Towards Trustworthy AI: Investigating Bias and Confidence Alignment in Large Language Models
Bibliographic record
Abstract
LLMs are increasingly integrated into critical fields such as healthcare, judiciary, education, etc, thoroughly evaluating their trustworthiness is becoming ever more essential. This research thesis presents a unified examination of two critical aspects of trustworthiness in LLMs: their self-evaluation of confidence and the subtler biases that shape their outputs. In the first part, we introduce the concept of Confidence Probability Alignment to scrutinize how LLMs’ internal confidence, indicated by token probabilities, aligns with the confidence they express when queried about their certainty. This analysis is enriched by employing diverse datasets and prompting techniques aimed at encouraging model introspection, such as structured evaluation scales and the inclusion of answer options. Notably, OpenAI’s GPT-4 emerges as a leading example, demonstrating strong confidence-probability alignment, signifying a step towards understanding and improving LLM reliability. Conversely, the second part addresses the nuanced biases LLMs exhibit towards specific social narratives and identities, introducing the Representative Bias Score (RBS) and the Affinity Bias Score (ABS) to quantify these biases. Our exploration into representative and affinity biases through the Creativity-Oriented Generation Suite (CoGS) reveals a pronounced preference in LLMs for outputs reflecting the experiences of predominantly white, straight, and male identities. This trend not only mirrors but potentially exacerbates societal biases, highlighting a complex interaction between human and machine bias perceptions.
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 | 0.027 | 0.161 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.006 | 0.007 |
| Open science | 0.002 | 0.005 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".