Zackary Okun Dunivin

Zackary Okun Dunivin, Bingyi Han, John Bollenbocher

LLMs act in the social world by drawing upon shared cultural patterns to make social situations understandable and actionable. Because identity is often part of the inferential substrate of competent judgment, ethical alignment requires regulating when and how systems invoke identity. Yet the dominant governance regime for identity-related harm remains bias mitigation, which treats identity primarily as a source of measurable disparities or harmful associations to be detected and suppressed. This leaves underspecified a positive, context-sensitive role for identity in interpretation. We call this governance problem bias negotiation: the normative regulation of identity-conditioned judgments of sociocultural relevance, inference, and justification. Empirically, we probe the feasibility of bias negotiation through semi-structured interviews with multiple publicly deployed chatbots. We identify recurring repertoires for negotiating identity including probabilistic framing of group tendencies and harm-value balancing. We also observe failure modes in which models avoid hard tradeoffs or apply principles inconsistently. Bias negotiation matters for justice because a positive role for sociocultural reasoning is required to recognize and potentially remediate structural inequities. But it is equally implicated in core model functionality as sociocultural competence is needed for systems that operate across heterogeneous cultural contexts. Because bias negotiation is a procedural capability expressed through deliberation and interaction, it cannot be validated by static benchmarks alone. To support targeted training, we introduce a broad but explicit framework that decomposes bias negotiation into an action space of negotiation moves (what to observe and score) and a complementary set of case features (over which the model negotiates), enabling systematic test-suite design and evaluation.

Zackary Okun Dunivin, Mobina Noori, Seth Frey et al.

Large language models (LLMs) allow for sophisticated qualitative coding of large datasets, but zero- and few-shot classifiers can produce an intolerable number of errors, even with careful, validated prompting. We present a simple, generalizable two-stage workflow: an LLM applies a human-designed, LLM-adapted codebook; a secondary LLM critic performs self-reflection on each positive label by re-reading the source text alongside the first model's rationale and issuing a final decision. We evaluate this approach on six qualitative codes over 3,000 high-content emails from Apache Software Foundation project evaluation discussions. Our human-derived audit of 360 positive annotations (60 passages by six codes) found that the first-line LLM had a false-positive rate of 8% to 54%, despite F1 scores of 0.74 and 1.00 in testing. Subsequent recoding of all stage-one annotations via a second self-reflection stage improved F1 by 0.04 to 0.25, bringing two especially poor performing codes up to 0.69 and 0.79 from 0.52 and 0.55 respectively. Our manual evaluation identified two recurrent error classes: misinterpretation (violations of code definitions) and meta-discussion (debate about a project evaluation criterion mistaken for its use as a decision justification). Code-specific critic clauses addressing observed failure modes were especially effective with testing and refinement, replicating the codebook-adaption process for LLM interpretation in stage-one. We explain how favoring recall in first-line LLM annotation combined with secondary critique delivers precision-first, compute-light control. With human guidance and validation, self-reflection slots into existing LLM-assisted annotation pipelines to reduce noise and potentially salvage unusable classifiers.

Zackary Okun Dunivin

Qualitative coding, or content analysis, extracts meaning from text to discern quantitative patterns across a corpus of texts. Recently, advances in the interpretive abilities of large language models (LLMs) offer potential for automating the coding process (applying category labels to texts), thereby enabling human researchers to concentrate on more creative research aspects, while delegating these interpretive tasks to AI. Our case study comprises a set of socio-historical codes on dense, paragraph-long passages representative of a humanistic study. We show that GPT-4 is capable of human-equivalent interpretations, whereas GPT-3.5 is not. Compared to our human-derived gold standard, GPT-4 delivers excellent intercoder reliability (Cohen's $κ\geq 0.79$) for 3 of 9 codes, and substantial reliability ($κ\geq 0.6$) for 8 of 9 codes. In contrast, GPT-3.5 greatly underperforms for all codes ($mean(κ) = 0.34$; $max(κ) = 0.55$). Importantly, we find that coding fidelity improves considerably when the LLM is prompted to give rationale justifying its coding decisions (chain-of-thought reasoning). We present these and other findings along with a set of best practices for adapting traditional codebooks for LLMs. Our results indicate that for certain codebooks, state-of-the-art LLMs are already adept at large-scale content analysis. Furthermore, they suggest the next generation of models will likely render AI coding a viable option for a majority of codebooks.