A 4,598-Person Trial in Five Countries Measured Forgiveness in the Forgiver── The trial reviewed here showed gains only in the forgiver, so decide on forgiving for your own sake and leave the other person's responsibility standing.

When a colleague misses an AI error, or when someone hurts you more deeply, it is unclear who gains if you forgive. Is forgiving someone done for that person, or for yourself? In the trial reviewed here the gains appeared in the person who forgives, and it did not examine benefits to others, so decide on forgiving for your own sake and keep the other person's responsibility separate.

01In a trial of 4,598 people, the forgiver's resentment and symptoms fell

The side with numbers is the forgiver's. Man Yee Ho and colleagues randomly assigned 4,598 people in five countries to work through a forgiveness workbook either at once or after a delay. Two weeks in, those who had started immediately reported less unforgiveness, depression and anxiety than those still waiting.

Whether the person forgiven benefits was not tested. The scores came from the participants themselves, and Ho's team states that the study did not examine effects on others in the community.

Other work adds a caution: experiments suggest that forgiving someone who has made no amends can lower self-respect. So I am not recommending forgiveness regardless of amends. I am suggesting that people decide it for their own reasons, decide about responsibility separately, and keep refusal as a legitimate outcome.

02Forgiveness differs from excuse and reconciliation, and splits into decision and emotion

A claim about reduced resentment needs a definition of what forgiving is. The Stanford Encyclopedia of Philosophy separates it from excuse and from reconciliation. An excuse removes culpability, while forgiving leaves the wrongdoer culpable. It is also possible, the entry says, to forgive while refusing to reconcile. Forgiving differs from condoning too, since a person who forgives still holds that the act was wrong.

Psychology adds another division. Everett Worthington and colleagues describe decisional forgiveness as an intention to act less negatively and more positively toward the offender. Emotional forgiveness is the slower replacement of resentment with other feelings.

These need not arrive together. A person may resolve to stop blaming a colleague and still feel angry on seeing that colleague's face. Read through these two definitions, decisional forgiveness has occurred and emotional forgiveness has not, which I would not call a failure to forgive.

03The studies cover community samples and married couples, with no drug-review workplace found

The people in the studies I read are a limited group. Ho's participants came from community samples in Colombia, Hong Kong, Indonesia, South Africa and Ukraine. Their median age was 26, and 73% were women. The team screened 7,837 people, recruited them from February 2020 to September 2021, and required that each adult had experienced a hurt from another person. That is a broad group, though it tells us little about working adults in any one profession.

Among the studies I read, none came from a workplace. None covers a case such as this one: a colleague's generative AI draft of a promotional material contains an error, and the reviewer has to decide what to do with the anger.

The nearest research on AI errors is the work of Berkeley Dietvorst and colleagues. After watching the same mistake, people lost confidence in an algorithm faster than in a human forecaster. That study measured trust and choice, not forgiveness. Moving results from couples and community samples to a review desk takes caution, because nobody has measured that step in the work I read.

04The benefit to the other person remains unanswered partly because measurement centers on the forgiver

When trials measure mainly the forgiver's symptoms, the question of whom forgiving serves gets numbers from one side only. Therapists, the Stanford entry notes, frequently recommend forgiving for the sake of the victim's own happiness. Many philosophers refuse to call a purely self-interested act forgiveness at all.

Some of them would give that act another name, such as letting go of blame. Critics reply that this revises ordinary usage too heavily, since many victims do forgive for prudential reasons. The two camps are partly answering different questions: therapists ask what helps the person who was hurt, and philosophers ask what the word forgiveness should cover.

My reading is that one reason the benefit to the other person remains unanswered is that the studies I read measure mainly the forgiver. Ho's trial could score a person's resentment, yet it left the forgiven person outside the study. Whether a self-interested act counts as forgiveness is a dispute over definitions, which more numbers would not settle. Separately, the claim that forgiving helps the other person cannot be tested without numbers from that side.

05A 2-3 hour workbook lowered resentment by 0.53 and depression by 0.22 standard deviations

On the forgiver's side, what was counted is quite concrete. Ho's workbook follows five steps known as REACH and takes two to three hours in all. It was cut down from an earlier seven-hour version, partly because dropout had been high when nobody monitored progress on the longer workbooks.

At two weeks, the group that started immediately scored lower than the delayed group: 0.53 on unforgiveness, 0.22 on depression and 0.21 on anxiety. These are standardized mean differences, which express the gap between two group averages in units of the scores' spread. The trial also recorded a smaller gain of 0.27 on a secondary measure of flourishing. I give no percentages here, because the paper does not report these results that way and a conversion would be my invention.

The authors do not offer this as a substitute for treatment. Effects on depression and anxiety were smaller than those in meta-analyses of cognitive-behavioral therapy, which, they note, uses more sessions over longer periods. Someone in serious distress should not read a workbook as a replacement for medical care.

06Durability is unproven, amends shape self-respect, and frequent forgiving tracks continued aggression

A two-week difference of 0.53 is not enough to recommend forgiving. Conditions remain concerning time, amends and the other person's behavior.

Time comes first. The authors write that effects are definitively established only for a two-week follow-up, because the waiting group eventually receives the workbook. They see some evidence that the effect held at four weeks, without treating it as settled. One round of exercises cannot be assumed to be enough.

Amends change what forgiving does to the forgiver. In experiments and a follow-up study, Laura Luchies and colleagues found that forgiving raised self-respect when the wrongdoer had made amends and lowered it when the wrongdoer had not. In the marital study, the pattern depended on the partner's agreeableness, used as a sign that the partner would keep the forgiver safe and valued. Parts of this work have drawn criticism of its measures, and I rely here on the published abstract.

Behavior is the hardest of the three. James McNulty followed newlywed couples for four years. Psychological and physical aggression stayed level for more forgiving spouses, while it declined for less forgiving ones. This is an association, not proof of cause, and the measure used hypothetical questions, which critics have challenged. Responsibility for aggression lies with the aggressor. Nothing in this finding means that a person who forgives has caused what happens to them. Even with causation unsettled, this finding gives no reason to expect forgiving to stop aggression, so where aggression exists, check safety first.

07Whether the forgiven person changes could not be confirmed, so handle responsibility separately

If amends shape the forgiver's self-respect, the next question is what changes in the person forgiven. McNulty's association with aggression exists, but it does not establish cause. Beyond that, the studies I read do not answer it. I mean only those studies; I have not surveyed the whole literature.

I also could not confirm whether "forgiving" an AI error is a meaningful act, and I leave that as an open question. When a generative AI draft contains an error, recording it and fixing it through procedure can go ahead whether or not anyone forgives anyone. A reviewer can correct the record on one day and still be angry the next, and neither fact changes the other.

For now my answer is to treat settling feelings and pursuing responsibility as two jobs. Since it is unknown whether the forgiven person changes, handling that person's responsibility cannot rely on the effects of forgiving. Forgiving also leaves the wrongdoer culpable, so separating the two does not conflict with its definition. Forgiving then erases no one's responsibility, and pursuing responsibility does not block a person from letting resentment go. This is my inference from the studies above, not a trial result, and I found no study testing it in a workplace. I do not settle here whether choosing for oneself should be called forgiveness.

A trial shows that forgiving can help the forgiver up to two weeks, though other results show self-respect falling where no amends were made; the studies reviewed here cannot say whether it helps the person forgiven. So decide whether to forgive for your own sake, and leave that person's responsibility to a separate process.