ConceptNote-non-maleficence-Peters
Non-maleficence
Definition of non-maleficence
Non-maleficence means the obligation not to cause harm, avoiding actions who damage individuals, groups, or society. In AI ethics it is required that systems are designed and deployed in ways that do not produce physical, psychological, social, or political harm. It is distinct from beneficence which is the positive duty to actively produce good outcomes, and from justice, which concerns the fair distribution of both benefits and harms across groups.
Implications of commitment to non-maleficence
Committing to non-maleficence in AI design means taking responsibility for foreseeable harms before they occur, not only responding after the fact. To execute this, some of the key requirements include risk assessment, and meaningful human oversight, particularly for systems used in high-stakes domains like healthcare and criminal justice. It also means continuing to monitor after deployment, since harms may only emerge at scale or over time. This is not always easy to predict or prevent, as tools built with good intentions can often be repurposed for harmful ends. For example, a tool designed to locate refugee camps in order to plan resource distribution could just as easily be used by governments to tighten surveillance and control over border regions.
…
Societal transformations required for addressing concern raised by non-maleficence
Taking non-maleficence seriously requires change at several levels. Regulators need the technical capacity to meaningfully audit AI systems, so they can maintain real oversight over how these tools are used. More than anything, developers need institutional incentives that reward caution over speed. Currently, the pressure to move fast and outpace competitors often works against careful, responsible development. This would only work if implemented globally, since no single country or region wants to be the one left behind or restricted while others move ahead. Last, the broader public needs enough AI literacy to recognize and articulate when harm is occurring, otherwise the burden of identifying problems falls entirely on those who built them.
…