Interpretability taxonomy: an approach for artificial intelligence developers technique selection
Bibliographic record
Abstract
As AI systems continue to increase their capabilities of performing human tasks there is a growing need to understand how the AI system determined its decisions. Interpretability is a concept in trustworthy AI research that is focused on understanding of the inner workings and decisions that come from the AI system. Our previous research revealed that AI developers lack consistent approaches or tools for implementing interpretability. There has been substantial theoretical interpretability research, yet the development of practical approaches in the form of tools to assist AI developers on interpretability remains underexplored in the research. This paper develops a taxonomy of AI interpretability techniques based on an analysis of the research literature using the survey of surveys method across 70 papers, examining 30 papers from 2019 to 2023. This paper develops a hierarchical taxonomy at a lower-level of abstraction to present and build upon relevant research literature. This paper also provides an approach as a practical tool in the form of decision trees to help AI developers identify and categorize interpretability techniques based on applicability and characteristics. Unlike existing interpretability taxonomies, this study introduces AI developer-oriented sample decision trees to aid in operationalizing the selection of interpretability techniques. This bridges theoretical research with practical implementation for AI developers. Preliminary testing and validation with AI developers was conducted to ensure applicability of the proposed taxonomy and decision trees. Preliminary validation demonstrated conceptual clarity and practical relevance, forming a foundation for future large-scale evaluation. This research contributes to further bridge the gap between AI systems and ensuring the practical implementation of interpretability.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.004 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 teacher head, 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".