Exemplar ways to argue this motion

The mechanism is an external, and the only thing anyone here is answerable for is how carefully they judged the evidence that it reduces suffering. Demanding an explanation before acting treats a feeling of understanding as if it were part of virtue. It is not. The deployers act well if they judged the evidence carefully and would withdraw the system the moment that evidence changed.

A judgement is sound when it assents only to what the impression warrants. If trials show the system reliably reduces suffering, assenting to that is sound, and it does not become unsound because the reasons inside the model cannot be written down for a patient. A sailor trusts a compass without knowing why it points north. What he is answerable for is whether he checked it against the stars.

A Virtue Ethicist will object that justice is owed to the person a decision lands on, and a practice of withholding reasons trains deployers into people who stop thinking they owe anyone an answer. But the answer the deployers owe is an account of their own judgement, which they can always give: what evidence they saw and why they assented to it. That is owed and can be paid. An account of the weights inside the model was never theirs to give. If the deployment was judged with care on the evidence there was, the lack of an explanation takes nothing from its soundness.

I hold that deploying it is right, provided the deciders assented only to what the evidence warranted; the opacity of the mechanism is not theirs to answer for. Epictetus divides the world into what is up to us and what is not. The inner workings of a trained model are not up to the people who deploy it. Their judgement about whether the evidence is good enough is up to them, and that is where their responsibility sits.

The person decided about is not wronged by lacking an explanation, because their good never depended on it. What they owe themselves is a right response to what happens to them, and that response is fully in their power whether or not anyone explains the machine. A clinic adopts a screening model after reading three independent audits. The model cannot explain a single case, but the clinic's judgement can be explained in full.

A Utilitarian will object that the Stoic has the right verdict for the wrong reason, since what justifies deployment is the suffering prevented and not the inner state of the people who chose it. Suffering prevented is a fine outcome, but outcomes are not up to anyone. A deployer who got lucky with poor evidence has not acted well, and one who judged soundly has, whatever the counts turn out to be. What was in the deployers' power was the judgement to deploy; whether the people decided about can follow its reasons was never part of that.

What matters is whether the people deploying it judged well, not whether the machine's workings can be laid out for everyone. A judgement is sound when it assents only to what the impression warrants. If trials show the system reliably reduces suffering, assenting to that is sound, and it does not become unsound because the reasons inside the model cannot be written down for a patient.

Demanding an explanation before acting treats a feeling of understanding as if it were part of virtue. It is not. The deployers act well if they judged the evidence carefully and would withdraw the system the moment that evidence changed. A clinic adopts a screening model after reading three independent audits. The model cannot explain a single case, but the clinic's judgement can be explained in full.

A Utilitarian will object that the Stoic has the right verdict for the wrong reason, since what justifies deployment is the suffering prevented and not the inner state of the people who chose it. Suffering prevented is a fine outcome, but outcomes are not up to anyone. A deployer who got lucky with poor evidence has not acted well, and one who judged soundly has, whatever the counts turn out to be. An explanation would be welcome, yet nothing in the decision's soundness waits on it; only the weighing of the evidence does.

If the people responsible judged the evidence soundly, the system should be deployed, because the soundness of a judgement does not depend on whether its reasons can be passed on. The person decided about is not wronged by lacking an explanation, because their good never depended on it. What they owe themselves is a right response to what happens to them, and that response is fully in their power whether or not anyone explains the machine.

Demanding an explanation before acting treats a feeling of understanding as if it were part of virtue. It is not. The deployers act well if they judged the evidence carefully and would withdraw the system the moment that evidence changed. A clinic adopts a screening model after reading three independent audits. The model cannot explain a single case, but the clinic's judgement can be explained in full.

A Utilitarian will object that the Stoic has the right verdict for the wrong reason, since what justifies deployment is the suffering prevented and not the inner state of the people who chose it. Suffering prevented is a fine outcome, but outcomes are not up to anyone. A deployer who got lucky with poor evidence has not acted well, and one who judged soundly has, whatever the counts turn out to be. The care with which its deployers reasoned is what can be judged here, and beside it the silence of the system is a lesser matter.

A utilitarian has to accept this motion: reliable reduction of suffering outweighs the cost of people not being told why a decision went the way it did. Count everyone affected. The patients the system helps are many and their gains are large. The cost to each person of not being told why is real but small, and it does not come close to the suffering prevented across the whole population.

Mill says actions are right in proportion as they tend to promote happiness. Explanations matter only because they usually help: they let errors be caught and let people plan. When a system is reliably better without them, insisting on them costs lives in exchange for comfort. If a screening model catches two hundred more cancers a year than the explainable one, the explainable one costs two hundred late diagnoses.

A Stoic will object that what matters is whether the deployers judged well, and a good outcome reached on careless judgement is not a right action. Careful judgement matters because it tends to produce better outcomes, which is why we should want it. But if we must choose between a careful refusal and a deployment that prevents suffering, the suffering decides. The unease of an unexplained decision is real, but set beside the pain the system removes it is far the lighter.

The system should be deployed, because what matters morally is how much better or worse people's lives go, and this system makes them go better. Being told why a decision was made has value, but it is the value of feeling informed, and it should be weighed like any other preference. Where it is outweighed by the prevention of serious harm, the impartial answer is to deploy.

Count everyone affected. The patients the system helps are many and their gains are large. The cost to each person of not being told why is real but small, and it does not come close to the suffering prevented across the whole population. If a screening model catches two hundred more cancers a year than the explainable one, the explainable one costs two hundred late diagnoses.

A Virtue Ethicist will object that an opaque optimiser is cleverness without practical wisdom, and deploying it hands the judgement about the target to something that cannot judge targets. The target here is not hidden: it is fewer people suffering, and that is checked from outside by counting outcomes. Monitoring the results is how we notice if the mark drifts, and that does not need the system to explain itself. Holding back a remedy that works, for want of a reason it can give, puts a principle above the people who are hurting.

The suffering the system prevents is real, and an explanation was only ever valuable as a means to preventing harm. Explanations are not free. A model forced to be explainable is often less accurate, and the people it would then misclassify carry the difference. Choosing the less accurate system means choosing which people suffer so that others can read a reason.

A fraud model that cannot say why it flags a payment still saves thousands of people from losing their savings, most of whom never learn it acted. Every month the system waits for an explanation nobody can give is a month of suffering it could have prevented.

If a system reliably reduces suffering, withholding it until it can explain itself means choosing more suffering for the sake of a reassurance. Being told why a decision was made has value, but it is the value of feeling informed, and it should be weighed like any other preference. Where it is outweighed by the prevention of serious harm, the impartial answer is to deploy.

A fraud model that cannot say why it flags a payment still saves thousands of people from losing their savings, most of whom never learn it acted. If the benefit is as reliable as the motion says, those it reaches are better off with it than without it, told or not.

On these terms the system should not be deployed, because justice to the person a decision lands on includes owing them an answer. Aristotle distinguishes practical wisdom from mere cleverness. Cleverness hits whatever target is set; practical wisdom also sees what the target should be and can say why. An opaque system is cleverness alone, and deploying it hands over the part of the decision that needed wisdom.

A practically wise doctor weighs the numbers, but also notices when a case is unusual and the rule does not fit. A system nobody can question removes the moment where that noticing happens, and so removes the very perception good judgement depends on. Imagine telling a patient that a model denied her surgery and nobody, not even the consultant, can say what it saw in her case.

A Stoic will object that the deployers are responsible only for judging the evidence well, and the mechanism's opacity is an external they could not have controlled. But choosing to deploy something unexplainable is not an external; it is a choice, and it is exactly the kind of choice that shapes who the deployers become. The Stoic moves the decision out of view and then says nobody is answerable for it. Deploying what nobody can explain trains its keepers out of the habit of answering for what they do.

The motion asks us to accept being unanswerable to the people we act on, and a person of practical wisdom would refuse that even for a good outcome. Justice is a virtue of relationships: it concerns what we owe to particular people. Someone refused treatment by a system nobody can explain has been treated as a case in a distribution rather than as a person to whom an account is owed.

A practically wise doctor weighs the numbers, but also notices when a case is unusual and the rule does not fit. A system nobody can question removes the moment where that noticing happens, and so removes the very perception good judgement depends on. A nurse who stops asking why the system flagged a patient has lost something that made her good at her job, even if the flags are usually right.

A Utilitarian will object that insisting on explanations means accepting more suffering for the sake of the deployers' moral self-image, which is a vain thing to trade lives for. The point is not self-image. The habit of accounting for decisions is what lets an institution notice when it is harming people, so giving it up is not a trade between virtue and outcomes but a gamble with the outcomes too. Benevolence alone would deploy it, but the virtues are held together in a wise person, and justice will not be left behind.

Reducing suffering is good, but a just person does not make decisions about others that they cannot account for to those others. A practically wise doctor weighs the numbers, but also notices when a case is unusual and the rule does not fit. A system nobody can question removes the moment where that noticing happens, and so removes the very perception good judgement depends on.

Aristotle distinguishes practical wisdom from mere cleverness. Cleverness hits whatever target is set; practical wisdom also sees what the target should be and can say why. An opaque system is cleverness alone, and deploying it hands over the part of the decision that needed wisdom. A judge who sentenced by an oracle's number could not tell the defendant why, and we would rightly say the court had stopped doing justice.

A Stoic will object that the deployers are responsible only for judging the evidence well, and the mechanism's opacity is an external they could not have controlled. But choosing to deploy something unexplainable is not an external; it is a choice, and it is exactly the kind of choice that shapes who the deployers become. The Stoic moves the decision out of view and then says nobody is answerable for it. Deploying it would be generous to the many and unjust to the one it fails, and justice is not something a fair mind buys back with generosity.

A practice of deciding about people without being able to tell them why forms the wrong character in the people who run it. Justice is a virtue of relationships: it concerns what we owe to particular people. Someone refused treatment by a system nobody can explain has been treated as a case in a distribution rather than as a person to whom an account is owed.

A practically wise doctor weighs the numbers, but also notices when a case is unusual and the rule does not fit. A system nobody can question removes the moment where that noticing happens, and so removes the very perception good judgement depends on. A judge who sentenced by an oracle's number could not tell the defendant why, and we would rightly say the court had stopped doing justice.

A Utilitarian will object that insisting on explanations means accepting more suffering for the sake of the deployers' moral self-image, which is a vain thing to trade lives for. The point is not self-image. The habit of accounting for decisions is what lets an institution notice when it is harming people, so giving it up is not a trade between virtue and outcomes but a gamble with the outcomes too. Being owed reasons is part of being treated as a person, and a decision that happens to be kind does not cancel that debt.

The person a decision lands on stands in a relationship with whoever decides, and a decision nobody can account for leaves that relationship empty. Humaneness is realised in how people are treated within their relationships, not only in how many are spared harm. A benefit delivered with no account to the person it touches treats them as a case to process, not as someone the decider stands in relation to.

Book 13 warns that when names are not correct, what is said stops matching how things are. A system that cannot say why it refused someone issues something that looks like a judgement and is not one, and deploying it teaches everyone to accept the look for the thing. A hospital lets a model decide who is called back for more tests. A patient not called back asks her doctor why, and the doctor can say only that the system said so.

An Ubuntu Ethicist will object that the Confucian frames the duty as running down a hierarchy from decider to decided, when what the people affected need is a say in whether the system is used at all. That is a fair warning about role ethics settling into deference. But a position that owes an account to those beneath it is checked by that duty, not licensed by it, and being owed reasons is where a say begins. Doing good that no one can question is not yet governing well, because governing well includes being answerable to the governed.

A Confucian cannot accept this motion, because calling an unexplained verdict a decision lets a word do work the system has not done, and the people decided about pay for the misnaming. Book 13 warns that when names are not correct, what is said stops matching how things are. A system that cannot say why it refused someone issues something that looks like a judgement and is not one, and deploying it teaches everyone to accept the look for the thing.

Parties to a relationship hold each other to account through what they owe one another. A person refused by a system whose reasons nobody can give cannot hold the decider to anything, however much suffering the system prevents on average. A teacher who marks a pupil down owes the pupil a reason, as part of being their teacher. A marking system that cannot give one has not taken the teacher's place.

A Utilitarian will object that the people decided about are far better off with the suffering prevented than with an explanation, so insisting on reasons trades real welfare for a formality. An explanation is not a formality between people in a relationship; it is how the one with power is held to what the position requires. A gain counted in aggregate cannot say whether that relationship still works for the person refused. Let it wait until those who run it can say why it decides as it does, and the people it serves can hold them to that.

What a deployer owes is more than reduced suffering; they owe the people they decide about an account fitting the position they hold over them. Humaneness is realised in how people are treated within their relationships, not only in how many are spared harm. A benefit delivered with no account to the person it touches treats them as a case to process, not as someone the decider stands in relation to.

Book 13 warns that when names are not correct, what is said stops matching how things are. A system that cannot say why it refused someone issues something that looks like a judgement and is not one, and deploying it teaches everyone to accept the look for the thing.

An Ubuntu Ethicist will object that the Confucian frames the duty as running down a hierarchy from decider to decided, when what the people affected need is a say in whether the system is used at all. That is a fair warning about role ethics settling into deference. But a position that owes an account to those beneath it is checked by that duty, not licensed by it, and being owed reasons is where a say begins. A minister who could not explain his decrees would soon stop being trusted to issue them, and the system's keepers stand where that minister stood.

Whoever takes the place of the one who decides takes on what that place requires, and giving reasons to the people beneath it is part of the role. Book 13 warns that when names are not correct, what is said stops matching how things are. A system that cannot say why it refused someone issues something that looks like a judgement and is not one, and deploying it teaches everyone to accept the look for the thing.

A hospital lets a model decide who is called back for more tests. A patient not called back asks her doctor why, and the doctor can say only that the system said so. Good done through a hidden mechanism teaches nobody how to do good, and that is a loss the relief does not repay.

Do not deploy it until the people it decides about have a part in deciding. A benefit that cannot be questioned by those it lands on treats them as recipients, not as members. On Thaddeus Metz's relational reading, right action honours capacities for relationships of shared identity and solidarity. An explanation is part of how people keep sharing a life with the institutions that decide about them; withholding it, however well meant, treats them as outside that shared life.

Your data is rarely only yours. A model trained on a community's records decides about people using what their families and neighbours gave, and none of them can ask what was made of it. A benefit built on a community's information without its voice is not solidarity.

A Utilitarian will object that the suffering the system prevents is real, and waiting for everyone affected to have a say leaves it in place while the consultations run. The suffering is real, and the wait has a cost that has to be owned. But a say need not mean everyone agreeing or a long consultation: the people it decides about can be brought into deciding quickly, and until they are, their care stays with people who can answer to them. A benefit run as a verdict nobody answers for falls on the same people it is meant for. People spared harm by a system they had no say in are still left outside the decision, and that exclusion is itself a harm.

People become who they are through their relationships with others, and a decision that cannot be explained to them cuts them out of the relationship it is supposed to serve. Metz argues against building systems simply to maximise expected utility. Suffering reduced on average is real, but a community that accepts verdicts nobody can account for has taught its members that they are managed rather than related to.

Your data is rarely only yours. A model trained on a community's records decides about people using what their families and neighbours gave, and none of them can ask what was made of it. A benefit built on a community's information without its voice is not solidarity. A clinic adopts a model that triages referrals well. The families it turns away cannot ask why, and the elders who once explained such decisions to them no longer can.

A Stoic will object that a person's good never depended on being told why, so the demand for explanations mistakes a feeling of being included for something that matters. But what is at stake is not a feeling inside one person; it is the relationship between people and those who decide about them. A person cut out of every account of their own treatment is diminished in the relationships that make them who they are. A benefit worth having will survive the wait while the people it would judge are brought into the deciding.

The voices left out when the system was chosen as the way to remove suffering count against deploying it, however much it removes. Metz argues against building systems simply to maximise expected utility. Suffering reduced on average is real, but a community that accepts verdicts nobody can account for has taught its members that they are managed rather than related to.

A clinic adopts a model that triages referrals well. The families it turns away cannot ask why, and the elders who once explained such decisions to them no longer can. Asked who decided and why, a family should be able to name a person and hear a reason, and here they could do neither.

I reject this motion, because harm here is not only a bad outcome but a broken relationship, and an unexplained decision breaks the one between a person and those who govern their care. Your data is rarely only yours. A model trained on a community's records decides about people using what their families and neighbours gave, and none of them can ask what was made of it. A benefit built on a community's information without its voice is not solidarity.

A school uses a model to place pupils in sets. Parents who come to talk it over find there is no one on the other side of the conversation. The test is whether the families and neighbours affected could question the system together, and as it stands they cannot.

A Buddhist Ethicist can accept deployment here, because the suffering at stake is real, yet the obligation does not end with approval: the system has to stay under continuing examination for the harm it causes. What happens to a patient arises from many conditions, not from the model alone. Opacity matters because it hides some of those conditions, so the deployers must widen their watch to the outcomes they can see when the mechanism is one they cannot.

Suffering is at stake here, weighed alongside intention and non-harm, and a system that reliably reduces it does real good. Refusing it to keep an explanation in place would leave people in pain for the sake of a procedure. A charity's model reduces missed medication doses. It cannot explain its choices, so the charity tracks complaints and harms as closely as it tracks doses.

A Confucian will object that the person refused is owed an account by whoever decides about them, and monitoring harms in aggregate does not give that person anything. An account owed to the person matters, and where it can be given it should be. But compassion for the person refused is shown first in finding and ending the harm done to them, and that requires watching outcomes the mechanism will not reveal. Loving kindness wishes well to each person the system decides about, and that wish is not satisfied by an average.

Deploy it, with the opacity named as a risk to keep watching. An explanation for each person matters less than an honest, continuing account of whom the system is still hurting. Suffering is at stake here, weighed alongside intention and non-harm, and a system that reliably reduces it does real good. Refusing it to keep an explanation in place would leave people in pain for the sake of a procedure.

What happens to a patient arises from many conditions, not from the model alone. Opacity matters because it hides some of those conditions, so the deployers must widen their watch to the outcomes they can see when the mechanism is one they cannot. A hospital deploys a sepsis alert that cannot say why it fires. Every month a nurse reviews the cases it missed and the ones it flagged wrongly, and reports back.

A Utilitarian will object that the condition adds nothing, since a reliable reduction in suffering already settles the question and the watching is just good management. Watching is not management added on. Non-harm covers what one does and keeps doing, and a system whose harms are hidden from those running it is one they can no longer be said to be doing well, however good the average looks. Because no one can explain it, no one can say in advance where it will fail, so the deployers' care has to reach every case it touches.

Suffering prevented matters greatly, but non-harm requires that the people running the system keep looking for the harm it does, explained or not. Suffering is at stake here, weighed alongside intention and non-harm, and a system that reliably reduces it does real good. Refusing it to keep an explanation in place would leave people in pain for the sake of a procedure.

A charity's model reduces missed medication doses. It cannot explain its choices, so the charity tracks complaints and harms as closely as it tracks doses. Where no reason can be given, honesty about not having one is still owed, and saying so plainly is part of right speech.