When it’s not smart to use Artificial Intelligence

Artificial Intelligence feels like it has been prominent in the news for some time. Stories about the potential replacement of human workers by AI or the potential economic crash caused by the bubble created by AI stocks have been everywhere. Post-secondary institutions have been grappling with how to address students’ use of AI in submitted work. Fake photos and videos created by AI have fooled many of us, sometimes with serious consequences. Courts have also been dealing with the implications of AI for matters that come before the courts.

In 2023 the Alberta Courts issued a Notice to the Profession called Ensuring the Integrity of Court Submissions When Using Large Language Models. This Notice urges parties to exercise caution in the use of AI, suggesting that parties only use case authorities from authoritative websites. The notice also requires that any AI-generated submissions be “verified with meaningful human control.”

Unfortunately, not all parties to Alberta Court proceedings appear to have been following the court’s clear guidance. In a recent case, the Court of Appeal dealt with a commercial law dispute about a party’s failure to respond to certain questions asked of him as part of the questioning process leading to trial. The trial judge found the party in contempt for his failure to respond and he appealed that decision to the Court of Appeal. The Court of Appeal concluded that the party remained in contempt, but not for all the same grounds and reasons provided by the trial judge.

The appellant’s written argument that was filed with the court prior to the appeal being heard contained, as expected, references to previously decided cases. The respondent’s counsel spent considerable time searching for the cases referred to in the appellant’s written argument. However, he was unable to find them, because seven of those cases did not exist. The appellant’s counsel had hired a contractor to prepare the legal argument and, despite initial assurances that no AI had been used to create the written argument, the cases in question were clearly generated (fabricated) by AI.

As noted by the court, the appellant’s counsel stated that, “he was ill, very busy, and it was the holiday season when the factum was due, which contributed to him failing to properly review the original factum and recognize the issue with several cited cases.” The court did not accept this as a compelling explanation and invited the parties to make submissions about costs payable directly by the lawyer for the appellant. It is very unusual for costs to be awarded directly against counsel. Such costs are almost always payable by the party.

Ultimately the court ordered the appellant’s counsel to pay $17,550 out of his own pocket – an expensive lesson!

In another recent decision, the Court of Appeal dealt with a party to a family law dispute who was not represented by counsel. This party provided written argument to the court and her written argument contained three authorities that did not exist. The appellant acknowledged that her submissions had been generated through the use of AI tools. The court dismissed the appeal and awarded an extra $500 in costs to be paid by the self-represented party. Another expensive lesson – AI didn’t win the appeal and it cost the person an extra $500.

These Alberta decisions follow similar decisions from other jurisdictions where lawyers who similarly claimed to be ill, very busy or overburdened by the holiday season submitted completely fabricated authorities to the courts and suffered the consequences.

None of this is to suggest that AI has no role in legal research or writing. It is clearly an important resource that can save lawyers time and clients money. However, all participants in the court system, trained lawyers and self-represented folks alike, must be aware of the limitations and pitfalls of using AI and must verify that the assertions and cases provided through AI searches actually exist and that they say what the AI model says they say.

As the Law Society of Alberta’s The Generative AI Playbook publication states, there are notorious examples of AI systems making up names, dates, historical events and legal cases, including actually writing fake reasons associated with the fake cases. If lawyers have to review AI generated material to determine if every fact, case or excerpt is correct (which is what the court currently requires), there may not be much savings at all in using AI to produce written argument to be submitted to a court.

The same limitations and cautions apply to the use of AI in any other area of work and we should all bear that in mind. 

Editor’s note: The views, perspectives and opinions in this article are solely the author’s and do not necessarily represent those of the AMA.