GCRI · Labs

L'IA dans les flux de travail scientifiques

Machine learning now enters science at every step: screening literature, designing experiments, analysing images and sequences, emulating simulations and drafting text. Each use can speed discovery and each can introduce errors that look like findings. This page sets out how models are evaluated for scientific use, where leakage, benchmark overfitting and silent drift arise, and what evidence editors, funders and regulators need before they rely on a model-assisted result.

01Canevas

Comment le travail se déroule.

systèmes · autour de ce sujetIllustratif

Testbed operations

AI in scientific workflows · Field lab coast · Campaign 5

Sensors deployedGCRI · Labs

Sensor deployment

littoral illustratif

BuoyStationGaugeMerBands: tidal rangeN

Sensors

S-01 tide940.0Nominal
S-04 salinity760.4Recalibrate
S-07 wave950.1Nominal

Sensor vs reference

Testbed operations: the screen’s structure, terms and states for AI in scientific workflows.Noms donnés à titre d'exemple et valeurs illustratives ; une conception, non un enregistrement ni un système en service.

Fig.Testbed operations: example names and illustrative values, not a record or a live system.

02La question

AI in scientific workflows: how the risk develops.

Model evaluation and scientific assistance. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.

Mécanisme

Leakage inflates performance

When information from the test set reaches training, through duplicates, shared patients, neighbouring sites or preprocessing fitted on all the data, a model appears more accurate than it is. Leakage has been documented in many fields that adopted machine learning.

Mécanisme

Benchmarks get overfitted

Repeated tuning against the same public benchmark selects models that fit its quirks. Gains on a leaderboard can vanish on new data, and contamination of training corpora with benchmark items makes scores hard to interpret.

Mécanisme

Drift and distribution shift

Instruments, protocols, populations and software change. A model validated on last year’s data or another lab’s scanner can degrade without any visible error. Monitoring after deployment is part of the evaluation, not an extra.

Mécanisme

Generated text and fabricated sources

Language models can produce fluent text, citations and code that are wrong. Unchecked, these errors enter manuscripts, peer reviews and proposals, and spread through citation. Disclosure and verification policies exist because such errors are hard to see.

03Méthodes et preuves

How AI in scientific workflows is modelled, measured and tested.

Le modèle, ses hypothèses, ses données et son incertitude, sous la forme qu'un scientifique ou un ingénieur vérifierait. Les structures suivent des méthodes publiées. Les valeurs représentées sont illustratives, ce ne sont pas des résultats.

Modèle, équations et hypothèses

The model structure used for AI in scientific workflows, the assumptions it makes and how each is tested.

Risque de l'élément iRi = f(Hi, Ei, Vi, Ci)aléa, exposition, vulnérabilité et capacité à faire face
Perte annuelle attendueEAL = ∑k pk · Lksur les événements k de probabilité annuelle pk
Systèmes couplésxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error
HypothèsePourquoi il est réaliséComment il est testé
Aléa et exposition sont indépendants au sein d'un événementGarde l'intégrale des pertes calculableComparer avec un ajustement de copule sur des relevés conjoints
Les couplages sont linéaires près du point de fonctionnementLaisse les chocs se propager à travers APerturber au-delà de la plage linéaire et comparer
Les courbes de vulnérabilité se transposent d'un lieu à l'autreLes courbes locales sont rarement disponiblesValider par rapport aux pertes observées lorsqu'elles existent
Entrées
jeux de données versionnés avec provenance
Productions
des distributions, jamais des nombres isolés
Code
publié avec la fiche de modèle et les tests
Reproduction
mêmes données, code, environnement et graine
Formulations standard du risque et de l'espace d'états ; les hypothèses énumérées sont typiques de la méthode, énoncées pour examen, et non des résultats.
Principes directeurs FAIR · GO FAIR et FORCE11W3C PROV · World Wide Web ConsortiumOrientations sur le traitement cohérent des incertitudes · GIECCEOS Analysis Ready Data · Comité sur les satellites d'observation de la Terre
04In the five domains

AI in scientific workflows changes each domain’s methods.

05Indicateurs

Ce qui est mesuré, et avec quelle qualité.

Chaque indicateur porte son unité, sa périodicité, le type de source qui le sous-tend et l'étiquette de preuve que porterait une valeur. Aucune valeur n'est affichée : un programme les fournit sous sa propre autorité.

IndicateurUnitéFréquenceType de sourceÉtiquette
Held-out evaluation design statedpercent of model-based studiesannuelAudit against the REFORMS checklistOBSERVÉ
Code and model weights availablepercent of studies with runnable code and weightsannuelRepository and article auditOBSERVÉ
External validationpercent of models tested on data from another site or periodannuelLiterature reviewESTIMÉ
Erreur d'étalonnageexpected calibration error, dimensionlessper modelHeld-out evaluationOBSERVÉ
Performance gap under shiftdifference in the primary metric between internal and external test setsper modelValidation reportsOBSERVÉ
AI use disclosurepercent of submissions with a disclosure statementannuelJournal submission recordsDÉCLARÉ
Contamination checks reportedpercent of benchmark evaluation reportsannuelEvaluation reportsDÉCLARÉ
06Sources

D'où proviennent les données probantes.

Jeux de données publics et registres institutionnels nommés, chacun avec sa catégorie de licence. Les sources fournies dans le cadre d'accords ne sont utilisées que selon leurs conditions et ne sont jamais republiées.

SourcesCe qu'elle apporteCatégorie de licence
OpenMLDatasets, tasks and evaluation runs with standardised splitsOpen, CC BY
Hugging Face Hub model and dataset cardsDocumented models and datasets with stated training data and limitationsLicence per model or dataset
MLCommons benchmarksShared benchmarks and measurement rules for model performanceOpen, Apache 2.0
AI Incident Database (Responsible AI Collaborative)Reported failures and harms of deployed AI systèmesOuvert, attribution
Software HeritageArchived source code with persistent intrinsic identifiersOpen archive, licence per repository
arXiv metadataPreprints across physics, computing and quantitative biologyMetadata CC0; licence per paper
Zenodo (CERN)Citable deposits of code, models, data and evaluation outputsOpen, licence per record
07Décisions

Qui décide, et à quoi servent les données probantes.

Laboratoires prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.

Qui décideLa questionDonnées probantes servies par cette page
Journal editors and reviewersWhether a model-based claim rests on an evaluation that excludes leakage and tests generalisationREFORMS or TRIPOD+AI checklist, split design, external validation, code availability
Bailleurs de fonds de la rechercheWhether applicants and reviewers may use generative AI, and on what disclosure termsFunder policy, disclosure records, documented verification of generated content
Medicines and device regulatorsWhether an AI component in a medical product is validated for its intended usePrespecified evaluation, external validation, predetermined change control plan
Research integrity officesWhether an allegation involving fabricated or generated content is substantiatedProvenance of text, data and code; version history; tool logs

Résultats que produit ce thème

  • Evaluation protocols that fix splits, metrics and baselines before training
  • Model cards and datasheets with intended use, limits and known failure modes
  • Leakage and contamination audits of model-assisted studies
  • Reproducible pipelines with code, weights, environment and data versions archived
08Questions essentielles

What experts ask about AI in research.

  • What task is the model doing, and what is it compared against?
  • Was the model evaluated on data independent of the data it was trained or tuned on?
  • How does performance change when conditions shift away from the training data?
  • Can the output be checked by a person or a simpler method, and is it?
  • Which steps of the workflow were assisted, and is that disclosed in the result?
09Mécanismes et facteurs

Where AI helps and where it misleads.

Machine learning models learn statistical regularities from data. They can be very effective where those regularities are stable and data are plentiful, for example classifying land cover from satellite imagery or emulating a climate model to explore many scenarios. They are weaker where data are sparse, where the system changes over time or where rare events matter most, which describes much of risk science.

Common failure modes are well understood: leakage, where information from the test set reaches training; glissement de distribution, where conditions in use differ from those in training; shortcut learning, where a model relies on a spurious cue; and overconfident outputs without calibrated uncertainty. Generative models add the risk of fluent but fabricated content, including citations and code that looks correct but is not.

Workflow effects matter as much as model accuracy. When a tool is used at several stages, small errors can accumulate, and reviewers may give automated outputs more trust than they deserve.

10Méthodes et données

How AI tools are evaluated.

  1. 01

    Held-out and out-of-distribution tests

    Evaluation on data separated by time, place or source. This is the minimum for any claim of generalisation.

  2. 02

    Baselines

    Comparison with simple statistical methods and expert judgement. A complex model that does not beat a simple baseline is not adding value.

  3. 03

    Calibration and uncertainty

    Checks that stated confidence matches observed accuracy, using quantification de l'incertitude.

  4. 04

    Provenance and disclosure

    A record of which model, version and prompts or settings were used at each step, so the workflow can be re-run or audited.

Benchmark scores rarely transfer directly to a new research setting, and a model’s behaviour can change between versions. Evaluation is repeated when either the task or the model changes.

12Where this page’s references

Where this page’s references lead.

Les organismes et documents à l'aune desquels cette page se lit. Les statuts et éditions changent ; chaque lien mène à l'organisme émetteur.

13Questions

Questions et réponses.

What is data leakage?
Any route by which information about the test data influences training or model selection. It makes reported performance higher than the model will achieve on genuinely new data.
Is a high benchmark score enough to publish a method?
Not on its own. Reviewers increasingly expect external validation, uncertainty on the score, ablations and an account of how the benchmark was protected from contamination.
Can a generative AI tool be listed as an author?
COPE and the ICMJE say no, because authorship carries accountability a tool cannot hold. Use of such tools is disclosed in the methods or acknowledgements, under the journal’s policy.
Why publish model weights and code?
So others can rerun the evaluation, test on new data and find errors. Where weights cannot be shared, for privacy or safety, a controlled access route and a stated reason are the accepted alternative.
Does Labs use generative AI to write findings?
Any assisted step is designed to be disclosed. Findings are the responsibility of named researchers, who check every generated claim, citation and line of code.
Can Labs evaluate a commercial AI model?
A model can be evaluated under a published protocol. The result describes performance on the tested task and is not an endorsement or approval.
Why compare with simple baselines?
Because complex models often appear impressive on benchmarks but add little over a well-chosen simple method on real research data.
RapportsContribuer

Apportez des éléments de preuve depuis cette page.

Rapports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.

GCRILaboratoiresPreuves des LabsMéthodes et technique
Laboratoires

Apportez une méthode, un jeu de données ou une question.

L'information et les demandes sont ouvertes dès aujourd'hui. Une demande reçoit une référence et indique l'étape suivante ; candidatures, paiements, allocations de calcul et inscriptions n'ouvrent que lorsque l'équipe, les conditions et la capacité d'examen existent.

The Global Centre for Risk and Innovation

Science · evidence · measurement

Not for profit

Our articles, maps and twins are free to read and free to republish. Readers and partners keep it that way.

Partner with us

Who funds us

GCRI is a not-for-profit institution. To ask who funds a piece of work, use the funding questions form.

Independence

Editors and the expert reviewers named on each article decide what we publish. Funders and partners do not review or approve articles before publication.

Licence

Articles: CC BY 4.0. Figures, maps, twins and data: CC BY 4.0. Other uses: ask through the permissions form.