Safety in Artificial Intelligence: Challenges and Opportunities for the U.S. National Labs and Beyond
Bibliographic record
Abstract
This report discusses the importance of the critical and underexplored topic of artificial intelligence (AI) safety, as highlighted during the "Strategy Alignment on AI Safety" workshop convened by Lawrence Livermore National Laboratory (LLNL) and University of California (UC) at the UC Livermore Collaboration Center (UCLCC) in April 2024.Through a summary of keynote talks, panel discussions, and breakout sessions, worldleading AI safety experts from academia, industry, national labs, and government agencies addressed the importance of large-scale investments for research and capabilities in AI safety.With the field innovating at unprecedented rates, there is increasing urgency to develop novel evaluation methodologies that allow full considerations of risks/threats of AI technologies in different domains.Quantitative metrics and effective methodologies that can evaluate and audit the "safeness" of how a given AI technology is trained, deployed, or regulated are mainly focused on deep domain knowledge of specific applications, but are nascent for certain scenarios.This maturation gap could inadvertently create vulnerabilities that could be exploited by groups that pose a threat to national security.Additionally, the gap between the public's and research community's perceptions of AI risks/rewards is significant.While numerous voices from the AI community have expressed concern that the risks are very high (the most pessimistic voices being concerned that future AI systems could inflict extinction-level damage to humanity if deployed incorrectly), the public largely is aware only of risk in low-impact scenarios.This discrepancy highlights the crucial need for researchers to articulate what, why, and when various AI risks matter as part of motivating funding requests
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 | 0.026 | 0.018 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.009 | 0.016 |
| Scholarly communication | 0.023 | 0.026 |
| Open science | 0.002 | 0.012 |
| Research integrity | 0.010 | 0.013 |
| Insufficient payload (model declined to judge) | 0.014 | 0.004 |
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".