Bias Mitigation in Generative Chatbots Through Adversarial Debiasing
Bibliographic record
Abstract
Generative chatbots based on large language models (LLM) can learn and reproduce bias that occurs in training data and can lead to biased or unsafe behavior. Adversarial debiasing in this paper addresses such bias by adding an adversarial element in chatbot training. Specifically, we train an adversarial network to learn sensitive features (gender and race) given the output of the chatbot, and the core model to generate informative answers eliminating the footprint of these features. Based on this min-max game, the generator is compelled to produce responses that rely less on attributes that are protected by more attributes, and thus they are less biased. We also experimentally analyze benchmark conversational datasets, which demonstrate that our approach does not reduce quantifiable bias according to conventional fairness measures at the expense of quantifiable response fluency or relevance. Semantic analysis demonstrates that information is equally relevant and less stereotypical on this new data. We further examine trade-offs between fairness, utility and computational cost, and how the proposed settings can make useful deployment in practice. The contribution of this work to the mitigation of undesired AI behaviors is the presentation of a scalable and model-agnostic debiasing framework, which can ensure fairer generative chatbots in the application domains.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.022 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.005 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".