The majority of Canadians are comfortable with artificial intelligence (AI) in the financial services industry if it means they will receive better services, according to a new report from TD Bank Group (TD).

The TD-commissioned survey found that 87% of Canadians said they expect their banks to be innovative, and 78% are open to financial services introducing new forms of technology — including apps and online services. Almost three-quarters (72%) were comfortable with AI, specifically, if it leads to better and more personalized service.

Respondents’ concerns lay in the risks AI poses to both themselves and society, the survey found. More than three out of four (77%) believe AI is advancing too quickly to understand the technology’s potential risks, and roughly two-thirds (68%) said they don’t understand the technology well enough to know the risks.

More than half of those surveyed (59%) said the lack of diversity among AI workers could lead to built-in biases.

When it comes to expectations of AI, 70% of Canadians said they should have control over how their data is used. More than half  (55%) also said companies should be transparent about how the technology is being implemented.

In addition to surveying Canadians about their attitudes toward AI, TD also held a roundtable discussion with experts from multiple sectors to explore the opportunities and challenges facing the financial services sector in adopting AI.

Richard Zemel, a computer science professor at the University of Toronto and research director at the Vector Institute for Artificial Intelligence, told the panel that balance is key in AI design.

“While removing data to protect against bias or to uphold privacy standards may be necessary in some cases, a lack of data can cause problems in the AI model,” Zemel said in the report. “Financial services need an approach that takes this challenge into consideration.”

Commissioned by TD, Environics Research Group conducted the online survey of 1,200 Canadian adults between May 17 and 25. Online surveys cannot be assigned a margin of error because they do not randomly sample the population.

Read the full report here.