Authored by Mario D. Bellissimo, Founder, Bellissimo Law Group.
This is not a paper. It is an opinion and more than that, an invitation.
An invitation to question whether the way we currently adjudicate immigration, citizenship, and refugee cases is still defensible in a world where everything else is accelerating. This is not an attempt to provide final answers, complete doctrine, or settled conclusions. It is an effort to start a conversation that is both necessary and, increasingly, unavoidable.
Because the reality is this: justice delayed is no longer an abstract concern, it is justice denied in real time.
Backlogs continue to grow. Processes become more layered. Outcomes are delayed far beyond what fairness should tolerate. And yet, much of our response remains incremental adjusting timelines, adding procedural steps, redistributing workloads while the underlying pressure continues to build.
At the same time, a parallel transformation is underway. Parties themselves, lawyers, consultants, and increasingly self-represented litigants are beginning to use technology to prepare cases faster, at greater scale, and with increasing sophistication. Records can be assembled more quickly. Submissions can be generated and refined rapidly. What once took weeks now takes hours.
This creates a structural imbalance: the inputs into the system are accelerating, while the adjudicative capacity remains largely fixed. That imbalance cannot hold. And if it does not hold, then the question becomes unavoidable:
What happens when the system can no longer keep up with the cases brought before it?
Some may argue it already has. If we accept that justice delayed is justice denied, then maintaining a model that cannot scale is not neutral. It is a choice, with consequences.
A Necessary Shift: AI as an Initial Decision-Maker
Against that backdrop, we should begin to seriously consider a shift that would have seemed implausible not long ago: the use of AI as an initial decision-maker.
This proposal will attract immediate resistance and it should. But it is important to be clear about what is being suggested, and what is not.
This is not about replacing lawyers, representatives, tribunal members, or judges. It is not about removing human judgment from the system. It is about repositioning human judgment where it is most needed, and allowing technology to manage what the system can no longer handle at scale.
In this model:
- Lawyers and representatives remain essential at the front end, diagnosing and preparing cases
- AI produces an initial, reasoned initial decision with reasons, based on that record
- Human adjudicators and judges remain at the back end, reviewing, correcting, and ultimately taking responsibility
The question is not whether AI can replace human decision-makers. The question is whether human decision-makers, as currently structured, can continue to meet demand without it.
From Concept to Practice: Where This Begins
If this idea is to be taken seriously, it must be grounded in practical implementation. In the immigration and citizenship context, high-volume, rule-bound decisions are the logical starting point, for example:
- Temporary resident visas, study permits, and work permits
- Sponsorship eligibility and admissibility assessments
- Citizenship grants based on statutory criteria
These are not trivial decisions but they are structured, repeatable, and already processed at scale. In the refugee context, the approach must be more measured for example in:
- Manifestly well-founded claims
- Clearly ineligible or abandoned claims
- Pre-hearing triage to identify determinative issues
The goal is not to hand over the hardest cases first. It is to begin where the system is already under pressure and where structured decision-making can be responsibly supported.
In each case, AI would generate an initial decision with reasons. And this raises a provocative but necessary question: If a machine can produce a structured, reasoned decision on a complete record, what exactly are we preserving by insisting that the first decision must always be human?
Rethinking Appeals and Judicial Review
If the front end changes, the back end cannot remain untouched. At the tribunal level, appeals could become more focused and disciplined, targeting specific errors:
- Misapplication of legal standards
- Failure to consider relevant evidence
- Credibility determinations requiring human reassessment
The process becomes less about starting over, and more about engaging with a defined, reasoned outcome.
At the Federal Court level, judicial review is not an appeal. But it is also not immune to change.
The current model, leave, judicial review, redetermination was not designed for a world of exponentially increasing volume and technologically accelerated inputs. And yet we continue to rely on it as though it were. So the question must be asked:
Why should judicial review remain structurally fixed if everything around it is changing? A reimagined model could include:
- AI-assisted triage at the leave stage
- More interactive engagement with the reasoning underlying decisions
- Greater use of issue-based determinations
- For example, sponsorship applications may need to be decided sooner than, say, a refused eTA matter. This suggests that access should not be organized solely on a “first filed, first decided” basis, but may require a hierarchy that accounts for urgency, impact, and the nature of the application.
- Legislative reform allowing courts, in appropriate cases, to finalize outcomes rather than defaulting to redetermination
These changes would not be simple. They would require legislative action, institutional coordination, and careful attention to constitutional principles. But the alternative is to preserve a structure that may no longer be capable of delivering timely justice.
Transparency and Accountability: The Non-Negotiables
The objections to this model are serious, and they are valid. AI is not neutral. It reflects the assumptions, data, and priorities embedded in its design. It can replicate bias. It can obscure reasoning. It can fail in ways that are difficult to detect.
For that reason, transparency cannot be optional.
Litigants must participate in the AI design, know when AI is used, how it is used, and what role it plays in the decision-making process. Systems must be auditable, explainable, and subject to independent review. Similarly, accountability must remain human. There is a meaningful distinction between:
- Automating decisions, and
- Using AI to structure and support reasoning
As AI moves closer to the substance of outcomes, the requirement for human responsibility must become more not less visible.
The decision-maker, in law, cannot be a machine at all stages. But at certain steps it can and the process that informs that decision-maker may increasingly be.
The Individual Experience of Justice
Even if the system works, will it feel just?
This may be the hardest question of all.
In immigration and refugee contexts, individuals often seek more than correctness. They seek recognition. They want to feel heard.
A negative decision delivered by AI may feel colder, more distant, and less legitimate.
But we must also confront the reality that delay, inconsistency, and opacity undermine trust in their own way.
So the question becomes:
Is a slower, human system necessarily more just, or simply more familiar?
The answer is not obvious. But it cannot be avoided.
The challenge is not choosing between human and machine. It is designing a system where timeliness, fairness, and human dignity are not in competition.
The Digital Divide: A Reason to Design Better, Not to Stand Still
One of the most legitimate concerns with any move toward AI-enabled adjudication is the digital divide. Not all litigants have access to technology. Not all have the ability to use it effectively.
And in the immigration and refugee context in particular, many individuals are already navigating language barriers, economic constraints, and systemic disadvantage. The risk is real: a system that relies too heavily on technology could deepen existing inequalities.
But this concern must be confronted honestly.
Because the current system already produces unequal outcomes just in less visible ways. Those with better resources, stronger representation, and greater familiarity with the system are already better positioned to navigate complexity and delay. Inequality is not introduced by technology; it is often revealed by it. The question, then, is not whether to move forward, but how. A human-centred AI model must include:
- Accessible, simplified interfaces, not increased complexity
- A strengthened role for counsel and representatives, especially for vulnerable populations
- Publicly supported access points, including clinics and community organizations equipped with the same tools
- And critically, the right to meaningful human engagement at key stages, regardless of technological access
If designed properly, automation does not have to exclude. It can, in fact, lower barriers, reduce delay, and make the system more navigable for those who are currently the most disadvantaged.
But that outcome is not automatic.
It must be built. And it will take time.
The Path Forward
How often do we hear we are at an inflection point?
In part because new inflection points are arising more frequently given the pace and the scope of automation. The core tension remains. We can continue to rely on a system that is increasingly strained, hoping that incremental adjustments will keep it functional.
Or we can acknowledge that the conditions have changed and that the system must change with them. From my perspective, we have not gone too far.
If anything, we have not gone far enough in intentionally designing how AI should function within adjudication.
Because doing nothing is not a neutral position.
If justice delayed is justice denied, then a system that cannot keep pace with demand is not simply inefficient, it is unjust.
The future should not be one where machines replace judgment.
It should be one where they enable it at scale where lawyers remain essential, adjudicators remain accountable, and individuals receive decisions that are not only fair in principle, but timely in practice and of value.
If we are serious about preserving the integrity of our system, then we must be equally serious about rethinking how it works.
Not someday.
Now.
