Contribution to the handbook
Priority · Assessed 2026-10-27
The question compares materially different human-control arrangements, from human-commanded action to reviewable or fully automated action. The evidence also warns that nominal review may not enable an effective challenge when a system is hard to understand, making this consequential to oversight design.
Does this challenge the goal?
In-scope gap
The question distinguishes levels of human control in an automated decision process. That informs the handbook’s open inquiry into authority, oversight, and correction without asserting that any one level is desirable.
The source example concerns tax enforcement and does not establish that the categories transfer unchanged to every governance function.
Why this category?
- Meaningful alternatives for allocating human control
- Consequential limits on review and override
- Relevance to authority and oversight
The source describes categories and an illustrative tax example, not the effectiveness or desirability of any arrangement in public governance.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
- How does the author illustrate different levels of human control in tax enforcement?
- Three levels of human control
- Review and possible override
- Human involvement may not be enough
- Algorithms may help find tax audit leads
- People may remain involved
- Human control in tax enforcement
- People can set or leave the search
Authority, accountability and resilience →
Read the exploration goalDescribe how the tax-enforcement example illustrates each level, from human-initiated action through human review to action without human involvement.
What we can say so far
Danaher illustrates three levels in tax enforcement: first, an algorithm selects audit targets only when a human agent requests that action; second, a system can process data continuously and produce recommendations, arrest warrants, or court summonses for human agents to accept or reject; and third, the process can be left entirely to machines, with no human involvement. The middle example gives people a review role, though the author’s framework distinguishes having a formal role from being able to challenge the system meaningfully.
What remains open
No material gap is recorded within this part’s current scope.
Conditions and limits
- The passages come from a readable author manuscript on an accessible mirror, not the publisher typeset version.
- The categories describe levels of human control; real systems may combine roles rather than fit neatly into one category.
- Formal human review or override does not establish that a person has enough information, understanding, or freedom to reject a system's output.
Objections to consider
- The author’s caveat limits what can be inferred from the review level: formal participation may not amount to effective control. The supplied passages do not resolve whether a reviewer can understand or successfully challenge the system.
Evidence and related ideas
Supporting evidence, qualifications and context play different roles in this answer.
context
People may remain involved An algocratic system may be fully automated, partly automated, or retain human supervision and input.
Establishes that algocratic systems can retain human supervision and input, providing general context for the examples.
The Threat of Algocracy: Reality, Resistance and Accommodation
Read full source: The Threat of Algocracy: Reality, Resistance and Accommodation (opens in a new tab) ↗Read source discussion: The Threat of Algocracy: Reality, Resistance and Accommodation →