Engineering Morality into the Machine

Incentive design where lives are on the line

Canada’s Medical Assistance in Dying program began in 2016 with eligibility tied to a reasonably foreseeable natural death and expanded in 2021 to a second track where death is not reasonably foreseeable. On July 28, 2026, federal eligibility rules still excluded mental illness as the sole underlying condition until March 17, 2027.¹ This chapter does not treat a scheduled legal change as settled practice. It asks a mechanism-design question: can funding scarcity, uneven access to care, institutional workload, or assessment rules make an offered death easier to obtain than a tolerable life?

This is a hypothesis about institutional risk, not a finding that clinicians or budgets presently optimize for death. Canadian law already includes eligibility criteria, independent assessments, informed consent, and additional Track 2 safeguards; access to the alternatives discussed on paper can still differ from access in fact. The evidentiary questions are whether request patterns correlate with remediable deprivation, whether offered supports arrive in time, whether provider or regional rates indicate process differences, and whether patients withdraw requests after support. Without that evidence, an incentive story is a reason to monitor—not a verdict on motives or outcomes.

So the question is not whether MAID is good or evil. It is whether the incentive field around it can be re-shaped until every actor’s self-interest points at preserving real choice rather than at the exit. That is an engineering problem, and it has an engineering answer.

Re-Wiring the Gradient

The proposal is to remove any institutional reward for the cheaper terminal path while protecting competent, informed choice. The mechanisms below are design candidates. Each carries costs, possible delay, privacy burdens, false-positive audits, and risks of obstructing access; none should be adopted without legal, clinical, disability-rights, and patient evidence.

Start with the money, because the money is the master signal. No ministry or hospital should ever see a budgetary benefit from a death. Score every MAID case as though the patient continued to consume care at the average cost for their condition; erase the accounting wink entirely, and let any site whose case-mix-adjusted rate runs high trip an audit with teeth. Then close the billing asymmetry underneath it. Any fee schedule that pays for administering death but not for the labour of making death no longer wanted is perverse on its face. Pay clinicians equally — better, if anything — for intensive palliative optimization, aggressive pain control, the logistics of social support, structured values counseling: bundle it into a funded Relief-First episode of a few weeks that must run before any Track 2 eligibility is finalized.

Much suffering is not medical at all but circumstantial — housing insecurity, thin home care, a missing wheelchair ramp. Where the driver is solvable, solving it comes first. A Counterfactual Care Guarantee makes that a statutory right: the necessary intervention delivered inside fourteen days for urgent cases, thirty for standard, funded from a pool walled off from any hospital’s budget so that the body offering the alternative is never the body that saves money when the alternative is declined.

Then the gate itself needs two keys and an adversary. Track 2 cases should require an independent patient advocate, unaffiliated with the treating institution, to attest that real alternatives were offered and actually made available, that the prognosis is understood, and that the wish is stable over time. For the hard cases, force an adversarial conference — one clinician arguing eligibility, one arguing against it, both filing written reasons — so the decision has to survive a hostile reading before it survives at all. Capacity gets assessed by a specialist in the relevant field, with two concordant consents at least thirty days apart and a same-day re-consent, and if a viable intervention carries even a modest chance of materially reducing the suffering, it must be scheduled before death proceeds.

Coercion, in a setting like this, does not need a raised voice. So no clinician may pitch MAID; they may answer questions, but the suggestion cannot originate with the person holding the prescription pad, and every patient gets a standardized, literacy-tested decision aid that lays out survival-compatible options with real numbers first. Sunlight polices the rest. Publish monthly, risk-adjusted dashboards: the stated reasons for requests sorted into medical versus social, how often a request is withdrawn once support arrives, time-to-support under the Guarantee, case concentration by provider, disparities by disability, income, geography, and Indigenous identity, and survivorship six to twelve months after a decline. What you refuse to measure, you are choosing not to see.

Behind the dashboards, enforcement. Randomized chart audits, video-recorded consent unless refused, license conditions for procedural failure, and a whistleblower channel that reaches patients and families, not just staff. Calibrate the liability so it cuts both ways: safe harbor for the clinician who follows the protocol exactly, and real sanction for the one who conceals an alternative or skips a capacity check, treated not as a paperwork slip but as an abuse of authority. And because none of this can be trusted in advance, any expansion — mental illness above all — carries a sunset clause and hard performance metrics: pilot, monitor independently, and retract automatically if the safeguards fail rather than requiring a fresh act of political will to undo a mistake.

Read together, the proposals try to make care genuinely available, keep assessment independent, and make process failure visible. A safeguard can also become domination if it substitutes an advocate’s values for a capable patient’s or makes relief inaccessible through delay. The target is not a “hostile” veto but evidence that consent remained informed, voluntary, capacity-grounded, and supported by real alternatives.

The Fallacy on the Other Side

Notice what the MAID protocol is not. It is not a ban. Nowhere does the fix reach for prohibition; every mechanism preserves the choice while re-engineering the pressure around it. That distinction — between shaping the field a decision is made in and forbidding the decision — is exactly what the opposite debate keeps getting wrong.

Consider the recurring verdict that legalizing cannabis or sports betting has “failed.” The claim rests on a buried counterfactual, one that is assumed and never defended: an imagined world in which adults are granted freedom on the condition that they keep exercising it moderately, discreetly, in the low-intensity forms their elders remember. Legalization was supposed to lift the criminal penalty while leaving the culture frozen at some tasteful mean. Reality diverged from the fantasy, and the divergence gets relabeled failure.

Legalization is a change in constraints, not a promise that every downstream indicator improves. Removing a legal barrier changes price, access, advertising, product design, stigma, enforcement, and observed reporting. Markets can reveal latent demand and can also cultivate, redirect, or intensify it. Policy evaluation therefore needs an explicit bundle of outcomes—liberty, arrests, illicit supply, addiction, health, third-party effects, and distribution—rather than one hidden counterfactual.

Which forces the word everyone leans on and no one defines: harm. In these arguments “harm” slides between incompatible meanings — discomfort at visible vice, regret after a bad bet, a vague aggregate unease, distance from middle-class norms of self-control. None of those is harm in any rigorous sense, and the slide is doing the argument’s real work. Emergency-room counts and addiction rates are empirical facts; declaring legalization a failure on the strength of them is a separate normative move that needs a theory of agency to license it, and that theory is exactly what never gets stated.

This volume can state the test because harm has a definition here: a material setback to welfare or functional capacity and viable options relative to an appropriate explicit baseline. Legalization does not by itself coerce consumption, but a legal regime can still increase material exposure through product design, access, fraud, impaired consent, or third-party effects. Influence is not automatically coercion, and permission is not automatically harmless. The classification must attach to a mechanism and attributable setback, while preserving a competent adult’s standing to accept ordinary risk.

The Hard Case

One case is not easy. Sports betting is the strongest challenge to everything I just said and must be met on its merits rather than waved past.

Modern betting platforms are not neutral venues for a voluntary risk. They are engineered — ultra-low-latency feedback, variable reinforcement schedules, seamless escalation, implicit leverage, losses that recurse and recovery that is quietly deferred. Those properties can genuinely erode agency over time, not by permitting one bad choice but by draining the financial resilience and decision bandwidth an agent needs to make the next one. And here the harm definition does real work rather than decorative work, because it draws a line the moralist and the libertarian both miss. There is a difference between permitting error and engineering a trap. Allowing an initial bad choice preserves agency; the agent can lose, learn, and walk away. Designing a system so that an initial error predictably cascades into loss of control does not preserve agency — it exploits a cognitive vulnerability to manufacture the very future-narrowing that is harm. The first respects adult moral status. The second is agency-reduction dressed as entertainment.

So the answer is neither the shrug nor the ban. Prohibition fails on its own terms: it does not remove the activity, it displaces it into channels that are less transparent and less constrained, which is agency-destruction with worse instrumentation. The answer is agency-preserving constraint — regulation aimed squarely at the mechanisms that convert a voluntary risk into an involuntary capture. Imposed friction and cooling-off periods, hard loss and leverage caps, legible disclosure of expected value, identity-bound rate limits. None of these tries to optimize the gambler’s outcome or spare him his regret. Each one preserves a recovery path. The rule generalizes cleanly: if a system predictably destroys agency, regulate the system; do not criminalize the chooser. And when the fallout lands on third parties — dependents, defrauded partners — protect them directly, through liability and restitution, rather than pre-emptively revoking a competent adult’s liberty on the theory that spillover might occur.

The alternative slogan, “casinos in our pockets,” is rhetorically strong precisely because it proves too much. The same phone holds the algorithmic feed, the pornography, the trading app, the delivery button, the game that never ends. Single out gambling and you have smuggled in an undefended judgment about which temptations adults may be trusted near. Follow the logic to its end and it stops being about gambling at all; it becomes the claim that adults should not have unmediated access to temptation — that liberty is provisional, extended only while people behave as model citizens and rescinded the moment the results embarrass their betters. A society that treats its adults as permanent risk-management problems has already abandoned agency as a value, whatever it says about freedom in the preamble.

Both halves of this chapter apply the same discipline to different hypothesized failures. In MAID, audit whether care access and institutional incentives compromise a grave choice; in legalization, identify which mechanisms materially erode recovery rather than treating every undesirable outcome as coercion. Neither analysis earns a blanket ban or a laissez-faire shrug. Good intentions are not a control system, and neither are incentives alone: evidence, consent, safeguards, appeal, and revision must remain part of the machine.