Beyond the Machine: 4 Key Limitations of AI in DAM

Photo by Pavel Danilyuk

We tend to think of AI as the great enabler, one of those pivotal technologies we know will transform our lives. The professional landscape will be re-shaped. Our approach to ideas, knowledge and thought, a once unique human process, will be fundamentally challenged.

Hovering on the horizon is the specter of an all-knowing entity, that never forgets, never gets tired of learning, that can analyze unimaginable volumes of information at split-second speeds and which, is by definition soulless. It’s scary and exciting.

In the world of DAM, there is much excitement about AI. Almost every system touts its AI prowess. Tagging, facial recognition, text recognition, AI translations, AI generated captions, search suggestions.

Sometimes it seems like AI has all the answers. Not sure of something? Ask ChatGPT. Want to impress a client? Talk about your latest and greatest AI features.

But let’s get real.

Yes, AI is amazing. AI does save time. It can do a pretty good job of tagging and its ability to identify faces, even in a crowd, is uncanny. But AI has its limitations too (apart from not being able to make tea and toast…yet). These are four key areas where archive owners and DAM providers need to recognize and think about the limits of AI:

4 Key Limitations of AI in Digital Asset Management

Created by Katya Mulvaney

1. AI Doesn’t Understand Context

It sounds blindingly obvious but AI doesn't understand context. Or let’s just say it doesn’t see like we do and can’t therefore know context – although technically it might be able to apply a certain degree of understanding to context if it was aware of all the variables.

Let’s take AI’s analysis of an image. A machine can identify elements and even describe those elements. A computer can recognize a tree, a car, the sun and clouds, and so on. That’s easy. But a machine can’t know relationships. It can’t see an image of two people and know if they are brothers, friends or enemies. These are things that give meaning to a picture but which AI can’t see. AI might be able to see a smile but it could still confuse it with a grimace – an important difference.

In a sense, AI offers a flat and soulless (how could it be otherwise) analysis of an image or video. Yes, some key elements can be identified, but if you’re looking for next-level tagging, that relates to the meaning of image, that references the relationships between people, or the significance of a scene, you’ll bump up against one of the principal limitations of AI.

Truly deep and comprehensive tagging still requires human input.

2. Facial Recognition Has Its Limits

AI does a scarily efficient job of recognizing faces. By this I mean it can identify the uniqueness of a face, without actually knowing who the face belongs to. Yes, you guessed it, it’s up to us humans to put names to faces. The good news is that facial recognition means you only have to put a name to a face once and AI will then identify that face every time it appears in your archive, which is brilliant and a massive time saver.

But even the unnerving accuracy of AI facial recognition has its limits. We encountered these in videos where people are moving around. Faces twist and turn, becoming only partially visible, making it hard for AI to recognize.

The moral of the story is that while AI facial recognition is super useful, there are some situations where it still trips up. Not a huge limitation, but a reality that users would do well to be aware of.

Photo by Andrea Piacquadio

3. AI Translates Without Understanding

It’s sounds odd, given how intelligent AI seems, but remember: AI isn’t human.; Its cleverness isn’t the product of an actual brain. AI services are the pure product of cold computational processes which lack deeper, contextual understanding.

When AI reads and translates, it doesn’t feel the weight and meaning of words. It doesn’t see context. It just sees a string of symbols that it decodes from one language into a another.

As we all know, the result feels stilted., The choice of words is often odd, and sometimes incomprehensible. You get a translation of a kind but it’s rarely usable right off the bat.

It will surely get better, but right now AI translation can’t be fully relied on.

4. AI Doesn’t Know What You’re Thinking or What You Want

The risk is that if we rely too greatly on AI, we’ll supplant human reflection and thought with machine generated ideas.

Many of us have a clear vision of our needs and objectives. AI can be helpful in defining and expressing these, but it can also be a distraction from our own strategic paths.

If you can’t think of an idea, reflexively turning to AI for an answer is likely to make our minds lazy. What if we lose the habit of thinking?

More than anything, our ideas are an extension of our selves, of our unique experiences and of how we have synthezized those into a world view. AI can’t replace that. It has no sensed and lived experiences to draw from and therefore falls short of being able to create ideas that connect on a deeper level.

Photo by Anastasia Shuraeva

AI is Just One Layer of Value

In short, AI is by definition impersonal when human intelligence is unavoidably unique and personal. The uniqueness of human thought cannot be replaced by AI.

At LightRocket Enterprise, we are focused on finding effective solutions to the challenges of digital asset management. AI is a part of that process but is just one layer of value in a multi-layered platform and while we are excited by its potential, we remain alert to its limitations.

Contact us now to find out how we can help you release the value of your archives: info@lightrocket.com

---

Written by Yvan Cohen | Yvan is a Co-Founder of LightRocket and has spent the past two decades immersed in the challenges and realities of digital asset management. As a professional photojournalist, Yvan uses his decades of media experience to help shape LightRocket's world-class DAM platform; focusing on collaboration, intuitive workflows and continuous innovation.

---