The Algorithmic Executioner: Automated Termination and Moral Reasoning

Overview

The workplace is increasingly mediated by algorithmic management, the use of computer- programmed procedures to coordinate, evaluate, discipline, and sometimes remove workers from a labor system (Baiocco et al., 2022). The development is not limited to platform work. Similar methods now appear in warehouses, logistics, call centers, retail, finance, and human resources departments. The ethical issue is not simply whether the technology is efficient. The central issue is whether the system treats workers as moral agents and moral patients, or merely as data-producing inputs in a corporate optimization process.

This module examines automated termination, including account deactivation, algorithmically generated discipline, and software-supported firing decisions. These systems may collect location data, productivity data, biometric information, customer ratings, keystroke patterns, and other behavioral indicators. They may also generate scores that trigger warnings, performance improvement plans, suspensions, or termination recommendations. The moral question is therefore direct: when a person loses income, professional identity, benefits, and social standing because a model assigns a low score, who is morally responsible?

To answer that question, this module utilizes an applied empirical ethics framework. Begin with empirically verifiable facts, identify the parties of interest, clarify the relevant axiology, and apply competing moral principles. Finally, formulate a prescriptive conclusion. The module gives special attention to John Rawls’s Justice as Fairness, Robert Nozick’s libertarian theory of self-ownership and voluntary exchange, and Virginia Held’s feminist ethics of care. These theories do not merely produce different answers. They ask different questions about fairness, liberty, responsibility, dependence, and institutional power.

Just the Facts, Ma’am

Before making a moral judgment, we need to know what the technology actually does. Algorithmic management refers to systems that use programmed rules, statistical models, or machine-learning tools to allocate tasks, monitor work, evaluate performance, and shape employee behavior (Baiocco et al., 2022). The phrase sounds technical, but the basic structure is familiar. A workplace sets measurable targets. Software collects data. A model compares the worker to the target. The system produces a decision or recommendation.

The following sequence captures the basic structure of many automated workplace systems:

1. Data The system collects digital traces, such as GPS location, delivery time, productivity output, customer ratings, keystrokes, biometric markers, or attendance data.
2. Algorithmic Software compares the data to predefined targets, historical patterns, or model-generated predictions.
3. Decision The system produces a warning, score, ranking, risk flag, or termination recommendation.

 

4. Execution or The employer either automatically acts on the output or routes it to a manager, human resources officer, or support team for review.

The moral risk increases as the distance between the human decision-maker and the affected worker increases. A system that merely summarizes attendance data is different from a system that disables a worker’s account without meaningful review. The difference matters because termination is not a trivial administrative action. It can remove a person’s income, health insurance, professional reputation, and ability to meet ordinary obligations.

The empirical record gives us reasons for concern. Reporting on Amazon Flex described drivers who were monitored and sometimes terminated by software-assisted processes with limited human intervention. Drivers reported that locked apartment gates, poor weather, app problems, and other factors outside their control contributed to poor ratings or account deactivation (De Chant, 2021; Soper, 2021). Ibrahim Diallo’s widely discussed case illustrates a different form of automation failure. Diallo described being locked out of company systems after an HR database process treated him as terminated, even though his immediate manager did not intend to fire him (Doctorow, 2018).

These cases do not prove that all algorithmic management is unethical. They show something more limited and more important: automated systems can convert data errors, design assumptions, and bureaucratic gaps into life-altering employment outcomes. Once a process becomes automated, the harm can be difficult to interrupt because each part of the organization treats the software output as if it were authoritative.

Algorithmic systems also raise discrimination concerns. The EEOC has warned that AI and other software tools used in employment decisions can violate federal civil rights laws when they create unlawful disparate impact or discriminate against applicants or employees with disabilities (EEOC, 2022, 2023). Barocas and Selbst (2016) likewise argue that data-driven systems can reproduce discrimination through biased training data, proxy variables, and apparently neutral classifications. A system can appear objective because it uses numbers while still embedding unjust patterns from prior human decisions.

Human review does not automatically solve the problem. Bartosiak and Modlinski (2022) found evidence of algorithmic conformism, meaning a tendency to accept biased algorithmic recommendations in workplace discipline. The issue is not that humans disappear. The issue is that humans may remain formally present while becoming deferential, passive, or institutionally unable to challenge the automated recommendation.

Normative Systems Evaluation

A normative system is an organized set of facts, values, principles, duties, and conflict- resolution rules used to determine whether an action is morally required, morally prohibited, or morally permissible (Holcombe, 2025). Algorithmic termination cannot be evaluated by asking only whether the employer had a legal right to fire the worker. Legality and morality overlap, but they are not identical.

The parties of interest include all entities whose well-being, autonomy, rights, or legitimate expectations are affected by the decision. In an automated termination case, the most obvious party of interest is the worker. However, the worker is not the only party. Employers,

managers, software developers, customers, shareholders, coworkers, regulators, and the broader moral community also have interests at stake.

 

Party of Interest	Primary Interests	Primary Vulnerabilities
Worker	Income, dignity, procedural fairness, reputation, employment stability, opportunity to respond.	Loss of livelihood, opaque accusations, inability to appeal, surveillance pressure, discrimination.
Human manager	Operational efficiency, consistent discipline, reduced administrative burden, legal compliance.	Overreliance on automated recommendations, loss of moral agency, pressure to defer to system output.
Software developer or vendor	Functional system design, client satisfaction, compliance, technical accuracy.	Distance from workplace harms, limited understanding of local context, incentive to optimize measurable targets.
Employer or owner	Productivity, cost control, risk management, scheduling flexibility, property rights.	Legal liability, reputational harm, worker distrust, distorted data from poorly designed metrics.
Customers and clients	Reliable service, safety, low prices, accurate delivery or product quality.	Little knowledge of how ratings affect workers, possible participation in unfair discipline.
Moral community	Fair labor standards, social trust, nondiscrimination, accountable institutions.	Normalization of opaque decision-making and erosion of due process in private life.

 

We must also distinguish between moral agents and moral patients. A moral agent is an entity capable of understanding moral reasons and acting in ways for which it can be held morally responsible. A moral patient is an entity capable of being harmed or benefited, and therefore deserving of moral consideration. Workers are both moral agents and moral patients. They make choices, but they can also be harmed by opaque institutional decisions.

Is an algorithm a moral agent? No. Treating it as one commits a category mistake. A category mistake occurs when one attributes a property to an entity that belongs to a different logical category. Algorithms do not understand moral reasons, assume responsibility, experience guilt, offer justification, or make amends. The moral agents are the people and institutions that design, purchase, deploy, interpret, and authorize the system.

Key Concept: The Category Mistake

A company that says, “The algorithm made the final decision,” may be describing a workflow. It is not identifying a moral agent. A mathematical model can produce an output, but it cannot bear moral responsibility. Responsibility remains with the human beings and institutions that built and deployed the decision system.

Rawlsian Justice as Fairness

John Rawls’s Justice as Fairness asks us to evaluate institutions from the standpoint of rational representatives choosing principles from behind a veil of ignorance. Behind this veil, no one knows whether they will be wealthy or poor, manager or worker, healthy or disabled, citizen or immigrant, technically skilled or economically vulnerable (Rawls, 1971).

 

The thought experiment matters because it blocks self-serving reasoning. If you did not know whether you would become the platform owner or the low-wage delivery worker, would you design a workplace system that can terminate workers automatically, offer no intelligible explanation, and provide no human appeal? A rational person would have strong reasons to reject such a system.

Rawls’s first principle, the Equal Liberty Principle, protects equal basic liberties compatible with similar liberties for others. In the workplace context, the principle does not mean that every worker has a right to a specific job forever. It does mean that institutions should not impose arbitrary, opaque, or unreviewable processes that undermine self-respect, procedural standing, and fair opportunity.

Automated termination without explanation fails this test because it creates decisional opacity. The worker is harmed without being able to understand the grounds of the decision, correct false data, or challenge the interpretation of the evidence. From behind the veil of ignorance, one would not choose a system in which one’s livelihood could be removed by a process one cannot inspect or contest.

The second Rawlsian concern is the Difference Principle. Social and economic inequalities are permissible only if they work to the greatest benefit of the least advantaged (Rawls, 1971). A company may argue that algorithmic termination increases efficiency. That claim, even if true, is descriptive. It does not by itself establish that the practice is morally permissible. Treating efficiency as sufficient moral justification commits the naturalistic fallacy because it derives an ought directly from an is.

Under Rawlsian analysis, the burden of proof rests on the employer. The employer must show that the automated system improves the position of vulnerable workers rather than merely shifting risk downward. If the system concentrates benefits among owners and executives while increasing surveillance, instability, and unreviewable error for low-wage workers, then the system fails the Difference Principle.

 

Rawlsian Test	Question	Likely Result for Unreviewable Automated Termination
Equal Liberty Principle	Does the process respect basic procedural standing, transparency, and the worker’s ability to challenge harmful decisions?	Fails when the worker cannot receive a meaningful explanation or appeal to a competent human reviewer.
Fair Equality of Opportunity	Does the system allow workers to compete under fair conditions rather than hidden or arbitrary metrics?	Fails when workers cannot know which data are used or when irrelevant factors distort the score.
Difference Principle	Does the inequality created by the system benefit the least advantaged workers?	Fails when efficiency gains primarily benefit owners while workers absorb the risks of error.

 

The Rawlsian conclusion is not that every use of algorithmic management is morally prohibited. The narrower conclusion is that automated termination is morally prohibited when it lacks transparency, human review, appeal rights, and safeguards against disparate impact. A

Rawlsian system could allow algorithmic assistance, but not algorithmic sovereignty over a worker’s livelihood.

Libertarian Self-Ownership and Contract

A libertarian analysis begins from a different moral starting point. Robert Nozick emphasizes self-ownership, private property, and voluntary exchange (Nozick, 1974). On this view, the worker owns his or her labor, and the employer owns the capital, platform, software, and business infrastructure. Employment is morally legitimate when both parties voluntarily agree to the exchange.

From this perspective, algorithmic termination can be morally permissible if the worker knowingly and voluntarily accepted the terms. If the contract clearly states that performance will be monitored by software and that falling below a defined threshold can lead to deactivation, the employer may argue that the worker consented. State interference may then appear paternalistic because it blocks adults from entering contracts using their own labor and property.

This libertarian argument is stronger than many critics admit. A worker is not automatically wronged merely because a contract is harsh. Libertarianism does not require equality of outcome or Rawlsian protection for the least advantaged. It requires legitimate acquisition, voluntary transfer, and respect for self-ownership.

However, the libertarian argument has an internal limit: valid consent requires adequate disclosure and contractual fidelity. A contract is not morally binding if it is secured by fraud, deception, coercion, or material misrepresentation. If the employer promises objective performance review while the system penalizes workers for locked gates, restaurant delays, discriminatory customer ratings, or software glitches, the employer has not merely used a harsh contract. The employer has failed to uphold the terms under which consent was supposedly given.

The key libertarian question is therefore not simply, “Did the worker sign?” The better question is, “What did the worker actually consent to?” Consent to fair measurement is not consent to arbitrary measurement. Consent to customer ratings is not consent to being blamed for restaurant delays. Consent to GPS tracking is not consent to being penalized for a mapping error.

Libertarian Internal Critique

A strict libertarian can defend automated termination only when the terms are clear, the metrics are contractually relevant, the data are accurate enough to sustain the promised evaluation, and the employer does not hide material facts about the system. Otherwise, the problem is not state paternalism. The problem is defective consent.

Feminist Ethics of Care

Feminist ethics of care shifts attention away from abstract consent and toward relationships, dependence, vulnerability, and the moral significance of care. Virginia Held argues that moral reasoning must take seriously the concrete conditions of human dependency and the responsibilities that arise within relationships (Held, 2006).

From an ethics of care perspective, algorithmic termination is troubling because it replaces a workplace relationship with a scoring system. The worker becomes visible as a set of metrics while becoming less visible as a person. A model can register late deliveries, low ratings, or reduced productivity. It cannot understand illness, disability, childcare disruption, unsafe weather, abusive customers, or the cumulative strain of precarious work.

The ethics of care does not deny that employers need standards. It asks whether the institution has abandoned the duties that arise from dependence. Workers often depend on employers for income, stability, benefits, recommendations, and future opportunity. Employers also depend on workers for labor, reputation, and continuity of service. A morally adequate workplace cannot treat these dependencies as irrelevant merely because a contract exists.

Under care ethics, termination should be relationally accountable. This means that before a person is removed from work, a competent human representative should examine the context, listen to the worker, identify preventable harms, and consider whether support or accommodation is morally required. The goal is not sentimentality. The goal is responsible attention to vulnerability.

Comparative Synthesis

Ethical Theory	Central Moral Concern	Primary Question	Likely Judgment
Rawlsian Justice as Fairness	Fair institutions, equal liberty, protection of the least advantaged.	Would rational agents choose this system from behind the veil of ignorance?	Unreviewable automated termination is morally prohibited. Algorithmic assistance may be permissible with transparency, appeal, and human oversight.
Libertarianism	Self-ownership, private property, voluntary exchange, contractual fidelity.	Did the worker knowingly and voluntarily agree to the actual terms of the system?	Permissible when consent is valid and the system follows disclosed contractual terms. Prohibited when hidden criteria, defective data, or misrepresentation undermine consent.
Feminist Ethics of Care	Relationship, vulnerability, dependence, empathy, contextual responsibility.	Does the system preserve responsible human attention to the worker’s concrete situation?	Morally suspect when it replaces care and accountability with impersonal metrics. More permissible when embedded in humane review practices.

 

These theories are not merely three opinions. They are three different ways of organizing moral attention. Rawls asks what rules would be fair under conditions of ignorance about one’s social position. Nozick asks whether each party’s liberty and property rights were respected.

Held asks whether the institution preserved or abandoned morally significant relationships.

Legal and Policy Context

The legal context is developing unevenly. In the European Union, Directive (EU) 2024/2831 requires digital labor platforms to provide transparency about automated monitoring and decision-making systems. It also requires human oversight and states that decisions to restrict, suspend, or terminate a platform worker’s account or contractual relationship must be taken by a human being (European Parliament & Council of the European Union, 2024).

The United States follows a different pattern. Federal civil rights laws still apply when automated systems are used in employment decisions. The EEOC has emphasized that employers using software, algorithms, or AI remain responsible for disparate impact and disability discrimination under existing law (EEOC, 2022, 2023). However, U.S. law does not currently create a general federal right to human review for every automated termination decision. The result is a patchwork of anti-discrimination law, contract law, state employment law, and emerging local regulation.

The policy debate therefore mirrors the ethical debate. One side emphasizes efficiency, flexibility, and freedom of contract. Another emphasizes procedural fairness, worker vulnerability, and institutional accountability. The ethically serious position must avoid two mistakes. It should not assume that every algorithmic tool is unjust. It should also not assume that efficiency, scale, or contractual language is sufficient to justify terminating a person without meaningful explanation or appeal.

Model Policy Recommendation

A morally defensible automated termination policy should treat algorithmic systems as decision-support tools rather than final moral authorities. Employers may use automated monitoring to identify potential performance problems, but termination, suspension, deactivation, or equivalent economic harm should require review by a competent human decision-maker with authority to override the system. The worker should receive a clear explanation of the evidence, the relevant metric, the source of the data, and the procedure for appeal.

This policy is justified across competing moral theories. Rawlsian theory supports it because no rational agent behind the veil of ignorance would choose an opaque system that threatens the least advantaged without appeal. Libertarian theory supports it when human review is necessary to preserve contractual fidelity and prevent misrepresentation. Feminist ethics of care supports it because employment relationships create dependence and vulnerability that cannot be morally reduced to numerical scores. The policy does not prohibit algorithmic tools. It prohibits using those tools as unaccountable substitutes for human moral responsibility.

Frequently Asked Questions

What is automated termination?

Automated termination occurs when software systems contribute to or directly trigger the firing, suspension, or deactivation of workers based on algorithmic analysis of performance data.

Can AI legally fire employees?

In the United States, employers remain legally responsible for employment decisions made with AI assistance. Federal anti-discrimination laws still apply.

Who is morally responsible when an AI system fires a worker?

The moral responsibility remains with the humans and organizations that design, deploy, purchase, and authorize the system. Algorithms are not moral agents.

What is algorithmic management?

Algorithmic management refers to the use of software, data analytics, and AI systems to assign tasks, monitor performance, evaluate workers, and influence workplace behavior.

What ethical concerns arise from AI-based firing decisions?

Major concerns include lack of transparency, inability to appeal decisions, algorithmic bias, discrimination, surveillance, and the erosion of human accountability.

About the Author

Mark T. Holcombe is an educator, AI ethics consultant, and author of Critical Moral Reasoning: An Applied Empirical Ethics Approach. His work focuses on artificial intelligence, moral philosophy, workplace ethics, and emerging technologies.

References

Baiocco, S., Fernandez-Macías, E., Rani, U., & Pesole, A. (2022). The algorithmic management of work and its implications in different contexts. International Labour Organization and European Commission.

Barocas, S., & Selbst, A. D. (2016). Big data’s disparate impact. California Law Review, 104(3), 671-732. https://doi.org/10.15779/Z38BG31

Bartosiak, M. L., & Modliński, A. (2022). Fired by an algorithm? Exploration of conformism with biased intelligent decision support systems in the context of workplace discipline. Career Development International, 27(6/7), 601-615. https://doi.org/10.1108/CDI-06-2022-0170

De Chant, T. (2021, June 28). Amazon is using algorithms with little human intervention to fire Flex workers.

Ars Technica.

Doctorow, C. (2018, June 20). Fired by an algorithm, and no one can figure out why. Boing Boing.

Equal Employment Opportunity Commission. (2022). The Americans with Disabilities Act and the use of software, algorithms, and artificial intelligence to assess job applicants and employees.

Equal Employment Opportunity Commission. (2023). Select issues: Assessing adverse impact in software, algorithms, and artificial intelligence used in employment selection procedures under Title VII of the Civil Rights Act of 1964.

European Parliament & Council of the European Union. (2024). Directive (EU) 2024/2831 on improving working conditions in platform work. Official Journal of the European Union.

Held, V. (2006). The ethics of care: Personal, political, and global. Oxford University Press.

Holcombe, M. T. (2025). Critical moral reasoning: An applied empirical ethics approach. ISBN: 979-8263846435

Nozick, R. (1974). Anarchy, state, and utopia. Basic Books. Rawls, J. (1971). A theory of justice. Harvard University Press.

Soper, S. (2021, June 28). Fired by bot at Amazon: It’s you against the machine. Bloomberg News.