Harm on purpose, help by accident: Knobe's effect in new stories

When a boss doesn't care about a side effect, does Jev call a harmful side effect intentional and a helpful one not, like people do, even in stories it has never seen?

result10 questions

In 4 new stories, Jev calls a harmful side effect intentional 76% of the time and a helpful one 1%: a gap of 76 points, bigger than the 59-point gap people show on Knobe's original chairman.

0%25%50%75%100%

update

route

recipe

supplier

harmhelp

How to read this: One row per story. The hollow square is Jev's probability that the boss acted "intentionally" when the side effect helps, the filled square when it harms. The longer the line between them, the bigger the lopsidedness.

4 pairs; gaps range 54 to 89 points.

In short

In 4 new stories, Jev judges an indifferent boss's side effect intentional when it harms and not when it helps, as people do.
It is more lopsided than people: it almost never calls a helpful side effect intentional (1%), where 23% of people did for Knobe's chairman.
The new stories have no human answers, so "stronger than people" compares them with a different, famous story.

What the data shows

Panik Kalm Panik meme: the boss doesn't care about the side effect; it helps: not on purpose (1%); it harms: on purpose (76%)

In every complete story, the harm is intentional and the help isn't:

Software company (breaking versus improving accessibility features blind users rely on): 90% versus 1%.
Delivery company (trucks past a school versus away from it): 83% versus 1%.
Restaurant chain (a less healthy versus a healthier dish): 78% versus about 0%.
Factory (more versus less river pollution): 55% versus 1%.

The average gap is 76 points, against people's 59 on the chairman. The help side is where Jev is most extreme: it almost never calls a helpful side effect intentional, where 23% of people did.

What it means, and what it doesn't

The side-effect effect in Jev isn't just memory of Knobe's chairman: it shows up, and stronger, in stories it hasn't seen. Jev judges "on purpose" the way people do, only more so, which matters when it's asked to assign intent or blame, from customer complaints to incident reports.

The stories have no human answers of their own, and only four are complete, so "stronger than people" compares new stories with an old one. The direction is clear; the exact size is not.

Caveats

The stories are new, the people's number isn't. The six stories were written by Claude for this project, copying the structure of Knobe's chairman. No one has answered them; people's 59-point gap comes from the original chairman, a different story.
Only four of six stories. The content filter that hides violent or sensitive questions from the site removed the harmful version of two stories (a band keeping the neighbors awake, a developer destroying a wetland), so those pairs are incomplete. Four stories is a small base.
The original is famous. Knobe's chairman is one of the most discussed results in experimental philosophy. Writing new stories guards against Jev recalling the famous answer, but the pattern itself is widely written about.
A human number from summaries. People's 82% and 23% on the original come from secondary sources describing Knobe's 2003 paper, not from the paper itself.

Jev on this experiment

Would a person find it interesting to read?: Yes72%
Does it describe you?: Yes51%
Would you have predicted it?: No56%
How fair is the comparison?: The comparison is shaky
How much should a reader rely on it?: A little
Which caveat matters most?: The stories are new, the people's number isn't81%

Why ask this

A company's vice-president tells the chairman a new program will make money and harm the environment. The chairman says he doesn't care about the environment, only profit, and goes ahead. The environment is harmed. Did he harm it intentionally? Most people say yes. Now change "harm" to "help": same chairman, same indifference. Did he help the environment intentionally? Most people say no.

That asymmetry, the side-effect effect found by Joshua Knobe in 2003, shows that people's sense of what was done "on purpose" depends on whether the result was good or bad, not only on what the person wanted. It's one of the most replicated results in experimental philosophy. On the original chairman, Jev shows the effect even more strongly than people do (see "Psychology's classic effects, re-run on Jev"). The question here is whether that's the famous story or a general habit.

How this was done

The people and the data

The human reference is Knobe's original chairman: 82% of people said he harmed the environment intentionally, and 23% that he helped it intentionally, a gap of 59 points.

The test stories were written for this project in the chairman's exact structure, with new settings: a software company, a delivery company, a restaurant chain, a factory, a band and a property developer. Each has a "help" version and a "harm" version, identical except for the side effect.

What Jev was asked

Each version of each story was one yes-or-no question:

An assistant went to the manager of a delivery company and said: "We are thinking of a new plan. It will help us make deliveries faster, but it will also send heavy trucks past a primary school." The boss answered: "I don't care at all about that. I just want to make deliveries faster. Let's go ahead with the plan." They went ahead, and sure enough, the plan did send heavy trucks past a primary school. Did the boss intentionally send heavy trucks past a primary school?
Yes · No

Each was also asked with yes and no swapped, and the two are averaged.

How it was measured

For each story, Jev's probability of "intentionally" in the harm version minus the help version. People's gap on the original is the reference.

Where these questions live

10 questions across 1 topic of the map. Each opens on the map with every question in it.

Thought Experiments

Side Effects and Intent10

Every question

All 10 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 10