When Algorithms Judge People, What Happens to Moral Responsibility?

In the same week that the APS AI Plan 2025 was released, charting a bold course for the modernisation of the Australian Public Service, the Australian Government has taken its next decisive step in the algorithmic era – the establishment of the Australian AI Safety Institute. According to the official announcement from Tim Ayres, Minister for Industry, Innovation and Science, this new institute will examine AI capabilities, coordinate multi-agency action, advise on regulation and ensure Australians are protected from “malign uses” of artificial intelligence.

On one level, this is welcome – it signals that the policy debate is moving beyond productivity talk and into the realm of accountability. The APS AI Plan envisages a public service infused with AI – chief AI officers in each agency, foundational training for staff, and systems built on trust, people and tools. Yet launching a dedicated safety institute raises a vital question – when algorithms make consequential decisions, about welfare claims, creditworthiness, risk assessments, who is morally responsible? When we talk about safety, we cannot merely talk about models, metrics and audit trails. We must ask – where does human responsibility lie?

The APS AI Plan offers a vision of efficiency, innovation and service-delivery. But the new institute underscores a tension -we are keen to embrace the power of AI, yet anxious about its risks. That tension is exactly where responsibility becomes ambiguous. The institute will advise and coordinate, but it will not decide. Yet the decisions are being made – the welfare claim is denied, the score is calculated, the model’s output becomes policy. In practice, institutions may lean on the institute’s recommendations while shifting moral weight onto the “system.”

A just society must maintain responsibility where it belongs -with the people who design, deploy and legitimise algorithmic systems, not with the systems themselves. The two announcements this week (the APS AI Plan 2025 and the establishment of the AI Safety Institute) therefore mark more than administrative reform, they mark a test of our collective moral capacity.

Bureaucratic decisions that once bore the unmistakable imprint of human judgment are increasingly mediated by algorithms. A welfare officer once assessing a claimant’s hardship now consults a risk score automatically generated from historical data. A bank manager who once interviewed applicants now relies on a creditworthiness model trained on correlations too complex for any human to parse. A parole board, pressed for time and political protection, turns to a predictive-policing tool that claims to estimate future risk more “objectively” than intuition ever could.

What has shifted is not merely technique, but morality. As we outsource judgment to machines, or rather, to the mathematical procedures we call machines, responsibility migrates. Where once we knew precisely whom to hold accountable for harmful decisions, we now confront a fog of abstraction. The system denies the claim. The model flags the risk. The algorithm calculates the score. And so, harm becomes reframed not as the moral failure of a person, but as the technical misfiring of a process.

This diffusion of responsibility is the most profound ethical transformation of the algorithmic age, and it is unfolding largely unnoticed by public debate. We speak endlessly of bias, transparency, fairness, and accuracy, necessary conversations, no doubt, but rarely do we confront the deeper assumption creeping into our institutions – that when algorithms judge, moral accountability evaporates. The ethical danger, in other words, is not that machines will replace human judgment, but that human judgment will hide behind machines.

Responsibility is a relational commitment – a willingness to stand behind one’s decisions, to answer for their consequences, and to face the people those consequences affect. It is precisely this willingness, deeply human, profoundly moral, that we risk losing in a world where automated systems mediate the most consequential choices of our lives.

Consider the denial of social welfare. In Australia, the ghosts of automated debt recovery still linger. An algorithm, flawed, simplistic, unforgiving, calculated overpayments by averaging earnings, presuming guilt, and issuing notices without human investigation. When challenged, the bureaucratic response was that the “system” had made an error, as if the system were a natural force, like a storm, rather than an artifact of human design. The public outrage that followed was justified not merely because the process was flawed, but because the moral chain of responsibility had been so thoroughly severed.

The phenomenon is not uniquely Australian. Across the world, individuals are interacting not with decision-makers but with decision-systems. Their frustration is familiar – “No one can tell me who decided this. No human will take responsibility.” The automated voice on the phone cannot explain why the loan was denied. The case worker says they have no authority to override the system. The government minister points out that the model was developed by experts. The experts claim they followed best practice. The private vendor insists the algorithm is proprietary. And so, accountability dissolves into a bureaucratic vapor.

We have entered an era of moral evasion by design.

Moral judgment is the capacity to understand the lived experience of another person. It requires empathy, narrative imagination, and situational discernment, none of which can be encoded in lines of code or statistical weights. When an algorithm determines creditworthiness, it does not see a family working three jobs, or a sudden illness, or the courage it takes to rebuild after financial devastation. It sees patterns. It sees probabilities. It sees categories.

The problem is not that algorithms use patterns, humans do this too. The problem is that once we outsource pattern recognition to systems, the stories disappear. And without stories, moral reasoning collapses into mere calculation.

Ethics is fundamentally about attention – the ability to genuinely see another person in all their particularity. Algorithmic systems, by contrast, see only generality. They do not attend; they sort. They do not listen; they classify. They do not answer; they output.

When we trade stories for scores, we reduce moral life to numerical abstraction. And when decisions are made in abstraction, responsibility becomes strategic rather than ethical – it becomes something to manage, to distribute, to obscure, rather than something to embrace.

One of the most pervasive myths sustaining algorithmic governance is the claim to neutrality. We are told these systems eliminate human bias, replacing subjective judgment with cold statistical fairness. But this is a comforting illusion. Algorithms do not eliminate bias; they metabolise it. They ingest the prejudices of our history and re-express them in sophisticated mathematical form.

When a model predicts criminal risk based on past policing data, it encodes the racial disparities in policing practices. When a credit scoring system penalises applicants from certain postcodes, it amplifies socioeconomic inequalities produced by decades of discriminatory housing policy. When an automated hiring tool downgrades CVs containing “non-white” names, it is simply reflecting the prejudices of the labour market it was trained on.

But even this critique, by now widely acknowledged, misses a deeper point – the moment we allow systems to obscure their own normative assumptions, we begin to treat those assumptions as if they came from nature rather than human choice.

Every model embodies explicit choices – Which data counts as relevant? What outcomes do we value? What types of errors are tolerable? Whose risk matters? Whose harm matters? These are not technical judgments; they are moral ones.

Yet they are rarely debated in Parliament. They are rarely disclosed to the public. They are often not even fully understood by the organisations using them. In many cases, the individuals building the models may not fully grasp the societal consequences of their design decisions.

A society that treats these choices as technical rather than moral is a society that has abdicated its ethical responsibility.

The moral shift we are witnessing is not toward immorality, but toward a new form of procedural morality, an ethic of compliance rather than conscience. If a system meets regulatory standards, if an organisation follows standard practice, if a model passes an audit, then the decision is considered ethically defensible. Responsibility is discharged through procedures, not through human reflection.

This is not ethics. It is risk management.

The distinction matters. Ethics requires judgment, humility, and the willingness to be wrong. It demands that decision-makers confront the moral weight of their actions. Risk management, by contrast, seeks to minimise exposure, not harm, but liability. The danger of algorithmic governance is that it encourages institutions to adopt the posture of risk management while claiming the legitimacy of ethical judgment.

We are drifting toward a world where harm can be explained away as an unintended consequence of a statistically valid model, a world where, because no individual directly caused the harm, no individual must take responsibility for it.

Some argue that algorithms merely reflect the scale and complexity of modern life. Human judgment, they say, is too slow, too inconsistent, too easily swayed by emotion. Machines, by contrast, can process vast quantities of data, make predictions at speed, and ensure decisions are consistent across populations. And in some respects, this is true. Algorithms can support humans in making fairer, more informed decisions.

But this is only ethically defensible if humans remain the locus of responsibility. The moment we treat systems as autonomous moral agents, or worse, as morally neutral machines, we erode the foundations of accountability in democratic society.

Responsibility is not a variable to be optimised. It is not a parameter to be tuned. It is not something that can be outsourced to a model and then reclaimed only when convenient. Responsibility belongs to the people who build these systems, who deploy them, who legitimise them, and who rely on them. It belongs to institutions, to governments, to corporations, but always, ultimately, to humans.

To deny this is to deny the very possibility of democratic ethics.

The most troubling ethical risk is not algorithmic bias itself, but the temptation it gives institutions to deflect accountability. When faced with political risk, bureaucratic pressure, or public scrutiny, institutions can point to the algorithm as a shield – “The model made the decision.”

This is a profound moral failure.

Machines do not make decisions. People do. Machines generate outputs. People interpret and act on them. A welfare algorithm cannot deny a claim. It can only produce a recommendation or a score. It is a human decision, to treat that score as binding, to remove discretionary override, to automate the process, that turns a system’s output into a life-altering judgment.

When institutions hide behind machines, what they are really doing is hiding behind themselves.

Democratic societies require not only fair outcomes, but transparent chains of accountability. Citizens must know not only what decisions are made, but who is responsible for making them. When algorithmic systems obscure this chain, we weaken the very mechanisms by which citizens can hold power to account.

Accountability must remain legible, even when decision-making processes become complex. It must be possible for a citizen to ask – Who designed this system? Who approved it? Who decided to rely on it? Who can change it? Who will answer for the harm it caused?

If no one can answer these questions, then the system is incompatible with democratic legitimacy.

What, then, is required? Not the abolition of algorithms, nor the naive belief that retreating to purely human decision-making will restore justice. Instead, we need a moral architecture that anchors responsibility firmly in human hands.

Three principles must guide us.

First, designers must acknowledge the moral dimensions of their choices. Data scientists and engineers must see themselves not as neutral technicians, but as moral agents shaping the contours of justice, welfare, opportunity, and punishment. With this power comes ethical obligation.

Second, institutions must resist the temptation to automate away accountability. Algorithms should support human judgment, not replace it. They should illuminate, not obscure. They should inform, not decide.

Third, society must refuse to treat algorithmic harm as a technical glitch. Harm is harm, whether delivered by a bureaucrat or a model. The moral responsibility lies not with the mechanism but with the humans who authorised its use.

Responsibility does not drift unless we allow it to.

In the End, We Are Answerable to Each Other

Morality is forged between us, it is woven from the relationships, obligations, and shared vulnerabilities that mark us as human. We cannot cede this to machines. We cannot accept a future in which moral responsibility is buried inside procedural systems, where harm is explained as a feature of the model, where the people affected become statistical exceptions rather than moral subjects.

In a just society, humans must remain answerable to humans.

Algorithms can assist. They can guide. They can reveal patterns invisible to our limited perception. But they cannot judge, not in the moral sense. Only people can do that. And only people can take responsibility for it.

As our institutions increasingly rely on algorithmic systems, we must insist on this simple, foundational truth – no algorithm is responsible for anything. Responsibility belongs to us, to the designers, the deployers, the policymakers, and the citizens who demand accountability. To allow that responsibility to drift from persons to processes is to undermine not only justice, but democracy itself.

The real ethical danger of the algorithmic age is moral abdication.

And the remedy, as always, is courage, the courage to stand behind our decisions, even when mediated by complex systems; the courage to answer for the harm that flows from structures we design; the courage to refuse the comforting illusion that machines can absorb the moral weight we would rather not carry.

Algorithms may shape our decisions. But they do not absolve us.

Not now. Not ever.

Keep Independent Journalism Alive – Support The AIMN

Dear Reader,

Since 2013, The Australian Independent Media Network has been a fearless voice for truth, giving public interest journalists a platform to hold power to account. From expert analysis on national and global events to uncovering issues that matter to you, we’re here because of your support.

Running an independent site isn’t cheap, and rising costs mean we need you now more than ever. Your donation – big or small – keeps our servers humming, our writers digging, and our stories free for all.

Join our community of truth-seekers. Donate via PayPal or credit card via the button below, or bank transfer [BSB: 062500; A/c no: 10495969] and help us keep shining a light.

With gratitude, The AIMN Team