Lawyers: AI's Usefulness May Lie in Culling Auditable Import Documentation
While AI technology providers often tout their ability to classify imported goods, AI might be more useful in helping importers and customs brokers document a large amount of entry data, according to Phillip Poland, who was speaking on an Oct. 22 webinar sponsored by the American Association of Exporters and Importers on AI and reasonable care.
Sign up for a free preview to unlock the rest of this article
If your job depends on informed compliance, you need International Trade Today. Delivered every business day and available any time online, only International Trade Today helps you stay current on the increasingly complex international trade regulatory environment.
Poland, a trade lawyer and head of regulatory compliance with Illuminate USA, a solar panel manufacturer, described a situation at an importer where the customs brokers were less interested in how quickly an AI tool was able to classify a good and instead wanted to understand how the software arrived at its conclusions. As a result, the software now helps brokers understand why a good might be classified the way it is -- which helps not only with collecting documentation but with making that documentation collection much more efficient, Poland said.
The AI software was "asking us questions, and it's prompting us with questions, and we may not have thought about those questions had we not been in the software," Poland said. "And the software is pulling rulings, and the software is pulling explanatory notes, and the software is pulling all the chapter notes. I don't have to go flip -- it's putting it all there, and it's giving me an organized question flow. Even if it saves us a fraction of the time, it's OK, because it's documenting every decision that we make. And every time we click yes or no, or we add in additional information, all of that is recorded."
Looking at all the assumptions that the software makes also gives the importers and brokers confidence that they are still meeting CBP's expectations of reasonable care, Poland said.
"The positive with the AI software is we now have a really solid, auditable record. Even if it took 20 minutes to classify today with the software, and it took 20 minutes in the book, it's worth it, because it's documented, and you can look in there and you can see, here’s every assumption," Poland said. "Here's everything that it looked at... . I actually think it really improves reasonable care, and I really believe that AI makes us a more responsible importer."
For importers, reasonable care may mean ensuring that the information they share with brokers, which may be AI-generated data, is accurate because the broker relies on that information, and importers may be held liable if that information is false, panelists said during the webinar.
However, the language in the law governing reasonable care is broad, and Congess eventually needs to update customs laws to differentiate between levels of negligence, especially as there are only a few court cases that attorneys can refer to that directly address reasonable care, panelists said.
"What has finally come out is that Customs has said it interprets the law to mean the importers are obligated to get the classification value and any other specifics correct at time of entry, hence the opportunity to file reconciliation entries if there's an error as a way of flagging for customs up front that there's going to be a change," said Su Ross, a lawyer with Mitchell Silberberg.
While an importer may disagree with CBP's assessment about a filing, it still has to follow the agency's directions, Ross said.
She added that AI technology software can more easily handle simple classifications, but it might still need more work extracting and then analyzing data elements from various documents related to valuation and country of origin.
"I think it's going to happen eventually, but it's going to be a while before that can get done. And similarly, I don't know that the AI, right now at least, is really good with concepts like essential character or substantial transformation. Is it possible they could be down the road? Yeah, but that's going to take a fair amount of work," Ross said.
Poland said there may be software that is capable of that, but the trade needs to "play" with it and see if the data is being analyzed properly. But what's also important is that the software gets the right data in the first place, the panelists said.
"What it's really good at doing is scraping data from documents ... [But] it's not analyzing it to say, 'Hi, I wonder if you actually gave me the correct data,'" Poland said. "So, the human being has to go in and do that next level of review."
And this is where reasonable care also has a role in ensuring the data analysis is correct, and no AI hallucinations have been made. "A computer cannot hit send, or you shouldn't have a computer hit send," Poland said. "There's still your responsible supervision, your reasonable care, because you're still going to go in and look at it quickly to make sure that it's correct."
Industry also needs to make sure the AI software they use receives frequent updates, as many trade announcements lately have occurred on a Friday evening and have an implementation date of Monday or just days later, Ross said.
Meanwhile, customs brokers need to be cautious about where they input the data they receive from importers so that the brokers don't break any confidentiality agreements, Ross continued.
"Do not use ChatGPT. Now, I'm going to tell you this, not because it's not a good thing to use, but because everything that you input into it goes into its machine learning. It goes into its own database. So if you give it names and dates and values or prices or whatever, the confidentiality of that is gone," Ross said.