Should an automated car explain more when a scene is harder to understand?
Across 1,103 annotated driving records, scenes judged to need more explanation were paired with longer human-written text. The association survived a 1,280-record rerun—but no occupants used the explanations in a moving vehicle.
The short version
- What happened
- The study matched 1,103 recorded-scene annotations with human-written explanations and an aggregate score for how necessary an explanation seemed. The texts contained 9,541 whitespace-delimited tokens, averaging 8.65 words and ranging from 2 to 20.
- Who it affects
- Passengers may benefit when a vehicle explains an unusual manoeuvre, but a message that is too long or badly timed could compete with the very attention needed to respond. Interface studies must measure that trade-off directly.
- What remains unproven
- The study analysed recorded scenes and written annotations rather than live vehicle use.
Human annotators wrote more when the scene seemed to demand it
A peer-reviewed study examined whether the amount of explanation in an automated-driving interface should vary with the scene. Researchers analysed 1,103 records pairing a recorded driving situation, an aggregate score for how necessary an explanation appeared and a human-written explanation. Those texts contained 9,541 whitespace-delimited tokens, averaging 8.65 per record.[1]
Who it affects
Passengers may benefit when a vehicle explains an unusual manoeuvre, but a message that is too long or badly timed could compete with the very attention needed to respond. Interface studies must measure that trade-off directly.[1]
Keep in mind
The study analysed recorded scenes and written annotations rather than live vehicle use.
The correlation between necessity and length was moderate and does not establish causation.
The four score partitions were descriptive; the analysis did not identify a unique latent-category count.
No comprehension, trust-calibration, takeover, reaction-time or safety outcome was measured.
The work received Korean government research support. The authors reported that funders had no role in design, analysis or writing and declared no competing interests.
Source analysis · 1 linked source. Original source date: 21 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 4 min The evidence, methods, implications and limitations
Human annotators wrote more when the scene seemed to demand it
A peer-reviewed study examined whether the amount of explanation in an automated-driving interface should vary with the scene. Researchers analysed 1,103 records pairing a recorded driving situation, an aggregate score for how necessary an explanation appeared and a human-written explanation. Those texts contained 9,541 whitespace-delimited tokens, averaging 8.65 per record.
Higher necessity scores were associated with longer explanations. The correlation was 0.374, with a template-cluster bootstrap interval from 0.266 to 0.448. Repeating the analysis with 1,280 records preserved the length, vocabulary and prediction patterns. The authors call the association 'progressive elaboration': people supplied more detail for scenes they judged harder to leave unexplained.[1]
Length carried a signal, but not the whole explanation
The researchers also tested whether wording could distinguish four descriptive score partitions. A text model reached balanced accuracy of 0.382, above a token-count baseline at 0.327 and a stratified dummy at 0.247. That is a modest exploratory result, not a reliable four-class explanation detector, and the analysis did not identify one defensible number of hidden explanation categories.
For designers, the useful idea is conditional detail: a routine lane change may need less explanation than braking for an obscured pedestrian. But content, timing and delivery may matter more than word count. Speech can distract; text can arrive too late; a technically complete account can still fail to answer the occupant's actual question.[1]
The distinction that matters
What the study tested—and what it did not
The evidence moves from recorded scenes to written annotations, not from a live vehicle to safer behaviour.
Observed
1,103 scene-and-text records
Necessity score linked with text length
Repeated
Rerun with 1,280 records
Main patterns persisted
Not tested
No live occupant experiment
Safety and understanding remain unknown
The missing evidence is a prospective occupant test
The records were retrospective annotations of recorded scenes. No passenger received the message while riding, and the study did not measure comprehension, trust calibration, takeover quality, reaction time or crashes. The necessity score and explanation text may also share the assumptions of the same annotation process.
A stronger next step would randomise short, adaptive and fixed-length explanations in a simulator, measuring whether people notice the cause of a manoeuvre without losing attention. Until then, the study provides a corpus-level design hypothesis—not evidence that a longer in-car explanation is safer.[1]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 2 min 31 sec · Background explainer
What is artificial intelligence? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A short introduction to artificial intelligence from the Royal Society. Start here for basic concepts, then use current reporting to explore what has changed.
Evidence check: Published in 2018, before today’s generative AI tools. Useful for basic concepts, rather than an overview of current capabilities.
Published 2018-11-28. Source checked 14 October 2026. Check captions on the publisher’s video page.
Explore current AI technologyPhoto: ThisisEngineering / Unsplash · Licence
Evidence check
The study analysed recorded scenes and written annotations rather than live vehicle use.
The correlation between necessity and length was moderate and does not establish causation.
The four score partitions were descriptive; the analysis did not identify a unique latent-category count.
No comprehension, trust-calibration, takeover, reaction-time or safety outcome was measured.
The work received Korean government research support. The authors reported that funders had no role in design, analysis or writing and declared no competing interests.
Reporting basis: Source analysis. Original source date: 21 October 2026.
Continue the story
Read next
Continue exploring explainable ai: compare the finding and its evidence limits.
Do AI explanations help people know when to trust a system?
A South Korean survey found that people who perceived an AI system as clearer also reported more understanding, trust and reliance. It did not test whether they actually understood the system or trusted it appropriately.
Read the next reportExplore the full source and evidence record
What this means for people
- Passengers may benefit when a vehicle explains an unusual manoeuvre, but a message that is too long or badly timed could compete with the very attention needed to respond. Interface studies must measure that trade-off directly.
Global context
Automated-driving interfaces will be used across languages, road cultures and vehicle-control regimes. A wording pattern from one annotated corpus must be tested with diverse occupants, accessibility needs and local traffic conditions before it becomes a design rule.
What the evidence does not yet show
- The study analysed recorded scenes and written annotations rather than live vehicle use.
- The correlation between necessity and length was moderate and does not establish causation.
- The four score partitions were descriptive; the analysis did not identify a unique latent-category count.
- No comprehension, trust-calibration, takeover, reaction-time or safety outcome was measured.
- The work received Korean government research support. The authors reported that funders had no role in design, analysis or writing and declared no competing interests.
What to watch next
- Simulator trials comparing adaptive, fixed and minimal explanations
- Effects on comprehension, takeover quality and attention
- Accessible delivery for hearing, vision and language differences
- Independent replication across road types, countries and vehicle systems
Living evidence record
Impact record IAI-0APC9HQ
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
21 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what Scientific Reports published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 21 October 2026
Research question, topics and regions
Whether perceived need for explanation in recorded driving scenes is associated with the length and wording of human-authored explanations
Topics: Autonomous vehicles · Explainable AI · Human factors · Vehicle interfaces · Trust
Regions: South Korea · Asia · Global
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.