The Law Firm AI hallucinations and Genealogy

 

As a former trial attorney, I have been amazed at the recent news reports of law firms being sanctioned in court for submitting pleadings containing AI hallucinations. See "Top Wall Street law firm embarrassed as AI hallucinations derail bankruptcy motion" After spending 39 years as a trial attorney, having opposing counsel submit briefs containing unsupported or inappropriate citations to court cases is not a surprise, but the idea that a law firm would turn over its pleadings to AI without substantiation is almost inconceivable. 

From my own experience, the proper procedure for drafting legal documents starts with vetting supporting case law. The legal system in the United States is based on English civil law and involves the principle of res judicata. That means that arguments in court are supported by existing legal decisions in other courts going back in time. In other words, you select your supporting cases before you begin drafting any kind of legal document. You do not include cases that have not been specifically reviewed. Research precedes writing. 

What is even more amazing about the law firm situation is that AI workflow options exist to eliminate hallucinations. For example, using Google GEMs, Gemini, and Google NotebookLM, the workflow would consist of:

1. Opening a new notebook in NotebookLM and then adding sources which had been previously verified as accurate.
2. Using the information acquired in the NotebookLM to support the drafting of the document in question.

The instructions for drafting the document would have been previously codified in a prompt created by the Google GEMS program. All of this would be supported by research done by Gemini, but because the research is focused into a NotebookLM notebook, the information generated will be forced to comply and operate within the previously examined and included sources. In short, this workflow eliminates any reference to a source that has not already been examined and verified.

Back to the lawyers. If the individuals in the law firm had not reviewed every single case included, that would have been negligence on their part and had nothing at all to do with AI and its possible hallucinations. 

Where does this leave us? As genealogists, we are dealing with the same type of situations where we find sources and we base our conclusions and our entries in our work product in the family tree entries or other presentations, other compilations such as family histories, based on the documents that we have found. Responsibility for the accuracy and applicability of the documents is not left to AI. That decision is the decision made by the genealogist. 

I recently wrote a blog post about the AI Research Assistant from FamilySearch.org. It is apparent that the accuracy of this genealogy research assistant appears to be based on matching names. The depth of research that's necessary to eliminate the same name, same person syndrome has not been programmed into this FamilySearch.org AI research assistant. Accepting the findings of these so-called AI assistants puts a genealogist in the same position as the law firms relying on hallucinations.