AI Safety for Everyone

TL;DR

The paper critiques the dominant existential-risk framing in AI safety and demonstrates, through a systematic review of 383 peer-reviewed papers, that AI safety encompasses a broad set of practical concerns across the system lifecycle. It maps eight risk types and a spectrum of methodologies, showing substantial engagement with applied algorithms, safe RL, interpretability, and evaluation frameworks, alongside theoretical work. The findings argue for an epistemically inclusive, pluralistic conception of AI safety that ties into traditional technological safety and governance practices. This reframing broadens stakeholder engagement, encourages robust safety engineering approaches, and informs more actionable policy and governance decisions. The work highlights open questions about sociotechnical AI safety and the integration of non-peer-reviewed literature to fully capture the field’s evolving landscape.

Abstract

Recent discussions and research in AI safety have increasingly emphasized the deep connection between AI safety and existential risk from advanced AI systems, suggesting that work on AI safety necessarily entails serious consideration of potential existential threats. However, this framing has three potential drawbacks: it may exclude researchers and practitioners who are committed to AI safety but approach the field from different angles; it could lead the public to mistakenly view AI safety as focused solely on existential scenarios rather than addressing a wide spectrum of safety challenges; and it risks creating resistance to safety measures among those who disagree with predictions of existential AI risks. Through a systematic literature review of primarily peer-reviewed research, we find a vast array of concrete safety work that addresses immediate and practical concerns with current AI systems. This includes crucial areas like adversarial robustness and interpretability, highlighting how AI safety research naturally extends existing technological and systems safety concerns and practices. Our findings suggest the need for an epistemically inclusive and pluralistic conception of AI safety that can accommodate the full range of safety considerations, motivations, and perspectives that currently shape the field.

Figures (6)

• Figure 1: World cloud of morphologically standardised terms occurring in the abstracts and titles of selected papers weighted by their tf-idf score. We excluded stop words from the analysis.
• Figure 2: Graph of term co-occurrence in abstracts with binary counting, a minimum term frequency of 10, and a relevance score of at least 60%. The figure was produced using the VOSViewer tool vaneckSoftwareSurveyVOSviewer2010. Due to the large number of nodes, not all labels are shown in the figure.
• Figure 3: The number and percentages of the selected papers that address various risk types.
• Figure 4: The number of publications over time since 2015 for the different risk types that were identified in the paper as a result of a systematic literature review.
• Figure 5: Comparison of papers that provide only theoretical results without significant empirical testing against those that give sufficient evaluation of their proposed system. Papers are further grouped by whether they propose a concrete algorithm. Note, that the number of papers overlap as some propose solutions in more than one category.