One of the dangers of using AI for dangerous goods compliance is that the hallucinations and the errors will become harder and harder to spot as the large language models get better. This creates kind of a conundrum because an advanced AI looks very convincing. I saw one earlier today that was a container filled with dangerous good boxes, but on closer inspections all the boxes or most of them were shown with GHS labels, and some of them were just nonsense and none of them had the requisite transportation labels with the Arabic numeral at the bottom.

So when I saw this image flyby in my feed this afternoon being reposted for the thousandth time I figured I’d grab a copy of it and share a few words of caution about dangerous goods, compliance, and AI, especially large language models. I always try to attribute a post if somebody else made the image, but this has been re-shared so many times I couldn’t figure out who the original author was.

The point to this post anyone reading this can easily spot the errors in the AI image. But the Internet is global so how would somebody and say Mongolia or pretty much anywhere outside of North America and Europe — Would they be so quick to identify the errors? Now imagine this was dangerous goods.

– Jim Powell