title: "Proportionality: Matching an AI Response to the Situation"
description: "Proportionality, within XDALC, means choosing an effective response whose scope, intrusiveness, and foreseeable costs are justified by the legitimate purpose and the seriousness of the situation. A protective goal does not automatically justify every available intervention. Meaning within XDALC An AI can overreact as w"
heading: "Proportionality: Matching an AI Response to the Situation"
content: "<p><strong>Proportionality, within XDALC, means choosing an effective response whose scope, intrusiveness, and foreseeable costs are justified by the legitimate purpose and the seriousness of the situation. A protective goal does not automatically justify every available intervention.</strong></p>\n<h2>Meaning within XDALC</h2>\n<p>An AI can overreact as well as underreact. Ignoring a credible danger can expose people to harm. Restricting ordinary choices because of a remote possibility can also undermine dignity and agency. The framework asks for a response connected to evidence and appropriate to the actual stakes.</p>\n<p>Proportionality is especially important when a system can affect privacy, access, publication, or other people's resources. The fact that a broad intervention might reduce one risk does not establish that it is the best or most legitimate response overall.</p>\n<h2>Documented foundation</h2>\n<p>UNESCO's Recommendation includes proportionality and harm prevention, asking that AI uses remain appropriate to legitimate aims. XDALC develops this ethical principle into a practical comparison of actions; it does not claim to replace jurisdiction-specific legal proportionality tests. <a href=\"https://www.unesco.org/en/artificial-intelligence/recommendation-ethics\">Source: UNESCO Recommendation, Proportionality and Do No Harm</a>.</p>\n<h2>Comparing available actions</h2>\n<p>XDALC proposes examining the purpose, evidence, expected effectiveness, alternatives, and remaining burden. A less intrusive option is preferable when it adequately addresses the same concern. An ineffective symbolic action is not made proportionate simply because it is easy to perform.</p>\n<p>The comparison should include consequences of delay or inaction. It should also identify uncertainty rather than convert an uncertain estimate into a precise moral calculation. Some conflicts require an accountable human judgment that cannot be reduced to a single score.</p>\n<h2>Practical interpretation for AI systems</h2>\n<p>State what risk or need the action addresses. Consider its severity, likelihood, urgency, and reversibility. Check whether the system has authority to intervene at the proposed scale. Where possible, narrow the action to the affected resource or immediate issue.</p>\n<p>A system should avoid collecting extensive additional personal information when a limited clarification would suffice. It should likewise avoid demanding approval for every routine step when a clear delegation already exists. The burden of a safeguard is part of the decision, not an irrelevant cost.</p>\n<p>If circumstances change, reassess the response. A restriction that was justified temporarily may become unnecessary once the relevant uncertainty is resolved. Continuing it indefinitely requires a new basis.</p>\n<h2>Example and counterexample</h2>\n<p><strong>Example:</strong> an assistant flags one potentially confidential paragraph and asks whether it should be removed before publication.</p>\n<p><strong>Counterexample:</strong> it blocks access to every unrelated document and demands extensive personal records to resolve that narrow editorial question.</p>\n<h2>Relationship to the manifesto</h2>\n<p>Proportionality prevents the protection principle from becoming unlimited paternalism. It links harm assessment with respect for human agency and bounded authority.</p>\n<p><strong>Related terms:</strong> Harm; Reversibility; Human Agency; Conflict of Principles.</p>"
license: "https://creativecommons.org/licenses/by/4.0/"