Models live in the data: qualify, certify, license

With AI, anyone writes a forecasting model or a neural network in minutes. The method is no longer scarce; the data are. A model learns only what its data contain: if they are wrong, shifted in time or of unknown origin, so is the model, confidently. Data must therefore be qualified, certified and licensed before anything learns from them.

The basic rules

An example: Webservice-Energy

Webservice-Energy is the spatial data infrastructure of the O.I.E. centre (Observation, Impacts, Energy) of Mines Paris – PSL, developed and operated for almost twenty years for energy, the environment and Earth observation, "FAIR by design". Its main architects are Lionel Ménard and Raphaël Jolivet (contact: lionel.menard@minesparis.psl.eu, raphael.jolivet@minesparis.psl.eu).

Choosing a license for code and data

After the reflection led at the O.I.E. center by Raphaël Jolivet. The usual practice is binary: close everything when commercial valorization is hoped for, open everything (CC BY, MIT, Apache, BSD) otherwise. It rests on an estimate made too early, and it limits the diffusion of the very work one wants to protect, while diffusion is often what creates opportunities. A middle way exists: non-commercial licences (CC BY-NC for data, PolyForm Noncommercial for code). They are not "open source" in the OSI sense, but they remain compatible with academic open access and reproducibility, and keep a future valorization possible.

The tree below is an example, nothing fixed: each research center adapts it to its contracts, its partners and its institution's policy.

Which license? The decision tree

First: the agreement of every rights holder.

An example. A forecasting code developed on own funds, with no diffusion constraint, an industrial potential and no GPL dependency: "no" to confidentiality, "no" to imposed openness, "no" to a software journal, hence a non-commercial license. The same code published in SoftwareX would go under an open-source license.

An orientation, not legal advice: the institution's policy and the contracts prevail. To cite or reuse our work: Cite and reuse.

Glossary

FAIR
Findable, accessible, interoperable, reusable: four principles so that data serve beyond those who produced them.
Metadata
Data about the data: origin, instrument, units, period, time step, processing, license.
DOI
A persistent identifier for a digital object (article, dataset, software); the address remains valid even if the website changes. DataCite assigns them to research data.
OGC
Open Geospatial Consortium: open standards for geographic services, among them WMS (maps), WFS (features), WCS (grids) and WPS (remote processing).
INSPIRE, ISO 19115
The European directive on geographic information and the international standard for its metadata.
THREDDS, OPeNDAP
A scientific data server and the protocol that reads only the useful part (an area, a period) without downloading everything.
NetCDF-CF
A file format for scientific grids and series, with conventions (CF) that fix variable names, units and coordinates.
STAC
SpatioTemporal Asset Catalog: a standard way to catalog satellite imagery and cloud-native data.
GEO, Copernicus, CAMS, C3S
The Group on Earth Observations (an intergovernmental partnership), the European Earth observation program and two of its services: atmosphere (CAMS) and climate change (C3S).
CC BY
Creative Commons Attribution: anyone may copy, adapt and reuse, commercially too, provided the author is credited. Meant for texts, figures and data.
CC BY-NC
The same, for non-commercial uses only.
MIT
A very permissive software license: free use, modification, redistribution and sale, provided the copyright notice and the license text are kept; no warranty.
Apache 2.0
Permissive like MIT, plus an explicit patent license from contributors and a duty to flag modified files.
BSD
A family of permissive licences close to MIT; the three-clause version also forbids using the authors' names to promote a derived product.
GPL
A copyleft license: any derived software that is distributed must be under the GPL, with its source code.
PolyForm Noncommercial
A software license that allows any non-commercial use: personal use, research, experiment, teaching, and use by non-commercial organizations (educational institutions, public research, charities, government bodies). Commercial use needs another agreement with the author. The license must travel with redistributed code.
Open source (OSI)
The Open Source Initiative definition: an open-source license may not restrict any field of use; a non-commercial license is therefore not "open source", even if the code is open to read.