Government science must take a comprehensive approach to integrating AI, say experts

Government departments and agencies must do more to integrate artificial intelligence in the research they support, but doing so demands a close look at ethical, security and data literacy issues, said panellists at the Canadian Science Policy Centre's annual conference, held online on November 18.

“Integrating AI into scientific research is progressing too rapidly to be playing catch-up at policy levels,” said Dr. Vik Pant (PhD), Chief Scientist and Chief Science Advisor at Natural Resources Canada.

Pant moderated the panel session “Bringing Digital Acceleration to Science: The Road to Adopting Artificial Intelligence in Science-Based Departments and Agencies.”

“It’s no longer viable to do science at scale without some kind of artificial intelligence [or machine learning],” said Dr. Carolyn Watters (PhD), Chief Digital Research Officer at the National Research Council.

Ethical issues are often incorporated as an afterthought, but they need to be built into and maintained in AI architectural structures, Watters said. Even when ethical considerations are addressed in early design, they can fall away as machines create algorithms dealing with software fixes and other changes, she noted. These subsequent changes may result in some “really nasty unintended consequences as you go forward."

Ashley Casovan, Executive Director of the non-profit organization Al Global, said that applications can be used incorrectly because the right outcomes haven't been properly inputted. She offered examples of AI facial recognition incorrectly identifying people, misdiagnosing health conditions or producing inaccurate predictions. Despite these issues, companies and governments as well as scientific researchers all need to integrate AI more quickly, said Casovan, former director of data and digital for the Government of Canada.

Dr. Gilles Bellefleur (PhD), Senior Research Scientist at Natural Resources Canada (NRCan), said AI enabled his organization to process data with ease and speed from across the country. NRCan incorporates huge datasets from physics, geology and rock mechanics.

But Bellefleur, like the other panellists, added the caveat of safeguarding information protection and intellectual property. Researchers must have access to expertise and specialization as they gather data from different sources, but as custodians, they have to ensure this data will be managed with privacy protection, he said.

Dr. Amir Ghasemi (PhD), Research Scientist at ISED's Communications Research Centre Canada (CRC), described how cloud computing and AI facilitated procedures to speed up certain labour-intensive processes, such as the granting of telecommunications licenses and ensuring compliance. Ghasemi said they were able to improve their decision-making by leveraging enormous amounts of field data.

Emphasizing the need to modernize data centres, he described offices prior to the arrival of cloud computing as “basically a lab” with multiple computers and servers under people’s desks.

With cloud computing, science is scalable and the community is sharing the information, Ghasemi said. “Scientists today are increasing production in difficult work because they’ve integrated a whole bunch of new technologies into their work.”

Ghasemi also underscored the need for data literacy, saying it is needed across all government agencies and departments. “From policy making to data collections [and] execution to data reporting, you need a certain degree of data literacy.”

R$