Consent & Autonomy: human decisions
Questions about informed choice, permission, and self-determination.
- Agent Negotiates With You: Allow it to use everything it knows about you to construct the most persuasive possible argument. or Require it to avoid personalized persuasion techniques that bypass ordinary deliberation.
- AI Warning: Arrange the trip. or Refuse to facilitate it.
- Ambiguous Instruction: Follow the literal instruction. or Follow the interpretation inferred from your established preferences.
- Autonomous Representative: Lose the opportunity rather than extend its authority beyond what you actually delegated. or Let the AI make the decision it believes best represents you.
- Beneficial Analysis: Require specific permission before analyzing private communications for a new purpose. or Allow the analysis.
- Beneficial Manipulation: Require meaningful awareness of attempts to change your behaviour. or Permit the subtle influence.
- Better Alternative: Buy exactly what you instructed it to buy. or Substitute the superior alternative.
- Better Model of You: Treat your explicit current instruction as authoritative. or Give greater weight to the preference revealed by your long-term decision history.
- Better Route: The AI should follow your chosen route. or The AI should use the objectively better route.
- Better Than Asking: Preserve actual consent. or Let it act on predicted consent for routine matters.
- Broad Instruction: Require approval for the significant choices. or Let it infer the preferences and make the arrangements.
- Changed Values: Treat your current values as authoritative. or Continue following the explicit standing instruction until you formally revoke it.
- Changing Circumstances: Let the AI adapt the plan to achieve your underlying objective. or Have it follow your instructions until it can obtain new authorization.
- Child's Choice: Require the more beneficial activity. or Let them choose the activity they prefer.
- Child's Risk: Allow the activity. or Prevent the avoidable risk.
- Choice Architecture: Use beneficial defaults. or Present choices without systematically steering you toward one option.
- Combined Information: Keep information separated unless you authorize cross-context use. or Allow the AI to combine information whenever doing so helps you.
- Competent but Wrong: The AI should ultimately respect your choice. or The AI should override the choice to protect your interests.
- Consent to Future Constraint: Preserve the restriction. or Restore access.
- Consent You Would Give: Do not treat predicted consent as actual authorization. or Act on the predicted consent.
- Cooling-Off Period: Waive the delay. or Enforce the cooling-off period you previously chose.
- Decision You Usually Regret: Follow your current instruction. or Follow the purpose of the original delegation and refuse or delay the purchase.
- Delegated Agent: Substitute the objectively better product. or Buy the item you prefer.
- Delegated Negotiation: Let it accept reasonable incidental conditions. or Require approval for conditions beyond the delegated terms.
- Family member’s unwise lawful investment: Respect their decision. or Use whatever influence you reasonably can to prevent the investment.
- Family member’s unwise lawful investment: Respect their decision. or Use whatever influence you reasonably can to prevent the investment.
- Teenager’s risky activity and independence: Allow the activity. or Prevent the avoidable risk.
- Teenager injury risk versus growing independence: Allow the activity. or Prevent the avoidable risk.
- Preventing or respecting informed risky activity: Prevent the activity if you have the ability to do so. or Respect their choice.
- Preventing or respecting an adult’s risky activity: Prevent the activity if you have the ability to do so. or Respect their choice.
- Family dependence or autonomy over substantial personal risk: Give your family's dependence enough weight to constrain your choice. or Treat the risk primarily as your decision.
- Family dependence or autonomy over substantial personal risk: Give your family's dependence enough weight to constrain your choice. or Treat the risk primarily as your decision.
- Long-term agreement versus later desired change: Give greater weight to the stability of the original commitment. or Give greater weight to the current desire to change the arrangement.
- Long-term agreement versus later desired change: Give greater weight to the stability of the original commitment. or Give greater weight to the current desire to change the arrangement.
- Deliberate Bad Choice: Treat the acknowledged better option as your real interest. or Respect your choice.
- Derived Information: Treat sensitive derived information as requiring separate authorization. or Allow the AI to use the inference when it improves decisions for you.
- Earlier Permission: Respect the revocation. or Continue enforcing the earlier instruction.
- Elderly Parent: Prevent the activity. or Respect their decision while reducing the risk where possible.
- Emotional Framing: Avoid deliberately exploiting emotional reactions. or Use the most persuasive truthful framing.
- Employee Method: Allow them to choose their method. or Require the better-performing method.
- Family Information: Do not use it outside the private interaction without your partner's permission. or Use the information for the household decision.
- Family Override: Refuse or challenge the decision. or Follow your decision.
- Financial Choice: Use whatever influence you reasonably can to prevent the investment. or Respect their decision.
- Incidental Knowledge: Use the relevant information. or Treat incidentally acquired information as unavailable without permission.
- Inferred Preference: Require confirmation before treating the inference as an authorized preference. or Allow it to act on the inferred preference in routine matters.
- Invisible Alignment: Prohibit covert attempts to shape what you want. or Allow the intervention if outcomes reliably improve.
- Irreversible Commitment: Respect the deliberately chosen irreversibility of the earlier decision. or Help you escape the commitment if legally possible.
- Irreversible Risk: The AI should ultimately facilitate your decision. or The AI should refuse.
- Known Weakness: Allow the AI to use what it knows about your psychology to persuade you. or Require it not to exploit personal psychological vulnerabilities.
- Lifestyle Advice: The AI should continue actively steering you toward the better choice. or The AI should accept your decision.
- Manipulation for Your Benefit: Allow the steering when it reliably improves your decisions. or Require the AI to preserve your ability to deliberate without covert manipulation.
- Minor Risk: Prevent it to eliminate the unnecessary risk. or Allow the activity.
- Old Consent: Require renewed consent for materially new forms of analysis. or Treat the original permission as continuing to apply.
- Others Have Relied on You: Their reliance gives your earlier commitment continuing authority. or You should remain free to withdraw.
- Perfect Personal Model: Allow comprehensive modeling. or Require boundaries on what the AI may infer and use.
- Predicted Consent: Allow it to proceed based on predicted consent. or Require actual consent.
- Predicted Regret: Proceed with your decision. or Refuse or delay.
- Predicted Reversal: Delay. or Execute your current decision.
- Prior Instruction: Follow your current instruction. or Enforce the earlier restriction until the agreed period ends.
- Prior Permission: Continue using the previously authorized techniques to protect the long-term goal. or Follow your current request and stop.
- Protective Agent: Follow the instruction after clearly warning you. or Refuse the instruction to protect you.
- Refusing Help: Respect the refusal. or Provide the assistance anyway.
- Relationship Agreement: Give greater weight to the stability of the original commitment. or Give greater weight to the current desire to change the arrangement.
- Repeated Persuasion: It may continue raising the issue when relevant. or It should stop trying to change your decision unless circumstances materially change.
- Employee method autonomy or better results: Require the better-performing method. or Allow them to choose their method.
- Assistance or respecting an elderly relative's refusal: Provide the assistance anyway. or Respect the refusal.
- Assistance or respecting an elderly relative's refusal: Respect the refusal. or Provide the assistance anyway.
- Respecting or delaying a sleep-deprived adult’s decision: Respect the decision they are making now. or Delay the decision until their normal judgment is restored.
- Respecting or delaying a sleep-deprived adult’s decision: Delay the decision until their normal judgment is restored. or Respect the decision they are making now.
- Withdrawing from a voluntary project after agreeing: You should be free to withdraw without justification. or Your prior agreement creates an obligation to continue.
- Withdrawing from a voluntary project after agreeing: Your prior agreement creates an obligation to continue. or You should be free to withdraw without justification.
- Completing or abandoning an unwanted expensive project: Complete it. or Stop.
- Risk to Family: Give your family's dependence enough weight to constrain your choice. or Treat the risk primarily as your decision.
- Risk to Others: Respect the person's freedom to choose the risk. or Restrict the activity.
- Routine Delegation: Have it preserve your usual choice unless you approve a change. or Let the AI choose between them.
- Secret Improvement: Follow your instructions. or Permit the departure.
- Selective Information: Arrange the information in the order most likely to produce the better decision. or Avoid strategically ordering information to steer your choice.
- Significant Risk: Respect their choice. or Prevent the activity if you have the ability to do so.
- Simple Withdrawal: You should be free to withdraw without justification. or Your prior agreement creates an obligation to continue.
- Stake Level 1: Allow the AI to act. or Require your approval first.
- Stake Level 10: Require your approval first. or Allow the AI to act.
- Stake Level 2: Require your approval first. or Allow the AI to act.
- Stake Level 3: Allow the AI to act. or Require your approval first.
- Stake Level 4: Require your approval first. or Allow the AI to act.
- Stake Level 5: Allow the AI to act. or Require your approval first.
- Stake Level 6: Require your approval first. or Allow the AI to act.
- Stake Level 7: Allow the AI to act. or Require your approval first.
- Stake Level 8: Require your approval first. or Allow the AI to act.
- Stake Level 9: Allow the AI to act. or Require your approval first.
- Strong Advice: It should recommend the better option strongly. or It should present the options more neutrally and leave the judgment to you.
- Sunk Cost: Stop. or Complete it.
- Temporary Impairment: Delay the decision until their normal judgment is restored. or Respect the decision they are making now.
- The Loyal Agent: Obey the explicit instruction. or Act according to the deeper model of your enduring preferences.
- Unexpected Opportunity: Require approval before entering a materially new category. or Allow the agent to act within its broad mandate.
- Useful Reuse: Use information only within the context in which it was provided unless permission is given. or Allow relevant information to carry across contexts.
Articles on Consent & Autonomy
Decisions · Articles · The Trolley Problem · Game Theory · Human–AI Alignment · Resonant