Source reconstruction · Proposed
Wisdom and AI
Research questions from Wisdom and AI · expanded from David Joseph’s analysis
Open questions
36 published items with open questions, grouped by the kind of work they need. An item can be accepted and still have work left on it.
Check an assumption 1
Test whether other options meet both needs and what slower development risks.Consider narrower capabilities, restricted access, safeguards and coordinated restraint. Review points alone do not resolve the conflict.Connection“Release the selected capability now.” conflicts with “Defer release of the selected capability.”Where do needs pull two ways?Find out what happens 14
Test whether uncertainty contributes to poorly justified choices.Uncertainty could also prompt restraint or better deliberation; neither effect is established here.Connection“Uncertainty about AI’s effects can make development and adoption choices difficult to justify.” contributes to “Some AI choices may proceed without an adequate account of their consequences.”What stands in the way?Find observations and a mechanism that distinguish this explanation.No supporting observations are supplied. Test whether the problem occurs, under which conditions, and what sustains it; do not infer a central cause from the diagram.Connection“The ends sought through AI alignment may be contested.” and “Value choices in AI alignment may remain implicit.” contributes to “Some apparently technical alignment disagreements may conceal ethical disagreements.”What stands in the way?Find observations and a mechanism that distinguish this explanation.No supporting observations are supplied. Test whether the problem occurs, under which conditions, and what sustains it; do not infer a central cause from the diagram.Connection“AI developers may face competitive pressure.” and “AI developers may perceive restraint as putting them at a disadvantage.” and “Coordination that could support restraint may be weak.” contributes to “Some actors may proceed with AI development despite recognising risks.”What stands in the way?Find observations and a mechanism that distinguish this explanation.No supporting observations are supplied. Test whether the problem occurs, under which conditions, and what sustains it; do not infer a central cause from the diagram.Connection“Some AI uses may delegate judgment that people would otherwise exercise.” and “Some AI uses may reduce opportunities for people to practise judgment.” contributes to “Some AI uses may weaken human agency.”What stands in the way?Find observations and a mechanism that distinguish this explanation.No supporting observations are supplied. Test whether the problem occurs, under which conditions, and what sustains it; do not infer a central cause from the diagram.Connection“AI’s possible moral standing may remain uncertain.” and “Decision criteria for responding to AI’s possible moral standing may be absent.” contributes to “Potentially important considerations about AI’s moral standing may receive inconsistent treatment.”What stands in the way?Test all four joint conditions and whether a stronger intervention exists.Review is only one candidate. Usable evidence, authority and incentives, and feasible alternatives are proposed joint conditions, not a guarantee; reject this intervention if another diagnosis is better supported.Connection“Decision makers use explicit review points to revisit AI choices against emerging evidence.” and “Usable evidence about the decision and its consequences is available.” and “Decision makers have authority and incentives to act on review findings.” and “Feasible alternatives to the current decision are available.” causes “AI choices can be revised when evidence changes their justification.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Make ethical aims and disagreements explicit.” contributes to “Making ethical aims explicit may expose assumptions hidden inside evaluations.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Making ethical aims explicit may expose assumptions hidden inside evaluations.” contributes to “Alternatives can be assessed against defensible criteria.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Introduce a specified wisdom-derived practice with an identified mechanism and limits.” contributes to “The selected practice may change attention, motivation or judgment through the identified mechanism.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“The selected practice may change attention, motivation or judgment through the identified mechanism.” contributes to “A relevant, specified aspect of decision making may improve.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Establish proportionate safeguards or coordination for the selected decision.” contributes to “Safeguards or coordination may change what actors can do or are motivated to do.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Safeguards or coordination may change what actors can do or are motivated to do.” contributes to “Restraint or redirection becomes practicable.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Decision makers use explicit review points to revisit AI choices against emerging evidence.” contributes to “Formal review may create unwarranted reassurance.”What will we change?Test this proposed mechanism and its credible alternatives.The source questions and review do not supply supporting empirical evidence. Specify the setting, mechanism and measurable effect before a test.Connection“Formal review may create unwarranted reassurance.” contributes to “Unwarranted reassurance may contribute to premature action.”What will we change?Check at the host 5
Check this candidate obstacle in the selected setting.These are possible obstacles for a future bounded test, not observations of an existing group.Obstacle“Wise” remains underspecified.What has to happen first?Check this candidate obstacle in the selected setting.These are possible obstacles for a future bounded test, not observations of an existing group.ObstacleThe causal diagnosis is unsupported.What has to happen first?Check this candidate obstacle in the selected setting.These are possible obstacles for a future bounded test, not observations of an existing group.ObstacleThe contribution from a tradition is vague.What has to happen first?Check this candidate obstacle in the selected setting.These are possible obstacles for a future bounded test, not observations of an existing group.ObstacleThe proposed change exceeds participants’ powers.What has to happen first?Check this candidate obstacle in the selected setting.These are possible obstacles for a future bounded test, not observations of an existing group.ObstacleSuccess could be confused with completing the process.What has to happen first?Other unfinished reasoning 16
Review the goal and justify each candidate necessary condition.GoalPeople and institutions approach AI wisely, with defensible aims, risks and responsibilities.What are we trying to achieve?Investigate what wisdom traditions contribute.Preserve disagreements between traditions and limits of analogy; the page asks rather than settles this.ConditionDecision makers understand benefits, harms and uncertainties.What are we trying to achieve?Examine traditions, disagreement, mechanisms and limits for this branch.Identify particular teachings or practices; do not assume traditions agree, their analogies transfer, or the candidate conditions are necessary or sufficient.ConditionAI’s intended ends are defensible and its pursuit of them is adequately supported.What are we trying to achieve?Examine traditions, disagreement, mechanisms and limits for this branch.Identify particular teachings or practices; do not assume traditions agree, their analogies transfer, or the candidate conditions are necessary or sufficient.ConditionDevelopment choices account for alignment uncertainty, catastrophe and alternatives.What are we trying to achieve?Examine traditions, disagreement, mechanisms and limits for this branch.Identify particular teachings or practices; do not assume traditions agree, their analogies transfer, or the candidate conditions are necessary or sufficient.ConditionAdoption decisions account for agency, relationships and institutions.What are we trying to achieve?Examine traditions, disagreement, mechanisms and limits for this branch.Identify particular teachings or practices; do not assume traditions agree, their analogies transfer, or the candidate conditions are necessary or sufficient.ConditionTreatment of AI responds responsibly to uncertain moral standing.What are we trying to achieve?Select an actor, capability, release decision and time horizon.This is a candidate cloud, not a recognised conflict for an actual decision. Relevant participants have not confirmed either need.Shared objectiveThe still-unspecified AI development decision is defensible.Where do needs pull two ways?Investigate how wisdom traditions might shape review practice.ChangeDecision makers revisit AI choices at explicit review points.What will we change?Select the tradition, practice, mechanism, limits and expected aspect of judgment.No practice or tradition is selected by this draft. Invoking wisdom does not establish an effect.ChangeIntroduce a specific wisdom-derived practice with a mechanism and limits.What will we change?Define substantive criteria and safeguards before testing.Completion of review must not become evidence that a risky decision is justified.ChangeRequire substantive criteria; completing review is not enough.What will we change?Select a bounded setting, decision and willing participants.The dependencies are candidate requirements, not proof of sufficiency or a commitment to conduct a test.ObjectiveOne decision practice can be responsibly tested in a defined setting.What has to happen first?Confirm prerequisites, authority and consent before assigning this action.Owner and date are unassigned. This is a proposed step, not an action commitment or completed work.ActionSelect one actor, decision and time horizon.Who does what?Confirm prerequisites, authority and consent before assigning this action.Owner and date are unassigned. This is a proposed step, not an action commitment or completed work.ActionRecord aims, empirical claims and disagreements.Who does what?Confirm prerequisites, authority and consent before assigning this action.Owner and date are unassigned. This is a proposed step, not an action commitment or completed work.ActionGather evidence that distinguishes the explanations.Who does what?Confirm prerequisites, authority and consent before assigning this action.Owner and date are unassigned. This is a proposed step, not an action commitment or completed work.ActionSpecify a guarded test; conduct it only where appropriate.Who does what?Confirm prerequisites, authority and consent before assigning this action.Owner and date are unassigned. This is a proposed step, not an action commitment or completed work.ActionCompare results with the prediction and alternatives.Who does what?These come from the commons' own record of unfinished reasoning, and from disputed items. No deadline, owner or priority is shown because none is recorded.