AI in DAM: the challenges
Of course, it isn’t all plain sailing. AI is in its infancy and the technology isn’t perfect yet. Here are some of the challenges you might encounter with AI-powered Digital Asset Management.
AI in DAM is still learning
Your DAM AI plug-in can only tag subjects it’s already learned about. This means it’s great on common subjects - like landmarks, famous faces, flora and fauna, weather, and colors. But it might struggle with more subtle or abstract subjects. This is especially true if an image is particularly complex and includes a lot of detail/different objects.
AI in DAM is generic
AI is trained to recognize generic subjects that are common to all experiences, cultures, and businesses. But it doesn’t know anything about your specific business - yet. You’ll need to train it. And that takes time and expertise.
For example, imagine you sell pottery. Your AI might recognize several different products as ‘plates’ but lack the nuance to tag them as side plates, dinner plates, porcelain, or stoneware… In the context of your DAM system, having lots of different products tagged as ‘plates’ - when you want to find images of a porcelain side plate - isn’t very helpful.
AI and DAM master lists don’t always mix
If your DAM uses a master keyword list, you may struggle to get it to accept the auto-tags generated by your AI. If you’ve locked down a controlled vocabulary for keywords - and your AI is suggesting tags that aren’t on the list - this can prevent them from being added. One way around this is to have two separate keyword fields, one for manual entry from your master list and one for keywords generated by your AI.
AI in DAM gets it wrong
AI and human image recognition are very different. Whereas humans understand the content of an image, AI just recognizes patterns in the pixels. That means it can make a strong guesstimate as to the content - but it won’t always get it right.
To prevent your search results from getting cluttered with irrelevant assets, you may need to edit and remove auto-tag errors. In a larger enterprise, this could require an additional team member to quality-check the AI tagging.
Sometimes VERY wrong
Sometimes, AI tagging doesn’t just result in mislabelled digital assets and the inconvenience they cause. Sometimes AI tagging can reflect biases and errors found in real life which, if admitted into your DAM unquestioned, may cause deeper issues.
For example, when this writer used a stock photo library to find images of breastfeeding, it was evident that AI tagging was in play. As well as appropriate keywords - such as motherhood, feeding, and infant - the images had been erroneously tagged with sexual keywords as well. In this case, we can assume the AI had correctly identified the anatomy but failed to understand the context.
More worryingly, some AI has been found to perpetuate racist and sexist prejudice. Not something any Digital Asset Manager wants to introduce into their business.