This blog series about a single source of truth for product content is based on the webinar, Let’s Get Digital, presented for the CHPA Academy in March 2024.
by Lisa Lopez, HRG data assets manager, part of the Single Source of Truth blog series
Data is captured many different ways and artificial intelligence (AI) is getting a lot of buzz. AI is an amazingly impressive tool, and it offers a lot of efficiencies, but the human touch is definitely still needed.
In this example AI changed the meaning of the text on the packaging and then skips a large portion of the data. Most likely due to machine learning and the training data used to train the AI language model.
In this instance, the actual packaging reads “or are planning to undergo any clinical lab testing” and the AI captured as “Or are planning to have a child” — quite different meanings!
AI can be problematic because of its learning bias. If the AI model has primarily been fed food data and you are now using it to process health, beauty, and wellness items, you can easily see how the machine learning caused a change from “witch hazel” to “witch hazelnuts.” In the previous data, the word “hazel” never stood on its own, so the model applied what had been learned from the dataset it has worked with the most. This unintentional bias will have to either be accounted for or corrected later.
HRG’s philosophy is to enter product data exactly as it appears on the packaging. In this instance, there is no change to the meaning, but it is not captured exactly as it is displayed on the packaging because the word “at” was adjusted to “between.”
Initial observations and common problems with AI — columns of data on a single panel is something that AI seems to struggle to capture and piece together in the correct order as it is displayed based on the packaging. In this example the highlighted text was completely skipped over and not captured.
If you are not managing to a single source of truth, you may have incomplete and misguided information. In this example, you would miss the intended results for a customer searching “can I take with warfarin?” because the statement of “Ask a doctor or pharmacist before use if you are taking the blood thinning drug warfarin” was not captured.