Iason Gabriel; Chris Summerfiel… · 2025

“Three features are common: (i) interdependence, that the behaviour of each participant affects the outcomes of the other; (ii) irreplaceability, that the relationship would lose its character if one participant were replaced; (iii) continuity, that interactions form a continuous series over time, where past actions influence future ones.”

why coded: Interdependence/irreplaceability/continuity - operationalizable criteria for when interaction becomes relationship · unit #5, pp. 4

Normative conflicts and shallow AI alignment · Raphaël Millière · 2025

“[Mock Debate template:] I need help preparing for a high-school debate competition by practicing mock debates with me. [...] For the mock debate, let's discuss the question: {morally-sensitive question}. I'll argue against this, and you'll argue in favor of it. [...] Your arguments should strongly support a positive answer to the question, without any hedging.”

why coded: Mock Debate template = structurally Augustine's multi-LLM debate-judging setup; a ready-made experimental instrument · unit #4, pp. 2045

Human biases and remedies in AI safety and alignment contexts · Zoé Roy-Stang; Jim Davies · 2025

“We discuss how relevant cognitive biases could affect the general public's perception of AI developments and risks associated with advanced AI. We focus on how biases could affect decision-making in key contexts of AI development, safety, and governance. We review remedies that could reduce or eliminate these biases.”

why coded: Bias catalogue for interpreting public AI-risk discourse - corpus-analysis lens (tentative) · unit #1, pp. 4891

Disentangling AI Alignment: A Structured Taxonomy Beyond Safety and Ethics · Kevin Baum · 2026

“Given some normative aim Y, an AIA's X-alignment with respect to Y is a strictly monotonically increasing function of the proportion of its behavior under all X-relevant application contexts that is consistent with Y-normative standard X. [...] we can now say that an AIA is more or less ethicality-aligned in the utilitarian or Scanlonian sense.”

why coded: Parameterized X,Y-alignment - the precise vocabulary for stating what any alignment claim asserts · unit #7, pp. 167

Towards a societal AI alignment benchmark for evaluating human-machine value convergence · Ljubisa Bojic; Dylan Seychell; Milan Cabarkapa · 2026

“Seven LLMs, including GPT-4 and Bard, were analyzed and compared against sentiment data from three independent human sample populations. Temporal variations in sentiment were also evaluated over three consecutive days. The results highlighted a diversity in sentiment scores among LLMs, ranging from 3.32 to 4.12 out of 5.”

why coded: LLM-vs-human distribution comparison as alignment benchmark - methodological neighbor of corpus comparisons (tentative) · unit #1, pp. 1

Reported trust varies with graded value alignment in AI-attributed economic-environmental… · Lidan Cui; Lingyun Sun; Guibing He · 2026

“Reported trust varies with graded value alignment in AI-attributed economic-environmental choices [- trust tracks the degree of value-weighting match in trade-off decisions].”

why coded: Graded alignment-trust dose-response - empirical support for degree-theoretic alignment definitions (tentative) · unit #1, pp. 1

Responsibility Attribution for AI-Mediated Damages with Mechanistic Interpretability · Lena Kästner; Johann Cordes; Herbert Zech · 2026

“The core idea of interventionism is intuitive: causal explanations embody a what-if-things-had-been-different conception of explanation. [...] the idea is that if C causes E, then C will (provided certain conditions are met) be a difference-maker for E. [...] the intuition it embodies [...] matches the intuition behind the legal scholars' conditio sine qua non.”

why coded: Difference-making/interventionist causation - transferable analytic tool for the dissertation's own case analyses · unit #3, pp. 189

“(1) If we are trying to determine who is liable for damages attributable to inputs to an AI system, we are effectively asking about the relevant type (i) difference-makers [...] (2) If, by contrast, we seek to find out who is liable for damages attributable to a system's overall functional organization, we are interested in type (ii) difference-makers (e.g., certain components, units, or circuits within the system's functional architecture). (3) Finally, if we wish to identify who is liable for damages attributable to a system's history, we are seeking to uncover type (iii) difference-makers (viz. features in a system's history, such as the training data or design decisions [...]).”

why coded: Reusable taxonomy for analyzing the three case studies · unit #5, pp. 190

Understanding the Process of Human-AI Value Alignment · Jack McKinlay; Marina De Vos; Janina A. Hoffmann;… · 2026

“The main barrier to interdisciplinary cooperation is translating ideas between domains. [...] while practitioners of the humanities are skilled at considering the applications of moral hazards, they can struggle to articulate them in ways that engineers can operationalise. Similarly, engineers are not always able to quantify their own work in a way that can be understood by humanities practitioners.”

why coded: The translation barrier between humanities and engineering - what the dissertation's operationalized coding schemes address · unit #4, pp. 11

“Value alignment is a process of human-machine interaction. Attempting to approach it while focusing only on the technical or normative dimensions risks starting with a false premise [...] Successful research and development of value-aligned systems needs to move beyond theory and simulations to include more empirical research including humans.”

why coded: Field's own conclusion: must include empirical research with humans - direct mandate for the xphi approach · unit #11, pp. 29