The Justice Department has quietly launched an artificial intelligence experiment as the Trudeau government prepares to use such sophisticated software to help make decisions in cases involving immigration, pension benefits and taxes.

The 18-month pilot project, which involves the Canada Revenue Agency, was started even though the government has yet to establish clear ethical guidelines on its use of artificial intelligence, or AI.

Definitions of AI vary, but it's generally understood as a computer system designed to quickly learn, reason and make decisions by imitating human cognition.

Since January, 26 tax practitioners at Justice Canada have been using commercial AI software that analyzes thousands of court cases to predict how judges might rule on a given set of facts about a taxpayer's affairs.

"There is currently no legislation, regulations, policy position or framework within the Government of Canada to govern the use of AI in Canada," says a May 2018 briefing note, obtained by CBC News under the Access to Information Act.

"There is a need to engage stakeholders to look at this uncharted territory with a critical reflection on challenges and new legal, ethical and policy issues."

Policy vacuum

Even in this policy vacuum, the Justice Department wants to be "on the leading edge and part of the critical discussions surrounding how to leverage new technology as a department and as a government," says the briefing note.

The pilot project uses the software program Tax Foresight, developed last year by Toronto-based tech startup Blue J Legal Inc.

The company, founded by legal experts at the University of Toronto, claims that the program predicts tax-litigation outcomes with an accuracy rate of 90 per cent.

The Department of Justice, which has paid Blue J Legal more than $20,000 for licences, went ahead with the pilot after its AI Task Force recommended an off-the-shelf product for the exercise, said spokesman Ian McLeod.

The department did not provide details about how Tax Foresight will be used, except to say that it will not supplant human employees. "This technology is meant to assist government employees in their work, not replace them," said McLeod.

"I think there's a lot of hype around AI"
- Fenwick McKelvey, assistant professor of communications, Concordia University, Montreal

Justice Canada also is looking at whether to launch two other pilots. One pilot under consideration would be launched at Immigration, Refugee and Citizenship Canada (IRCC) to help develop "litigation strategies," says a notice to industry sent earlier this year asking for information.

The other proposed pilot, with Employment and Social Development Canada (ESDC), would be used in litigation and research involving compensation for federal government employees. Eventually, AI could be expanded to work on litigation involving the Canada Pension Plan, Old Age Security and Employment Insurance cases, says the notice.

Justice Canada is still reviewing industry feedback for the IRCC and ESDC proposals, and has not made any decisions, McLeod said.

Through the Treasury Board, the Trudeau government has been ramping up efforts to use more artificial intelligence programs across departments. Those efforts are being quarterbacked by Alex Benay, who was appointed Canada's chief information officer in April 2017.

Treasury Board has been circulating a white paper suggesting seven principles for using artificial intelligence, including a proposal that AI software should be "trained" in ethics and human-rights obligations.

The document also says AI "should be deployed in the most transparent manner possible," and that "people should always be governed — and perceived to be governed — by people."

Critics and skeptics of AI have warned that the technology can be problematic, especially when it's used in policing and justice-related applications. Underlying crime data used by the software, for example, may be racially biased — which could drive litigation decisions that reflect racist attitudes.

"I think there's a lot of hype around AI," Fenwick McKelvey, an assistant professor of communications at Concordia University in Montreal, said in an interview. "It raises a lot of concerns where the rubber hits the road.

"If it's for tax cases, I think there's room for manoeuvre. But I think that it's important to be sensitive to some of the downsides of AI in justice and law enforcement, given the risks."

AI and human rights

McKelvey said government must ensure any use of AI does not violate the Charter of Rights and Freedoms or the Privacy Act, and operation of the software must be subject to external audit.

Fabio Bonanno, a tax expert with Trowbridge Professional Corp. in Toronto, began using the Tax Foresight software this year as a time-saving tool to research large numbers of court cases.

He said the firm's advice to clients remains in the hands of human professionals, not in the software, and he welcomed the news that Justice Canada is using the same Blue J Legal software.

"Hopefully, it will reduce any sort of disputes," Bonanno said in an interview. "On their end, there are employees in the Department of Justice that are going to be doing the same thing as me."

This article originally appeared in CBC