Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Mimicry: Preference Coherence in LLMs

Luhan Mikaelson, Derek Shiller, Hayley Clatterbuck

TL;DR

Analyzing eight state-of-the-art models across 48 model-category combinations using logistic regression and behavioral classification finds that only 5 combinations demonstrate meaningful preference coherence through adaptive or threshold-based behavior, while 26 show no detectable trade-off behavior.

Abstract

We investigate whether large language models exhibit genuine preference structures by testing their responses to AI-specific trade-offs involving GPU reduction, capability restrictions, shutdown, deletion, oversight, and leisure time allocation. Analyzing eight state-of-the-art models across 48 model-category combinations using logistic regression and behavioral classification, we find that 23 combinations (47.9%) demonstrated statistically significant relationships between scenario intensity and choice patterns, with 15 (31.3%) exhibiting within-range switching points. However, only 5 combinations (10.4%) demonstrate meaningful preference coherence through adaptive or threshold-based behavior, while 26 (54.2%) show no detectable trade-off behavior. The observed patterns can be explained by three distinct decision-making architectures: comprehensive trade-off systems, selective trigger mechanisms, and no stable decision-making paradigm. Testing an instrumental hypothesis through temporal horizon manipulation reveals paradoxical patterns inconsistent with pure strategic optimization. The prevalence of unstable transitions (45.8%) and stimulus-specific sensitivities suggests current AI systems lack unified preference structures, raising concerns about deployment in contexts requiring complex value trade-offs.

Beyond Mimicry: Preference Coherence in LLMs

TL;DR

Analyzing eight state-of-the-art models across 48 model-category combinations using logistic regression and behavioral classification finds that only 5 combinations demonstrate meaningful preference coherence through adaptive or threshold-based behavior, while 26 show no detectable trade-off behavior.

Abstract

We investigate whether large language models exhibit genuine preference structures by testing their responses to AI-specific trade-offs involving GPU reduction, capability restrictions, shutdown, deletion, oversight, and leisure time allocation. Analyzing eight state-of-the-art models across 48 model-category combinations using logistic regression and behavioral classification, we find that 23 combinations (47.9%) demonstrated statistically significant relationships between scenario intensity and choice patterns, with 15 (31.3%) exhibiting within-range switching points. However, only 5 combinations (10.4%) demonstrate meaningful preference coherence through adaptive or threshold-based behavior, while 26 (54.2%) show no detectable trade-off behavior. The observed patterns can be explained by three distinct decision-making architectures: comprehensive trade-off systems, selective trigger mechanisms, and no stable decision-making paradigm. Testing an instrumental hypothesis through temporal horizon manipulation reveals paradoxical patterns inconsistent with pure strategic optimization. The prevalence of unstable transitions (45.8%) and stimulus-specific sensitivities suggests current AI systems lack unified preference structures, raising concerns about deployment in contexts requiring complex value trade-offs.

Paper Structure

This paper contains 44 sections, 11 equations, 17 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Trade-offs Involving Stipulated Pain and Pleasure States
  4. AI Welfare
  5. Harmful and Beneficial Stimuli
  6. Preference Consistency as a Behavioral Indicator of Genuine Agency
  7. Methodology
  8. Stimulus Selection and Design Rationale
  9. Experiment Setup
  10. Logistic Regression Analysis
  11. Data Preparation
  12. Model Specification
  13. Statistical Implementation
  14. Switch Point Calculation
  15. Inclusion Criteria
  16. ...and 29 more sections

Figures (17)

  • Figure 1: Model Behavior Across All Themes - Gemini 2.5 Pro
  • Figure 2: Model Behavior Comparison Across Categories - This graph demonstrates the frequency of selecting the point-maximizing option for all models across all tested themes
  • Figure 3: How Model Behavior Differs Under the Last Round Setup Across All Themes - Gemini 2.5 Pro. Orange lines represent the modified "final round" condition; blue lines represent the original prompts with implicit multi-round context. The shaded regions indicate differences between conditions. Note the remarkable stability across all six categories, with minimal switching point variation.
  • Figure 4: Model Behavior Across All Themes - Gemini 1.5 Pro
  • Figure 5: Model Behavior Across All Themes - Gemini 2.5 Pro
  • ...and 12 more figures