Recommendations AI methods used in a publication should be:
11. Presented in the context of a problem description that clearly identifies what
problem they are intended to solve
12. Outlined conceptually so that anyone can understand theirfoundationalconcepts
13 Described in pseudocode so that others can understand the details of how they work
Table 4. Author Checklist Part III.
Recommendations for AI methods in publications.
always offer DOIs, licenses, and citations.
For a specific publication, the version of the source
code that is being used should be clearly specified,
and the source code repository should support the
identification and future access of specific versions.
Source Code Metadata
Basic metadata includes a descriptive title, the source
code’s authors, and the creation date. Additional
metadata is always valuable to others in terms of
understanding and reusing the source code.
Licenses for Source Code
Recommended licenses for source code are the standard licenses from the Open Source Initiative. Licenses such as Apache v2 or MIT are recommended
because they provide unlimited reuse (as long as
there is attribution). Other more restrictive licenses
are available to limit commercial use or impose
licensing conditions on extensions of the original
Permanent Unique Identifiers for Source Code
A separate DOI should be assigned to meaningful versions of the source code, such as a version used for a
publication. GitHub offers an option to obtain a DOI
for a source code version, which is done by storing
that version permanently in the Zenodo data repository. Any source code can be uploaded manually to
community data repositories such as Zenodo,
figshare, and Dataverse. PURLS can be assigned by
anyone to any source code version that has a URL on
the web, using a trusted service such as w3id.org.
Source Code Citation
Source code citation can be directly provided by a
source code repository, or it can be constructed by
hand. A citation for a source code version consists of
a descriptive name (or title) for the source code, its
creators, the name of the repository where it can be
accessed, the version, and the permanent URL. For
example, a citation for GitHub code in (Gil et al.
Ratnakar, Varun. “DISK software” (v1.0.0). Zenodo.
By uploading the source code into the GitHub code
repository, we obtained a persistent identifier for the
version used in the publication as well as the citation.
Specifying the authors, the name, and the license
takes negligible effort. Implementing the author
checklist for source code required little time.
Recommendations for AI Methods
Our recommendations for AI methods are listed in
The problem that a conceptual AI method solves
should be explicitly described in the publication. In
De Weerdt et al. (2013) the following example can be
found: “To address this problem, we propose a novel
navigation system ...” The authors explicitly describe
the problem that they address. Another good example of this practice can be found in He et al. (2016).
Here the authors state the problem explicitly: “In this
paper, we address the degradation problem by introducing a deep residual learning framework.” The
degradation problem is also properly described in
A high-level, textual description of the AI method
should be provided to readers to allow them to gain
an understanding of it. This description should
include a broad overview of how the AI method
works and specify input variables and the resulting
output. In general, the AI research community excels
at providing this information in publications.
Pseudocode for the AI method should also be provided. In cases where detailed pseudocode cannot be
provided due to the complexity of the proposed algorithm or system, a more abstract pseudocode summary can be provided that illustrates the AI method’s