91视频

As reliance on artificial intelligence (AI) seeps into everyday life, it was bound to seep into the world of land use. One of us got a cold call from a prospective client indicating that ChatGPT had instructed the person to not just hire an attorney, but to reach out to this attorney in particular, and provided the author鈥檚 personal number rather than office number. By searching for and combining lengthy and complicated land use records 鈥� local zoning regulations, applications, testimony, and staff reports in numerous cases, AI tools are now used by the public and attorneys alike as a convenience to create land use applications and supportive testimony submitted at the local level, as well as briefs submitted to the Land Use Board of Appeals (LUBA) and the appellate courts. This 鈥渃onvenience,鈥� although perhaps providing greater access, is causing clients to question attorney fee estimates, as well as the use of planners and attorneys at all, assuming that AI is equally up to the task. LUBA and the courts have now responded, evaluating these AI-generated efforts. The results are sobering.

What is well-known is that generative AI models hallucinate. Not only do they fabricate cases that do not exist, but they also provide false case quotations and inaccurate descriptions of those cases and yet give all appearances that they are supportive of a party鈥檚 arguments. As the Oregon Court of Appeals recently observed, 鈥済enerative artificial intelligence is not perceiving nonexistent law as the result of a disorder. Rather it is generating nonexistent law in accordance with its design.鈥� Large-language platforms, such as ChatGPT, Copilot and Claude, are trained by observing how phases or words appear together, and to use statistical modeling techniques or extrapolate from its input, so as to support whatever the user desires or its AI device is programmed to desire. However, when AI generates a case citation or holding, it is less prone to test its existence. Because the words sound so right, it is easy to believe that its proposition is, in fact, true.

Now the courts are weighing in. In Ringo v. Colquhoun Design Studio, the Oregon Court of Appeals identified signs of AI hallucinations in an appellate brief and imposed a $2,000 sanction against counsel presenting the brief. The court highlighted an attorney鈥檚 professional duty not to perpetuate falsehoods where the result undermines the rule of law. Notwithstanding the sanction, the court allowed the attorney leave to file a new brief.

In Doiban v. Oregon Liquor and Cannabis Commission, when questioned in oral argument about the use of AI in a brief, petitioner鈥檚 counsel acknowledged that the brief fell short, citing health conditions leading to reliance on staff work that was insufficiently investigated in advance of filing. Petitioner鈥檚 counsel explained that AI was used to create a brief outline, and traditional legal databases 鈥� Lexis and Westlaw 鈥� were consulted for case support. Finding nothing helpful, the effort turned to general search engines, which seemed to generate what appeared to be 鈥渓egitimate legal analysis.鈥� According to counsel, when asking Google whether the fabricated cases are real, the search engine will respond that 鈥渢he fabricated cases are in fact real.鈥� Rather than rely on the fee schedule used in Ringo that would reveal an assessment of $16,500, counsel鈥檚 acknowledgment of the mistake and corrective action to avoid these mistakes in the future, the sanction was reduced to $10,000. Again, a lawyer鈥檚 duty to present a case with candor and truthfulness were the bedrock of the court鈥檚 ruling. Finding that striking the brief in its entirety would harm the client more than their counsel, the court accepted a redacted brief removing all the nonexistent case law coupled with an affirmative statement that the brief did not use generative AI and that all sources of law have been verified.

It is not just the courts; LUBA has now weighed in on the use of AI in briefing. In Bartholomew v. Clackamas County, a self-represented petitioner submitted a brief including case citations that did not exist or relied on authorities that did not stand for the propositions stated. In response to challenge, petitioner explained that the reliance on these sources was pursued in good faith, urging LUBA to overlook the mistakes and consider the arguments without the nonexistent citations and incorrect authorities.

The petitioner noted lack of access to legal databases and a desire not to be 鈥渟hut out of the process.鈥� This argument is interesting because LUBA, unlike the courts, was designed and intended to be open to public participation, including by nonlawyers. Highlighting the additional time necessary for LUBA to decipher these hallucinations as well as the need for 鈥渇air and predictable process,鈥� LUBA concluded that self-represented parties must be held to the same 鈥渞easonable diligence to verify鈥� sources that applies to attorneys. The LUBA appeal process鈥� integrity and efficiency demand as much.

Unlike court accommodations to restore non-hallucinatory briefing, LUBA struck petitioner鈥檚 brief outright and did not consider it, citing statutory short decision-making deadlines. Unlike Oregon courts, LUBA does not have authority to sanction parties for misrepresenting authorities in a brief, although it will be interesting to see if other parties to this case seek recovery of their fees expended in defense, something that the law does allow.

It is likely that the courts and LUBA will adopt rules prohibiting the use of AI or, more likely, demanding an affirmative statement that all sources have been validated. Local land use planning departments may well do the same. Attorneys and land use planners must help bridge the education gap by explaining the significant limitations of AI to their clients. It may well be that reliance on AI may become as common as turning to the dictionary for reference, but participants must acknowledge full responsibility for the products they produce. Land use regulations and laws, and the cases applying them, is a highly developed, sophisticated area. Although ChatGPT can spit out a LUBA brief just as easily as it creates your next vacation itinerary, all participants, including attorneys, must understand the limitations.

Edward Sullivan is a retired practitioner of land use and municipal law with more than 50 years of experience. Contact him at [email protected].

Carrie Richter is an attorney specializing in land use and municipal law at Bateman Seidel. Contact her at 503-972-9903 or [email protected].

The opinions, beliefs and viewpoints expressed in the preceding commentary are those of the authors and do not necessarily reflect the opinions, beliefs and viewpoints of the Daily Journal of Commerce or its editors. Neither author nor the 91视频 guarantees the accuracy or completeness of any information published herein.