From Ethical Principles to Lifecycle Oversight: A Library-Based Analysis of AI-Enabled Health Research in Canada

In the modern world of health research, scientists often rely on computer programs that learn from vast amounts of information to find patterns, predict illnesses, or suggest treatments. These programs, known as artificial intelligence, are powerful tools, but they operate differently than the traditional experiments researchers have used for decades. In a standard study, a scientist might ask a group of people for permission to use their medical records for a specific question, run the analysis, and publish the results. Once the study is done, the work is usually considered finished. However, artificial intelligence changes this timeline. These systems can be trained on data, then updated, improved, and redeployed for entirely new tasks long after the initial research is approved. They can also create new information from old records, or be used by companies outside the research team, raising questions about who is responsible if something goes wrong. Canada has a well-respected set of rules called the Tri-Council Policy Statement that guides how researchers treat human participants. These rules are built on three solid ideas: respecting people, caring for their well-being, and ensuring fairness. While these principles are strong, they were written before artificial intelligence became a common part of health research. The big question facing Canada's research community is whether these old rules are enough to handle the new, shifting nature of these smart computer systems.

A researcher named Abdolreza Babamahmoodi set out to answer this question by looking closely at how Canada's current rules apply to artificial intelligence in health. This was not a study that tested new drugs or surveyed patients; instead, it was a careful examination of existing policies, laws, and guidelines from Canada and around the world. The researcher gathered twenty-eight different documents, including Canada's main research ethics policy, international guidance from the World Health Organization, and rules from Australia and the European Union. The goal was to see if the current Canadian rules provide clear instructions for researchers and ethics committees on what to do when artificial intelligence is involved. The analysis focused on six key areas: deciding what counts as research, where the data comes from, whether the computer model works fairly, how humans keep control, how to handle changes to the system later, and how to respect the rights of Indigenous communities.

The study found that while Canada's existing rules are good at setting the moral tone, they are not detailed enough to guide the daily decisions researchers and ethics boards must make. The current policy says researchers must respect people and ensure fairness, but it does not specify what evidence a researcher needs to show to prove a computer model is fair. It does not say what happens when a model is updated six months after approval, or who is responsible if a third-party company uses the data in a way the researcher did not expect. The researcher noted that without these specific instructions, ethics committees are left guessing. Some might ask for too much information, slowing down important work, while others might miss critical risks because they do not know what to look for. The paper argues that the principles are sufficient to judge right from wrong, but they are not operationally adequate to tell people exactly what to do.

To solve this, the researcher proposed a new way of looking at the process, called a "lifecycle review." Instead of treating research as a single event that ends when the study is published, this approach views it as a continuous journey with six distinct checkpoints. The first checkpoint is the purpose: defining exactly what the artificial intelligence will do and who it will help. The second is the data: mapping where the information came from and who has the right to use it. The third is the development: checking how the computer model was built and tested. The fourth is validation: ensuring the model works well for different groups of people and not just the average. The fifth is deployment: making sure a human can understand the computer's suggestions and stop it if it makes a mistake. The final checkpoint is change and retirement: planning for what happens when the model is updated, moved to a new place, or turned off.

This new framework suggests that researchers should submit a specific set of information at each of these six stages. For example, when a model is first proposed, the researcher must explain its intended use. Later, if the model is updated or if the data sources change, the researcher must report this to the ethics committee to see if the change is significant enough to require a new review. The paper emphasizes that this is not about creating a mountain of paperwork, but about ensuring that the right people are making the right decisions at the right time. It suggests that a low-risk study might need only a few details, while a high-risk study that affects patient care would need a much more thorough plan.

A crucial part of the paper addresses the rights of Indigenous peoples in Canada. The current rules already recognize that Indigenous communities have collective rights over their data, not just individual rights. The researcher found that while the existing policy acknowledges this, it does not explain how these rights apply to artificial intelligence. For instance, if a computer model is trained on data from an Indigenous community, does the community have a say if that model is later sold to a company or used for a different purpose? The paper argues that these questions cannot be answered by a general policy written by outsiders. Instead, the researcher explicitly states that these specific governance questions must be led by the Indigenous communities themselves. The proposed framework leaves a space for these communities to define their own rules for how their data and the resulting computer models are treated, ensuring that their authority is respected without being overridden by a one-size-fits-all solution.

The researcher also looked at how other countries handle these issues. The World Health Organization and Australia have started to create specific guides for artificial intelligence, and the European Union has laws that require strict monitoring of high-risk systems. The paper suggests that Canada does not need to copy these laws exactly but can learn from their structure. By adopting a similar "living" guide that can be updated as technology changes, Canada can keep its core ethical principles stable while adding the necessary details to handle new challenges. The study concludes that the current system is not broken, but it is incomplete. It needs a bridge between its strong moral principles and the practical realities of how artificial intelligence works.

The proposed solution is a working model that can be tested and improved. The researcher is not claiming that this framework is perfect or that it solves every problem immediately. Instead, it is presented as a tool for discussion and future validation. The paper suggests that before this model becomes a standard rule, it should be tested with real researchers and ethics committees to see if it actually helps them make better decisions without creating unnecessary delays. The ultimate goal is to create a system where artificial intelligence can be used to improve health, but where the people involved are protected, the data is used responsibly, and the communities affected have a real voice in how the technology is governed. By breaking the process down into clear, manageable steps, the framework aims to make the complex world of artificial intelligence research more transparent and accountable for everyone involved.