Trustworthy requirements for foundation models—A comprehensive survey and roadmap
Bibliographic record
Abstract
Foundation models are being broadly adopted for downstream tasks and then deployed in real-world systems due to their diverse and generalization capabilities. This versatility allows them to excel across various domains, providing a strong basis for building specialized models and solutions, thus, accelerating the process of artificial intelligence (AI) transition from research to real-world deployment. However, these foundation models and implemented AI systems driven by these models present many challenges, particularly in the area of trustworthiness. They might be vulnerable to adversarial attacks, output incorrect answers or decisions, biased against certain groups, prone to privacy leakage etc. This can cause severe outcomes, especially with the application of AI in high stake areas such as finance and healthcare. Thus, developing trustworthiness of foundation models-based AI systems has become important and necessary. Trustworthy AI systems ensure reliability, safety, and fairness, making them crucial for successful real-world implementation and user acceptance. The core questions under this survey topic are: How to define trustworthiness in foundation models? What trustworthy aspects should we take into consideration regarding foundation models? What approaches can enhance their trustworthiness? What are challenges and what future directions? In this survey, we present a comprehensive analysis of what constitute trustworthy foundation model. We summarized, analysed and discussed highly relevant trustworthy aspects for foundation models. To structure our analysis, we also formalized lifecycle of foundation model-based AI systems. This allowed us to specify the requirements and approaches for each stage of the lifecycle. Lastly, we outlined challenges and future directions towards trustworthy foundation models. The main contributions of this paper are four-fold: (1) Formalization of the lifecycle for foundation models and definition of each phase. (2) Summary of key trustworthy aspects of foundation models and define them. (3) Examination of approaches for each aspects across the lifecycle. (4) Identification of challenges, gaps, and future directions. • Aligning Key Trustworthy Criteria with Foundation Model Lifecycle Stages. • Comprehensive Survey on Approaches for Trustworthy Foundation Models. • Identifies trust challenges and future research paths for trustworthy foundation models.
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 | 0.016 | 0.056 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.006 | 0.004 |
| Science and technology studies | 0.002 | 0.005 |
| Scholarly communication | 0.009 | 0.017 |
| Open science | 0.003 | 0.005 |
| Research integrity | 0.004 | 0.006 |
| Insufficient payload (model declined to judge) | 0.004 | 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".