-
The first step to alleviating researcher and REB concerns and
anxiety is to provide some level of advice and guidance right now.
Develop and release a draft document outlining preliminary guidance
for AI research. Make this document truly "living" by stating that
it will be updated as new information is gathered and new advice is
proposed. Seek feedback about this draft document from experts and
REBs across the country and update the document as that feedback is
reviewed.
-
Use the powers within your control to seek feedback from the best
and brightest. As is permitted by the Terms of Reference for the
Panel on Research Ethics (PRE), create an ad-hoc expert group to
explore, in more depth, the ethical implications of AI research and
what that means for the TCPS2. This expert group could be
responsible for updates to the draft preliminary guidance document,
with the goal of providing a more permanent document in the future.
-
Stop the silence. The lack of information from the PRE is a major
area of concern for researchers and REBs alike. Rather than keeping
silent until a new version of the TCPS2 can be released, or an
official interpretation document is released, keep members of the
research community updated on the progress being made on AI research
guidance. Use such regular communications to gather feedback and
gauge areas of concern.
-
While it is not yet law, Bill C-27 has generated a lot of concern
and discussion. Some of that discussion should be coming from the
PRE regarding how Bill C-27 might impact the TCPS2. Commitments do
not have to be made, but an analysis of how Bill C-27 might impact
the TCPS2 would be beneficial to the PRE, researchers, and REBs
alike.
-
At the moment, all REB and REO activity at institutions is funded by
that institution. While the Tri-Council may claim that the funding
for such areas comes from overhead obtained through grants, those
funds are not dedicated to ethics. Dedicated funds to REBs and REOs
to ensure they are properly staffed, trained, and able to do their
important work is crucial to ensuring research in Canada is
following the TCPS2.
-
Another pain point often heard from REBs is the lack of consistency
between the TCPS2 and other Tri-Council policies, especially where
data is involved. More care and attention should be paid to align
the TCPS2 with these policies to ensure researchers are not being
asked to do conflicting things. In addition, both the TCPS2 and
these policies need to be reviewed from an expert perspective.
-
The Tri-Council should provide specific guidance and oversight to
public researchers for data use in public/private research
partnerships. Public researchers engaging in research with industry
partners need guidance to ensure they understand their obligations,
and a clear enforcement mechanism.
-
The 2022 version of the TCPS2 does contain enough information to
draft some conclusions about the impact of ethics approval on AI
research. Rather than wait for guidance from the PRE, GREBs can
develop high-level guidance in the forms of checklists or
infographics for researchers to use when thinking about the ethical
implications of their AI research. Like the recommendations for the
Tri-Council, this guidance can be considered dynamic and be updated
as new information is obtained and considered.
-
One tangential point seen from the survey results is the fact that
there seems to be a disconnect between what university
administration is doing with regards to AI and the involvement of
GREBs. While policies associated with students, courses, and
teaching and learning may not be directly relevant to GREBs, being
involved in these discussions would be valuable in order to gather
information that may eventually be used as guidance for researchers.
-
GREBs should create a sub-committee of board members and experts
from their institution to make decisions about the ethical
implications of AI research. This sub-committee could be tasked with
creating and updating the guidance documents noted above, and
evaluating ethics applications involving AI. Having AI researchers
on such committees would ensure the proper expertise is received and
valuable perspectives are heard.
-
Research support personnel have a unique and wholistic view of the
research being conducted in a given institution. Training them on
research ethics helps to ensure consistent ethics messaging is being
heard by researchers across the institution. It also provides GREBs
with collaborators who can provide insights on the ways in which
researchers think about and plan their projects, which might help
with the development of future guidance.
-
AI models that include distributed representations of their training
data, where those data are subject to leakage, should be subject to
the same consent, security and privacy regulations as the original
data.
-
Since one of the main reasons why AI researchers avoid seeking
ethics approval is that they (correctly) perceive the process to be
time-consuming, persnickety, and opaque, part of the solution has to
be making the process more researcher friendly. Developing
streamlined processes for simpler cases, increased flexibility about
immaterial details, and making the process more collaborative could
all improve the experience for researchers without adding to the
workload of GREB staff.