AI in Family Law: Beyond the Hype

The legal profession's relationship with AI has been, to put it diplomatically, complicated.

On one hand, the 2024 Legal Trends Report shows solo practitioners leading AI adoption, and 65%+ of legal practices now prioritize automation. On the other hand, Ko v. Li — the Ontario case where fabricated AI-generated case law was cited in a family law factum — demonstrated exactly what happens when AI is used without understanding its limitations.

Both of these things can be true simultaneously. AI is genuinely useful for certain tasks in family law. It is genuinely dangerous for others. The challenge for practitioners is distinguishing between the two.

What AI Is Good At in Family Law

AI's strengths in legal practice are well-suited to several specific pain points in family law work.

Data extraction from documents

Family law matters generate mountains of financial documents — T1 General forms, Notices of Assessment, pay stubs, financial statements. Manually transcribing data from these documents into calculation software is tedious, time-consuming, and error-prone. Modern AI can read these documents and extract structured data with high accuracy. The key word is “extract” — the AI is identifying and copying information that already exists in the document, not generating new information.

Plain-language explanations

One of the most time-consuming parts of family law practice is explaining calculation results to clients. “Your spousal support range is $1,200 to $1,600 per month” is a number. What the client needs is a full explanation of why that range applies to their situation. AI can generate these explanations accurately because the underlying data is deterministic and structured.

Scenario exploration

“What if custody goes from sole to shared?” “What if we include the Section 7 expenses for hockey?” “What if the payor's income increases by $15,000 next year?” Each of these questions, in traditional software, requires manually re-entering or adjusting data and running a new calculation. AI can interpret these natural-language questions, make the appropriate adjustments, and present the comparison — dramatically reducing the time to explore alternatives.

Anomaly detection

“This self-employment income seems low relative to the business revenue.” “The reported childcare expenses are above the 90th percentile for this region.” AI can identify patterns that warrant a closer look — not to make judgments, but to flag things the lawyer might want to investigate.

What AI Should Never Do in Family Law

The Ko v. Li case made one thing clear: AI should not generate legal content that practitioners rely on without verification. But the lesson extends beyond case citations.

AI should not generate calculations

Child support table lookups, SSAG formulas, tax implications — these must be computed by deterministic engines. A deterministic engine produces the same output for the same input every time. It follows the Federal Child Support Guidelines tables exactly. It applies the SSAG formulas as written. There's no probability, no approximation, no hallucination risk.

AI should not determine legal entitlement

Whether a spouse is entitled to support — and on what basis (compensatory, non-compensatory, or contractual) — is a legal determination that requires understanding the specific facts, the applicable law, and the relevant case law. AI can surface relevant considerations, but the determination belongs to the lawyer.

AI should not make recommendations about positioning within ranges

The SSAG give a range. Where to recommend within that range involves weighing competing factors, understanding local judicial tendencies, and exercising professional judgment. AI can present the factors. The recommendation is the lawyer's.

The Architecture That Matters

The distinction that matters isn't “AI vs. no AI.” It's the architecture — specifically, how AI and deterministic calculation engines interact.

Layer 1: Deterministic calculation engine

Pure functions. No AI. Child support tables, SSAG formulas, tax calculations, Section 7 apportionments. Given the same inputs, it always produces the same outputs. This is the mathematical foundation.

Layer 2: AI orchestration

The AI handles intake and extraction (reading documents), scenario exploration (interpreting natural-language questions), explanation generation (translating results into plain language), and anomaly flagging (identifying data that warrants a closer look). The AI never touches the math.

Layer 3: Human-in-the-loop

Every AI suggestion — every extracted data point, every flagged anomaly — requires lawyer review and approval before it enters the calculation engine. The lawyer is always the final authority.

Layer 4: Audit trail

Every input, every AI extraction, every lawyer approval, every calculation result is recorded in an immutable, time-stamped log. If a judge asks “how did you arrive at this number?” — the answer is a complete, traceable chain from source document to final calculation.

The Law Societies Are Watching — And That's a Good Thing

Every major Canadian law society has now published AI guidance. The Law Society of Ontario's April 2024 practice note on generative AI. The Law Society of Alberta's Generative AI Playbook. The Law Society of BC's practice resource on professional responsibility and AI. The CBA's national ethics toolkit published in November 2024.

The common thread: lawyers have a technological competence obligation that extends to understanding the tools they use, including AI tools. Using AI doesn't relieve the duty of diligence — it creates a new dimension of it.

Beyond the Hype

AI in family law isn't magic. It won't replace lawyers. It won't eliminate the need for professional judgment. It won't solve the access-to-justice crisis on its own.

What it can do — if built responsibly — is eliminate the drudge work that consumes hours of a practitioner's day. The data entry. The re-entry. The manual scenario modeling. The time spent translating calculation results into language clients understand.

That's not hype. That's automation applied to the right problems, with the right safeguards, in a profession that needs it.

Divorce Copilot is built on the architecture described in this article: deterministic calculations, AI orchestration, human-in-the-loop verification, and a complete audit trail. See it in action — start your free trial →