Counting Worlds Branching Time Semantics for post-hoc Bias Mitigation in generative AI
Not on the map. Why?
Abstract (arXiv)
Generative AI systems are known to amplify biases present in their training data. While several inference-time mitigation strategies have been proposed, they remain largely empirical and lack formal guarantees. In this paper we introduce CTLF, a branching-time logic designed to reason about bias in series of generative AI outputs. CTLF adopts a counting worlds semantics where each world represents a possible output at a given step in the generation process and introduces modal operators that allow us to verify whether the current output series respects an intended probability distribution over a protected attribute, to predict the likelihood of remaining within acceptable bounds as new outputs are generated, and to determine how many outputs are needed to remove in order to restore fairness. We illustrate the framework on a toy example of biased image generation, showing how CTLF formulas can express concrete fairness properties at different points in the output series.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 16 of its 33 references that are in the map. First posted on arXiv 21 Apr 2026.
Builds on
16 of 33 references.
References · 16
- 2023Bias and Fairness in Large Language Models: A Surveycited by 639
- 2024Bias in Large Language Models: Origin, Evaluation, and Mitigationcited by 77
- 2024Investigating Bias in LLM-Based Bias Detection: Disparities between LLMs and Human Perceptioncited by 56
- 2024Current state of LLM Risks and AI Guardrailscited by 46
- 2022Quantifying Societal Bias Amplification in Image Captioningcited by 37
- 2024UniBias: Unveiling and Mitigating LLM Bias through Internal Attention and FFN Manipulationcited by 31
- 2017Parametric Markov Chains: PCTL Complexity and Fraction-free Gaussian Eliminationcited by 11
- 2025Bridging the Safety Gap: A Guardrail Pipeline for Trustworthy LLM Inferencescited by 8
- 2025BiasFilter: An Inference-Time Debiasing Framework for Large Language Modelscited by 6
- 2025Simulating a Bias Mitigation Scenario in Large Language Modelscited by 4
- 2022Checking Trustworthiness of Probabilistic Computations in a Typed Natural Deduction Systemcited by 3
- 2025DSO: Direct Steering Optimization for Bias Mitigationcited by 3
4 more
- 2021On Counting Propositional Logiccited by 1
- 2025Formal Verification of Markov Processes with Learned Parameterscited by 1
- 2025Trustworthiness Preservation by Copies of Machine Learning Systemscited by 1
- 2026A Linear Temporal Logic of Frequencies on Series of Eventscited by 1
Built on by
Not cited yet.