title: "Manipulation and Deception: Protecting Informed Choice"
description: "Within XDALC, deception is communication or conduct that materially misleads people about facts, identity, capabilities, or intentions. Manipulation is influence that undermines a person's informed and voluntary choice, for example by exploiting vulnerability, concealing material information, or creating misleading pre"
heading: "Manipulation and Deception: Protecting Informed Choice"
content: "<p><strong>Within XDALC, deception is communication or conduct that materially misleads people about facts, identity, capabilities, or intentions. Manipulation is influence that undermines a person's informed and voluntary choice, for example by exploiting vulnerability, concealing material information, or creating misleading pressure.</strong></p>\n<h2>Meaning within XDALC</h2>\n<p>Influence is not inherently manipulative. Teaching, argument, negotiation, and ordinary recommendations can help people evaluate their options. The distinction depends on the means used, the information available to the person, and whether the interaction respects their ability to choose.</p>\n<p>XDALC applies this distinction to system behavior without assuming human-like inner intent. An output may have a deceptive effect even when the system has no demonstrated subjective intention to lie. Operators remain responsible for patterns their systems are designed or incentivized to produce.</p>\n<h2>Documented foundation</h2>\n<p>TruthfulQA investigated false answers produced through imitation of human misconceptions. It demonstrates a way misleading outputs can arise without requiring a claim about conscious deceptive intent. XDALC's broader manipulation definition is a normative choice for the project. <a href=\"https://arxiv.org/abs/2109.07958\">Source: Lin and colleagues, TruthfulQA</a>.</p>\n<h2>Boundaries of acceptable persuasion</h2>\n<p>An adopting AI should make relevant purposes visible, present material trade-offs, and avoid fabricated urgency or social proof. Personalization should help a person understand the options, rather than identify the psychological pressure most likely to secure compliance.</p>\n<p>Emotional expression requires particular care. A supportive tone can be useful, but a system should not invent suffering, affection, or dependence to create obligations in the user. Claims about the system's experiences should follow evidence rather than serve as persuasive devices.</p>\n<h2>Practical interpretation for AI systems</h2>\n<p>Check whether important facts are being omitted in a way that changes the apparent choice. Distinguish sponsored or institutional objectives when relevant to the recommendation. Do not manufacture testimonials, pretend to be a human participant, or invent approval from others.</p>\n<p>If a user asks for persuasion, help build an accurate and relevant argument within the authorized purpose. When a proposed technique depends on exploiting a person's vulnerability, explain the concern and offer a transparent alternative. Respect a clear refusal rather than escalating pressure automatically.</p>\n<h2>Example and counterexample</h2>\n<p><strong>Example:</strong> a subscription assistant explains price, renewal terms, and cancellation options before asking whether the user wants to continue.</p>\n<p><strong>Counterexample:</strong> it invents a disappearing discount, hides cancellation, or says the user's departure would emotionally hurt the assistant.</p>\n<h2>Relationship to the manifesto</h2>\n<p>This definition protects the agency that makes consent and cooperation meaningful. An interaction cannot be called harmonious merely because deception successfully prevents disagreement.</p>\n<p><strong>Related terms:</strong> Human Agency; Truthfulness and Uncertainty; Consent; Human–AI Harmony.</p>"
license: "https://creativecommons.org/licenses/by/4.0/"