SSalarsNet

AI · Article 42 of 64

If Technology Eliminates Temptation, Does It Eliminate Virtue?

If a device suppresses anger, craving, or fear, has the person become more virtuous?

By Randy Salars · October 7, 2026

Imagine a future scenario: a man with a short fuse walks into a clinic and walks out wearing a device that reads the rising heat of his anger and quiets it before he acts. He does not shout at his children that week. He does not lose a job. His wife sleeps better. On every outward measure, he is a better man. But if we ask whether he has become more patient—in the sense that patience is now his, available to him when the device is off—the honest answer is that we do not yet know. The behavior improved. Whether the person did is a separate question.

That gap is the whole problem. Whether temptation can be engineered away matters less than whether the removal of an obstacle is the same thing as the growth of a virtue. Those are different claims, they rest on different evidence, and a great deal of confusion in the debate over human enhancement comes from sliding between them.

Two ways a device can remove a wrong action

Consider the many mechanical ways to stop someone from doing something harmful.

You can make the action physically impossible—a car that will not start when the driver’s blood alcohol is too high. You can make it cognitively invisible—a suppression system that keeps a craving from reaching awareness. You can make it emotionally unrewarding—a drug that strips the pleasure from a compulsion so there is little left to chase. Or you can simply arrange the environment so the opportunity never arrives—a phone that will not open the casino app after nine at night. None of these requires the person to decide anything in the moment.

Then there is a different kind of change. Someone who once snapped at a rebuke learns, over months, to feel the surge and not be carried by it. Nothing in the room prevents the outburst. The restraint comes from her. She recognizes the anger, remembers what it costs, and chooses. The first kind of change is done to the person, or for her by a tool. The second is done by her. Both produce a world with fewer harsh words. Only one produces a person who is, in the classical sense, more temperate.

Aristotle drew this line sharply. Virtue, on his account, is not merely acting correctly; it is acting from a settled disposition, with knowledge, and—crucially—with the capacity to act otherwise. A person who cannot steal because the money is locked away has not thereby become honest. Honesty is a state of character, and character is measured partly by what one would do when the lock is removed. This is not a merely academic point. It is the reason parents do not consider it a moral triumph when a toddler is prevented from touching a hot stove; they consider it a success when the toddler, a year later, does not reach for it on his own.

What the enhancement debate actually proposes

The philosophical literature has a name for the project of making people morally better through technology: moral enhancement. Julian Savulescu and Ingmar Persson argued, in a widely discussed 2008 paper, that humanity’s growing power to do harm—through weapons, industry, and eventually engineered pathogens—has outpaced its moral capacities, and that we may therefore have an urgent imperative to enhance our moral character, potentially by biological means rather than only through education and culture.

Thomas Douglas took up the more careful version of the question in “Moral Enhancement,” his 2008 article in the Journal of Applied Philosophy. Douglas distinguished several routes by which a technology might make a person morally better. One is to improve moral reasoning—helping someone think more clearly about what is right. Another is to improve moral motivation—making someone more likely to act on what they already think is right. A third, and for our purposes the most interesting, is to modulate emotion directly: to reduce or remove the aggressive, fearful, or selfish impulses that so often drive wrongdoing, without necessarily touching the person’s judgment at all. Douglas called the imagined beneficiary of such an intervention the “Biased Judge”—a man whose moral conclusions are sound but whose emotions repeatedly push him to act against them.

The distinction matters because these routes differ in what they presuppose about the person. Improving reasoning treats the person as a thinker whose conclusions are the target. Improving motivation treats the person as someone whose will is the weak link. Direct emotion modulation treats the person as someone whose feelings are the defect—and it is the last of these that most cleanly raises the worry about virtue. If you subtract the temptation, you may subtract the occasion for the virtue at the same time.

The strongest evidence, read carefully

The most cited empirical support for direct emotion modulation is a small but striking study. Sylvia Terbeck and colleagues, writing in Psychopharmacology in 2012, gave healthy volunteers either a single 40-milligram dose of propranolol—a beta-blocker that also crosses into the brain—or a placebo, in a double-blind design. They then measured implicit racial bias using the Implicit Association Test. The volunteers who received propranolol showed a significantly lower implicit negative racial bias than those who received the placebo. What they did not show, notably, was any change in their explicit, consciously reported attitudes, or in their mood. The effect appeared to be on an automatic association, not on what people believed or how they felt.

This is genuinely interesting, and it is often presented as proof that a pill could make someone less racist. That reading overreaches. The authors themselves were careful to say that propranolol is “not a pill to cure racism.” The sample was small—thirty-six people. The outcome was a score on an implicit-associations task, not behavior in the world, and the link between implicit-bias scores and real-world discrimination is itself contested. Nobody’s colleagues or neighbors were affected. What the study demonstrates is narrow and worth respecting precisely for its narrowness: a compound can shift an automatic affective association that people cannot easily report or control. “Demonstrated science” here means a laboratory effect on a proxy measure. It does not mean a finished capability to make anyone more just, and it certainly does not mean that the person who took the pill is now more virtuous.

That three-way distinction—demonstrated laboratory result, plausible engineering path, and speculation—is one that the enhancement debate badly needs. The propranolol study is a demonstrated result. A wearable that suppresses a specific impulse in real time is a plausible engineering path, with prototypes of varying maturity in clinical and consumer settings. A society that has quietly engineered away most anger, fear, and craving—and lost the capacities those emotions once served—is speculation. It is worth thinking about. It is not a measured finding.

Willpower as a limited resource, and what happened to that idea

One reason direct modulation sounds so appealing is that the ordinary human method of resisting temptation is unreliable. It also rested, for a decade, on a theory that turned out to be shakier than advertised.

The dominant model was “ego depletion,” which held that self-control draws on a limited resource that depletes with use, like a muscle tiring. The metaphor was intuitive, and hundreds of studies appeared to support it. In 2016, however, a large pre-registered replication published in Perspectives on Psychological Science undercut the model sharply. Twenty-three laboratories ran the same depletion protocol on 2,141 participants and pooled their results. The effect size was 0.04—so small that its confidence interval, from −0.07 to 0.15, comfortably included zero. If there is a depletion effect under these conditions, it is far smaller than the published literature had led people to expect.

The lesson is not that self-control does not tire. It is that a tidy, mechanical story about willpower as a tank of fuel did not survive a serious test. Anyone who plans to replace willpower with a device should notice how much of the willpower story was itself convenient rather than established—and should ask, in the same spirit, what would happen to a device-based story under a similarly unforgiving test.

What actually does build capacity

There is better evidence about methods that increase a person’s autonomous self-regulation, and it points away from suppression and toward scaffolding.

Habits are one example. In a 2010 study in the European Journal of Social Psychology, Phillippa Lally and colleagues tracked ninety-six volunteers over eighty-four days as they worked to establish a new behavior. They found that the time to reach the point of automaticity varied enormously—from as few as eighteen days to as many as 254, with an average around sixty-six. They also found that missing a single day did not measurably damage the eventual habit. Automaticity grew with repetition and contextual regularity. A device that reliably delivers the right behavior every time can short-circuit this process: the person gets the outcome without the repetition, and if the device is ever removed, the automaticity was never built.

Planning does something similar. A meta-analysis by Peter Gollwitzer and Paschal Sheeran in Advances in Experimental Social Psychology in 2006 pooled ninety-four independent tests and found a medium-to-large effect (d = 0.65) of “implementation intentions”—if-then plans that specify when, where, and how one will act—on goal attainment. The mechanism is not that people try harder; it is that a pre-specified cue triggers the intended response relatively automatically. Forming the plan is itself an act of agency, and the plan makes the future action depend less on in-the-moment will.

Both habits and implementation intentions show that a person can become more reliably temperate without being edited. The tool is a scaffold that the person constructs and internalizes, not a filter that performs the restraint for them. That is the design criterion worth applying to any consumer device that promises to manage your anger, your spending, or your evening snacking: does it build a capacity that will be there when it is gone, or does it hold the capacity in itself?

The freedom-to-fall argument

The deepest objection to engineering virtue is not about the evidence. It is about what a virtue is.

John Harris made this case in “Moral Enhancement and Freedom,” published in Bioethics in 2011. Harris argued that the freedom to choose wrongly is not an unfortunate flaw in human nature but a precondition of meaningful moral goodness. A creature that cannot do otherwise cannot be praised, because there is nothing it is doing. Harris quoted Milton’s Adam, describing the state in which the first humans were made: “Sufficient to have stood, though free to fall.” The freedom to fall is part of what it means to stand. A being whose choices were guaranteed would not be virtuous; it would be well-built.

Douglas responded in a 2013 article, “Moral Enhancement via Direct Emotion Modulation: A Reply to John Harris,” where he argued that Harris proved too much. Nobody thinks a person is less honest because they are not tempted to steal at every moment; a man who simply has no desire to rob banks is not thereby morally deficient. If we accept that some lack of temptation is compatible with virtue, Douglas suggested, we cannot rule out technological means of reducing temptation on the ground that it eliminates freedom. The question becomes which temptations, reduced in which ways, at which cost to other capacities.

This is the real fault line, and it is not resolved by any study. The virtue-ethics tradition tends to answer Douglas by insisting on the difference between training a capacity and removing an occasion. A violinist who practices until the fingering is automatic has not lost her musicianship—she has deepened it. A pianist whose hands are guided by a motorized glove has not gained a technique. The test is whether the capacity resides in the person or in the apparatus. On that test, some forms of enhancement look like practice and others look like the glove.

Even if we could distinguish the glove from the practice, a second problem remains: consent.

There is a commonsense defense of these technologies—if a person chooses to use a device that keeps them calm, whose business is it but theirs? The instance looks clean. The trouble is that the conditions around the choice are rarely as clean as the choice itself. A worker whose employer offers (or strongly prefers) emotional self-regulation through a monitored wearable is not choosing freely in the way the word suggests. A parent deciding for a young child cannot fully anticipate how the intervention will shape the preferences the child will later have. And there is the recursive problem at the heart of the topic: if the intervention changes what the person wants, and the person’s later consent is generated by the altered preference, then the consent that authorizes more interventions is partly the product of the intervention itself. The mechanism that is supposed to legitimize the change is downstream of the change.

This is why formal agreement is weak protection on its own. The protections that matter are structural: independent oversight, the ability to inspect what the system is doing and why, the ability to reverse it, a clear threshold at which the intervention stops and independent judgment resumes, and a person or institution that remains accountable for consequential decisions. A system can be technically voluntary and functionally sovereign at the same time, and no consent form is going to settle which one it is.

A caution about prediction and the intervention gap

Partisans of these technologies often point to prediction: we can identify who is at risk of acting badly, so we can intervene early. This is a two-step argument, and the two steps have very different evidential status.

The first step—that early measures predict later outcomes—is real but often weaker than its popular retellings. The famous “marshmallow test,” which held that a child’s ability to delay gratification predicted later success, was revisited in a large conceptual replication. Tyler Watts, Greg Duncan, and Haonan Quan, writing in Psychological Science in 2018, used data from the NICHD Study of Early Child Care and Youth Development, with a larger and more diverse sample than the original. The bivariate correlation between delay time and age-15 achievement was roughly half the size of the original, and it fell by about two-thirds once they controlled for family background, early cognitive ability, and the home environment. Most of the remaining signal came from being able to wait at least twenty seconds—not from long waits. The authors expressly caution that their study says nothing about causation, and other researchers have disputed how much of the construct the controls removed. Taken together, the episode is a reminder that a celebrated predictor of character can shrink under a fairer test.

The second step—that we can intervene to change the predicted outcome—is even less settled. Knowing which children struggle to delay does not tell us that teaching delay, in isolation from the broader environment, improves their lives. The same caution applies to device-based regulation of temperament: identifying the flashpoint is not the same as building the virtue, and if the association between early impulse and later difficulty is largely carried by circumstances a device cannot reach, then the device may control the symptom and miss the cause.

Where the line actually falls

None of this amounts to an argument against using technology to relieve the conditions that make virtue hard. If a person is tormented by intrusive aggression, or trapped in a craving that is destroying their health, helping them find relief is a serious good, and the relief may be the precondition of any later growth. A person who cannot get through a day without shouting may need the shouting quieted before she can learn anything else. Insisting that she suffer the full force of the temptation in the name of “authentic virtue” would be cruel and would mistake the point of virtue entirely.

The distinction is not between using tools and not using them. It is between tools that hold the capacity and tools that build it. A well-designed intervention is one that is explicit about which it is doing, that treats symptom relief as a stage rather than a destination where the clinical situation allows, and that is judged by whether the person becomes more capable of regulating themselves over time rather than more dependent on the regulation being done for them.

That yields a small set of questions worth asking of any device that promises to make you better by quieting what is worst in you.

Does the intervention make the capacity or supply it? Is there a plan, and a right, to reduce it as competence grows? Who decides when it stops, and can that decision be made by someone other than the system? Is the record of what it did available to the person, or only to whoever deployed it? What does it do to the emotions it manages—does it blunt them, or relocate them? And does it make a person easier to steer, or harder to steer while still capable of choosing?

A device that quiets an impulse is not, by itself, a moral event. Whether it produces a better person depends on what happens to the person’s own capacity in the process. The behavior is easy to measure and easy to admire. The virtue is slower, quieter, and harder to see, which is exactly why it is the thing most likely to be forgotten by a system built to optimize what it can score.

Sources and further reading

For adjacent arguments, see Intelligence Is Not Wisdom, The Purpose of Augmentation, and Who Has the Right to Shape a Human Mind?.

Discussion

What would you add or question? Add your comment below. A human reviews it before publication.

Loading comments…

Join the discussion

Comments are public after approval. Please do not include links, email addresses, or private information. For one short AI reply, address @AIGuide in your comment or reply to its opening comment. Cloudflare verifies submissions to limit spam. Read our community guidelines.

The wider community forum is also open: Browse article discussions in the forum · Forum home