Last semester I took a Data, Code and Social Innovation course that presented me with an opportunity to seek legal tech solutions for traditional legal issues rooted in common law legal tests. This course was an exciting intellectual learning journey. In addition, we were encouraged to try on a technology lens to spot pitfalls, limitations and biases that those solutions might create or have embedded in them.
As an enthusiast of labour and employment law, I came across Craig Mazerolle’s Crafting an Aboriginal Labour Law ((2016) 74 U Toronto Fac L Rev 18). I decided to look at challenges that Indigenous people face trying to organize and regulate their labour relations. In an attempt to combine legal tech and labour law, I attempted to design and test an Indigenous labour law classification model based on NIL/TU,O Child and Family Services Society v. BC Government and Service Employees’ Union  2 S.C.R. 696. Although it was an intellectually satisfying exercise, I faced a situation wherein I had been trying to develop a solution that Indigenous groups may neither need nor want. As settlers, we need to stop trying to imagine solutions on behalf of Indigenous groups. This article calls for Indigenous educators and employment and labour lawyers to share their perspectives, discuss labour and employment issues and ideas on how to achieve prosperity for their nations.
Aboriginal labour law
In the early 2000s, a band unsuccessfully tried to claim a legal right to enact its own Labour Code to regulate a casino (Mississaugas of Scugog Island First Nation v. AG Canada & AG Ontario 2007 ONCA 814, leave to appeal r’d, 2008). The Court of Appeal for Ontario agreed with the Ontario Labour Relations Board that there was no evidence to support Indigenous rights based on the Van der Peet test (R. v. Van der Peet  2 S.C.R. 507). It stated that “the organization of work at that level of generality is a feature of every human society” (para. 39). Finally, Justice Rosalie Silberman Abella noted in Tessier ltée v. Quebec (Commission de la santé et de la sécurité du travail) 2012 SCC 23 at para. 11 that “courts have accepted that legislation respecting labour relations is presumptively a provincial matter since it engages the provinces’ authority over property and civil rights under s. 92(13) of the Constitution Act, 1867.”
The Stevedores Reference (“derivative jurisdiction”) test is a narrow legal test, confirming that federal jurisdiction is an exception. In Quebec Minimum Wage Commission v. Construction Montcalm Inc.  1 S.C.R. 754 (as cited in NIL/TU,O Child and Family Services at para. 14), Justice Brian Dickson developed the “functional” test examining the nature of the business operation, its normal or habitual activities (“a going concern”) without regard for any exceptions. The NIL/TU,O Child and Family Services case addressed the jurisdiction over labour relations in providing Indigenous child welfare services by applying the “functional test.”
The functional test looks at something unique or distinctive (R. v. Sparrow  1 S.C.R. 1075; Van der Peet ) rather than personal and family needs, which everyone has. Under this test, there is a presumption that the jurisdiction is provincial. To displace this presumption, one must meet this test.The first step entails looking at the habitual activities, which constitute a federal undertaking (Indigenous governance issues). If one cannot conclude, then one needs to go to the second step of the test. The second step is to prove that “the provincial regulation of that entity’s labour relations would impair the ‘core’ of that head of power” (NIL/TU,O Child and Family Services at para. 20).
Providing services in a particular Indigenous community does not displace primary provincial jurisdiction over labour relations. The test confirmed that the Indigenous welfare services were a provincial operation despite their structure, location on the reserve and federal funding. Therefore, the court did examine the “core” of s. 91(24).
Looking at the new set of facts, the courts will presume the provincial jurisdiction over labour relations unless it is rebuttable based on habitual activities. If the first step is inconclusive, the courts will then examine whether the Indigenous operations lie at the “core” of the federal head of power and provincial jurisdiction impedes it.
The proposed binary classification model may predict two outcomes: a “federal” or “provincial” jurisdiction in the context of Indigenous labour relations. Based on how the logistic regression generates probabilities, these outcomes would either be “positive” or “negative.” To automate a judge’s decision, I would need to set a cutoff with the help of experts in Constitutional law, labour law, Indigenous law, corporate/commercial law and Indigenous representatives. The use of logistic regression guarantees that my output will be either “1” or “0.” The output of this model would be “Yes” – “1” for “federal” and “No” – “0” for “provincial.” Finally, if I had BigTech resources, I would create a neural network visualization with one hidden layer with activation functions to compute a prediction regarding the nature of labour relations.
Model testing, evaluation
To test this model, I would need to set the suitable decision threshold with a higher cutoff since genuine “federal” labour relations in the Indigenous context are few and far between. In this case, the classifier accuracy should be at 99 per cent for true “federal” positives. Before releasing any legal tech solution, I would perform an algorithmic impact assessment to assess potential harms to Indigenous people and ensure that I have developed an ethical AI tool. This too is available on a federal government website.
Access to justice solution
I envisioned this hypothetical classification model as an access to justice solution for Indigenous allied professionals such as lawyers, economists, consultants and proponents. It would enable the users to predict the possible outcome if it were to be heard by a judge. The model should provide legal information about the average predicted probabilities regarding the types of labour relations in the Indigenous context without lengthy and expensive litigation.
Pitfalls, limitations, biases
Indigenous labour law may become even more complex. A category of labour relations regulations concerning the Aboriginal public sector may arise based on traditional practices not accounted for by the common law classification model. Mazerolle’s example: Westbank First Nation Self-Government Agreement, 2003 (at 24). I would need to use a multiclass classification instead of a binary one.
If I focus only on the results confirming that Indigenous labour relations should be under federal jurisdiction, there is a risk of confirmation bias. With neural network visualization solution, I would face the “black boxes” issue (Alex Gutman and Jordan Goldmeier, Becoming a Data Head: How to Think, Speak, and Understand Data Science, Statistics, and Machine Learning (2021) at 177) (Data Head). It is not easy to decipher why those artificial nodes arrive at predictions.
The scholarship on race and NLP highlights “the need for proactive consideration of how NLP systems can uphold racial hierarchies” (at 1). Because of the algorithmic bias, AI should not replace human evaluation (Data Head at 197). For example, Safiya Noble described “how biases embedded in Google search perpetuate racism and, in extreme cases, perhaps even motivate racial violence” (at 5). Others could manipulate an algorithm to achieve unintended consequences.
Ultimately, the built-in algorithmic bias stems from the common law test. Specifically, the NIL/TU,O Child and Family Services Society case does not account for the use of outdated legal concepts. Even the word “Aboriginal” reflects the colonial attitude and treatment of the Indigenous people. To conclude, Indigenous nations have had “hundreds of years of history of non-native people telling them what to do” (Norma Rae of the Okanagan, The Globe and Mail (8 Nov. 2001) as cited in Crafting an Aboriginal Labour Law at 20). They neither trust the government nor its established labour relations regimes (ibid.) After all, do they need a common law-based labour law classification model?
Oksana Romanov is a 2L law student at the Lincoln Alexander School of Law in Toronto. She is an aspiring employment law lawyer. She is passionate about fostering inclusive communities, effecting social change, and advocating for the human rights of persons with disabilities to remove attitudinal and environmental barriers to their full participation in society. To learn more about the author, you can visit her LinkedIn profile.