Can AI tools and cloud computing coexist with the principles of open, reproducible science? That was the question driving a UBC think tank on 7 May. While the afternoon didn’t produce easy answers, by bringing together researchers, bioinformaticians, and data management professionals in the same room, it generated insights that none of these groups would have reached alone.
We welcomed a diverse group of not only researchers, but also data management support professionals and privacy experts. One of our key takeaways was the added value of these different perspectives in a dialogue about infrastructure – Ronnie de Jonge
The timing had a certain irony to it: on the same day, a fire at an Almere data center disrupted systems at Utrecht University, illustrating rather concretely the vulnerabilities the group had gathered to discuss.
After an opening by Ronnie de Jonge, three invited speakers framed the conversation: Wilson Silva working on trustworthy AI for life sciences, Mercedes Beltràn on responsible data management, and Patrick Kemmeren sharing practical experience from the Princess Maxima Center demonstrating what well-designed cloud infrastructure can achieve, including reducing analysis time from six months to two weeks.
From there, the conversation moved into facilitated group discussions that ran over time and continued during drinks. Three themes stood out.
A shared concern was how to build resilient workflows that don’t leave research dangerously dependent on a handful of commercial providers. The risks of such dependencies include a provider changing their model in ways that affect results, adjusting their pricing in ways that make the service unaffordable, or going down at a critical moment. Switching providers is possible in theory, but costly and disruptive in practice. Participants called for contingency plans and regular backups as a baseline to ensure there is always an alternative when a platform fails or becomes unavailable.
At the same time, realistic risk management was raised: allowing every hypothetical risk to block progress is counterproductive. Not adopting technologies while others do, will pose a risk in itself, namely that of falling behind. The goal is a pragmatic approach that enables work today while building resilience for tomorrow. European-level coordination was seen as a necessary step towards this goal.
Good legal agreements, and steps like anonymising data before it enters the cloud at all, were seen as practical foundations. Two fundamental trade-offs proved harder to resolve. The first is between model quality and privacy: better AI systems benefit from access to more data, yet privacy considerations limit what can be used. The second is between patient privacy and institutional interests: in collaborative research settings, protecting individuals and protecting organisational interests can pull in opposite directions. Building trust between partners was seen as important in overcoming these challenges and keeping collaboration moving.
Beyond these tensions, broader questions of governance and autonomy were raised: who ultimately controls and processes data, in whose interest, and how consent for data sharing should be organised in a landscape that is constantly evolving. These illustrate the need for clear and adaptive guidelines.
There was genuine concern about the effect of AI on critical thinking and creativity. The response, participants agreed, is not to reject tools that are already in use, but to adopt them deliberately: keeping humans in control, defining the scope of AI use clearly, and investing in training so researchers can engage with these tools critically. Positive examples of responsible use, such as the Sturgeon tool for patient care, matter here, as do explicit guidelines on when not to use AI.
Looking back
Although no list of solutions emerged, that was never really the aim. One of the key takeaways was the added value of bringing different perspectives into a dialogue about infrastructure: the kinds of insights that emerge when researchers, bioinformaticians, and data management professionals are in the same room are ones that none of these groups would reach alone. We look back on a productive afternoon of dialogue and exchange, and hope participants left with new perspectives and connections. A shared recognition of the challenges, and a stronger foundation for the ongoing conversation, is a meaningful outcome in itself.