AI: Don't Trust And Verify

Yet another set of attorneys are being properly sanctioned for using an AI tool that created fake caselaw and citations that the attorneys filed with the court.

The attorneys did not verify that the answer was actually truthful and the cited cases actually both existed and stood for the proposition the response claimed they did.

The citations in fact neither stood for that position nor even actually existed in reality as AI had completely fabricated them.

That is a problem. 

Filing fraudulent cases then led to appropriate sanctions on the attorneys when court staff checked and saw they did not actually exist. 

This is not the first time this has occurred, and you would think attorneys would have wised up by now and figured out that AI provides a response to your prompt, but that response is not necessarily going to be truthful and at times is completely fraudulent and it is incumbent on you to verify any answer it may give.

At this point if AI tells you the sky is blue, you need to stick your head out the door and verify it, otherwise you're going to miss the very visible oncoming storm that will bite you. 

When you're filing a document with the court, it is on you to verify what you are filing is truthful and correct and crucially that the citation supporting your position actually exist.

Too many attorneys are taking shortcuts trying to save time and using AI tools that do not give valid and correct answers but are not just spouting answers it thinks you want to hear, but are actively producing fabricated citations to back up the false information.

Since AI is this provably bad at providing factual answers for legal questions, what other things are people relying upon and using it for that also will be shown to be fabrications and false?

Chat GPT - Don't Trust, But Verify

Chat GPT as a large language model AI that is supposed to make your research easier.

Just for fun, I decided to give it a try and asked it a legal question regarding a point of law on a case I'm working on. 

The answer is known to lawyers quite well as it is a pretty basic property principal that a possessing joint owner of a property does not owe rent to a non-possessor joint owner except under certain circumstances.

Finding Michigan state case law on this principle is somewhat difficult - because everyone knows the rule so there really hasn't been cases litigating it (there's some good cases on this situation when they are tenants in common but that's not the same as a joint tenancy).

So, I asked Chat GPT the question and it came back with the rule as expected, that rent isn't owed except under certain circumstances, and very quickly, too. 

It also came back with two Michigan cases with quotes supporting this rule - complete with very nice authoritative and proper legal cites to the two cases it helpfully cited.

Which do not exist.

Nor do the quotes themselves exist. 

Yes, I checked.

The citations, while looking completely correct, do not go to the case names Chat GPT identified as the cases, but to other totally unrelated cases, and those case names themselves either do not exist or are cases with completely different citations and for completely unrelated issues.

Chat GPT completely made up two case quotes and legal citations to them, probably trying to be helpful.

It's a good thing I did not blindly accept such good sounding quotes as factual but tried to verify them on a proper legal research tool.  I'm rather suspicious like that, and as an attorney you can never use made up cases to support your position.

Chat GPT is certainly not ready for prime time when it comes to legal research and I would be rather skeptical of it being an authoritative source for anything else without using it as a springboard to find verifiable and authoritative sources. 

Caveat usor.