What does a human-AI team mean — sorting 53 papers found five distinct kinds under one name
Research on human-AI teams is growing, but what counts as a team differs between studies. This paper sorts 53 papers into five kinds and shows that distinct arrangements are being studied under one definition, raising whether findings transfer between them.
Paper overview (our summary)
- Field (arXiv category)cs.HC（+1）
- AuthorsNathan Hughes, Ibrahim Habli
- Submitted2026-07-02
- arXiv ID2607.02198v1
Key points
- Fifty-three papers on human-AI teams were sorted into five main clusters using psychological taxonomies of teaming.
- The five are an AI assistant arrangement, ad-hoc dependency, ad-hoc forced dependency, paired equanimity and group equanimity.
- Distinct kinds of team are shown to be studied under the same definition.
- That raises whether insights are truly transferable between papers.
- The paper offers a way to identify the kind, a checklist for reporting, and routes toward synthesising the field.
1One name, different contents
The phrase human-AI team is widely used, yet what it points to differs by study. An arrangement where AI assists a person in charge and one where judgment is genuinely shared are not the same situation.
| The kinds identified | Character |
|---|---|
| An AI assistant arrangement | The person leads and the AI supports |
| Ad-hoc dependency | Dependency arises as needed |
| Ad-hoc forced dependency | Dependency is structurally unavoidable |
| Paired equanimity | A person and an AI work as equals |
| Group equanimity | Several parties work as equals |
The authors sorted 53 papers into five clusters using psychological taxonomies of teaming. Each represents a distinct combination of team-level characteristics, showing that different things are handled under one definition.
2Whether findings transfer
The question the authors raise from this is the core of the paper. May a finding from one study be carried into another? Sharing a term does not make the premises the same. A field can look as though it is accumulating while in fact running several separate enquiries in parallel.
3What to do about it
- 1IdentifyDetermine which kind of team you are dealing with
- 2ReportDescribe it following the checklist for reporting
- 3CompareSet findings against studies of the same kind
- 4SynthesiseConsider how the field can be brought together
The paper does not stop at diagnosis: it offers a way to identify the kind and a checklist for reporting. The contribution takes the form of making concrete what must be written down for a study to be comparable with others.
4When a term spreads before it settles
In a new field, words start being used before they are defined. This site separately covers work warning against reporting the effect of human-AI collaboration as a single average. Both point the same way: check what is being aggregated before aggregating. This article is our own summary of public research information and does not warrant its contents.
Why it matters
Findings gathered under one term may describe quite different situations. Anyone consulting research to inform adoption needs to establish which kind their own situation belongs to.
FAQ
Why does the kind of team matter?
What does the paper propose?
Sources (primary)
Source: arXiv (descriptive metadata is CC0 public domain). Summaries are our own; see arXiv for the original text and PDF.
- arXiv abstract page (original, official)
- PDF (arXiv)
- arXiv ID: 2607.02198