MGMT 8825M Week 2 Principles Brief Example

Reasons and a Human Before Any Adverse Decision: A Principles Brief for a Staffing Agency's Automated Systems, Grounding One Principle in Explicability, Weighing the Charge That Principles Lack Teeth and Ranking Where It Must Apply First

Student Name

Walden University

MGMT 8825M: Ethics of Artificial Intelligence and Autonomous Systems

Instructor Name

Month Day, Year

Reasons and a Human Before Any Adverse Decision: A Principles Brief for a Staffing Agency's Automated Systems, Grounding One Principle in Explicability, Weighing the Charge That Principles Lack Teeth and Ranking Where It Must Apply First

The Principle

Brightline Staffing may not let an automated system make a decision adverse to a worker unless the worker can obtain the reasons for it and contest it before a human with authority to reverse it.

Why This Organization Needs It

Brightline, a composite staffing agency placing about 3,000 temporary workers in warehouses and call centers across Ohio and Indiana, now uses automated systems for three decisions that can harm workers: ranking who is offered shifts, deducting pay for late clock-ins and suspending workers whose client ratings fall below a threshold. Each system was bought to save coordinator time. None was designed with a route for a worker to ask why.

Grounding

The principle draws on the explicability principle that Floridi and Cowls (2019) added to the four principles long used in bioethics: beneficence, non-maleficence, autonomy and justice. They argued that AI requires a fifth because its decisions can be unintelligible to those affected and because someone must be accountable for them. Explicability, in their account, combines intelligibility, knowing how a decision was reached, with accountability, knowing who is responsible for it. The principle above turns that idea into an operational rule: reasons supply the intelligibility, and a human with power to reverse supplies the accountability.

A second ground lies in respect for persons. A worker suspended by a rating threshold she cannot see is being treated as an input to a process, not as someone owed an account of what is done to her. The demand for reasons is old; what is new is how easily automated systems let organizations skip it.

Context

Brightline would not be the first to adopt a principle of this kind. Reviews of published AI ethics guidelines have found frequent mention of transparency, fairness and accountability. Hagendorff (2020), evaluating 22 sets of guidelines, also found that they seldom include mechanisms to enforce their own commitments and that their effect on practice appears small. A principle that names a specific right, reasons and human review, and a specific trigger, an adverse decision, is easier to check than one that commits to transparency in general.

The Rival Worry

Mittelstadt (2019) argued that principles alone cannot guarantee ethical AI. Medicine's principles work, he observed, because they sit within a profession with common aims, a fiduciary duty to patients, long-established norms and real sanctions for breaches. AI development has none of these, so principles risk becoming statements that commit to little.

The worry applies directly to Brightline. The agency has no professional body watching it, and the coordinators who would carry out human review are measured on fill rates. A principle adopted without changes to those measures would likely produce reviewers who approve whatever the system decided. The brief accepts this criticism and answers it in part: the principle must come with a named owner, the operations director, a monthly count of contested decisions and reversals, and review time protected in coordinators' targets. These are not a profession, but they are the parts of one that a single organization can build.

Ranking Where It Applies First

The three systems differ in how much harm they can do and how hard their decisions are to reverse. Ranked by that measure:

1. Suspensions based on client ratings come first. A suspension removes income entirely, ratings are given by client supervisors with no obligation to explain them, and a worker currently has no way to see her score.

2. Pay deductions for late clock-ins come second. They are smaller and more frequent, but they are mechanical, so reasons are easy to provide: the clock-in record and the rule applied. The main gap is the lack of a human who can correct a faulty time clock.

3. Shift ranking comes last. Being offered fewer shifts harms workers, but the harm is spread out, and explaining a ranking to each worker raises practical difficulties the other two do not. The principle applies here too, but the first step is to publish the factors the ranking uses.

The order also follows the principle's own trigger. The more clearly a decision is adverse to a named person, the sooner the agency should be able to give that person reasons and a hearing.

References

Floridi, L., & Cowls, J. (2019). A unified framework of five principles for AI in society. Harvard Data Science Review, 1(1). https://doi.org/10.1162/99608f92.8cd550d1

Hagendorff, T. (2020). The ethics of AI ethics: An evaluation of guidelines. Minds and Machines, 30(1), 99-120. https://doi.org/10.1007/s11023-020-09517-8

Mittelstadt, B. (2019). Principles alone cannot guarantee ethical AI. Nature Machine Intelligence, 1(11), 501-507. https://doi.org/10.1038/s42256-019-0114-4