Aldis, digital asset management


Artificial Intelligence for DAM is not scary (or perfect)

Artificial Intelligence (AI) is a hot topic, and in the DAM field, it’s sparking a lot of excitement…and a lot of anxiety:

  • Is AI going to replace human librarians? 
  • If I resist using AI, am I missing out?
  • What does it mean to use AI?

The short answers to those questions:

  • No.
  • Yes.
  • Let’s get into it. Here’s a DAM AI primer.

The Big Tent of AI: Definitions and Concepts

Artificial Intelligence is not one thing. It’s a set of systems with one thing in common: they execute tasks that used to exclusively require human intelligence to complete. (They don’t replace human intelligence, and in fact they augment it, but we’ll get into that later.) First, a few key definitions relevant to the DAM industry:

Facial Recognition & Object Recognition

Algorithmic tools that compare visual assets and recognize faces and objects, based on existing visual data.

Optical Character Recognition (OCR)

Software that recognizes text visible in an image and translates it into machine-readable and searchable text.

Duplicate Detection

An algorithmic tool that compares assets and identifies duplicate or similar assets.

Natural Language Processing

Software that parses speech and text, and recognizes meaning in the phrases and sentences. It allows for more complex and sophisticated searches and helps power translations, transcriptions, and chatbots.

Machine Learning (ML)

Broadly speaking, ML means that software can learn from existing datasets by recognizing patterns and analyzing the components of those assets, and then apply that learning to new material, without explicit instruction to do so. Practically speaking, it means a tool can improve.

Human Intelligence is Still Necessary

(and we’re not just saying that because we’re human)

“AI is fantastic, but it’s definitely not a magic bullet,” said Phil Seibel, Aldis librarian.

He frames a cautionary example: “Let’s say you have a DAM system heavy on images of vehicles, lots of different kinds of vehicles, and a brand-new AI tool. This is gonna be great. We'll just aim the AI at all of our footage and it's going to label all this stuff. But here’s what you’ll probably get back: ‘truck, truck, truck, car, truck, truck, car, truck.’

“100% accurate, 100% useless.

“That’s when a human brain comes in, to course-correct and teach the AI to be more specific: maybe it’s model numbers based on distinguishing design features, or recognizing serial numbers through OCR, or looking for key information on license plates.

“You need somebody there to hold AI’s hand and teach it, like you teach a child: ‘this is right; this is wrong; this is close, but we actually want to use this other, different word, not the one you chose.’

“It’ll get better and better as you go, as the machine learning advances, but the need for human hand-holding never stops. It just gets more nuanced, more sophisticated.”

AI Struggles with Context

AI is good at identifying objects, including human objects (aka people), but it struggles with recognizing the subjective context that brings meaning to those objects. In other words – not surprisingly – AI struggles to recognize and interpret human content.

“AI can recognize two people and a taco in a picture, but it’s not going to help anyone understand what’s actually going on,” said Kyle Henke, Aldis librarian. “Are they arguing about who gets to eat the taco? Is one of the people disappointed or envious because the other person isn’t sharing the taco?


“That kind of real interpretation requires human eyes.”

Aldis librarian Ben Zamora-Weiss frames it this way: the more transactional the task is, the better AI is. 

“Transcriptions are almost flawless these days, and translations are also really well done. Facial and object recognition, or things like putting time markers in when a scene is changing in a video – fantastic.

“But when it comes to understanding the context for why a photo was taken, or why a video was created, AI isn’t there yet.”

AI Pro Tips

So, okay, like a lot of new tools, AI has a ton of potential, but it needs help to fulfill it. Below are some tips (and some real-world caveats) to get the most out of AI:

Up your internal audits

Plan to invest time upfront – and indefinitely – in training, monitoring, and correcting AI output. An AI’s performance will change over time, and its development may not be linear. Plan for ongoing audits of your AI tools and the value they bring to your DAM system.

Same goes for machine learning

Despite all the scary stories of machines “learning how to learn” in the near dystopian future, the truth is they still need a lot of management and oversight. (At least for now.) Build a lot of machine-hand-holding into your machine-learning plans, early and often.

Invest in training

Each AI tool requires its own training. Some vendors allow custom training, others do not. Determine if the ability to do your own training or use your own data models is necessary for how you’ll be using AI. For many, the out-of-the-box capabilities are sufficient.

Separate metadata

Draw a bright line between metadata generated by humans and by computers. It’ll make it easier to maintain quality control.

Don’t rely on AI vendors for benchmarking

There is no standard or defined approach for benchmarking AI performance. When you’re choosing which tools and vendors to adopt, it’s up to you to supply a rigorous set of questions and requirements.

An AI Guide

AI can feel overwhelming and intimidating, but, really, it’s just another set of tools to make your DAM better.

We leave you with some big-picture takeaways that apply to any new tool:


Begin with the end in mind. Start with well-defined goals for incorporating the AI tools you’re considering, and commit to a clear strategy. Your overall DAM strategy should include clear goals and a strategy/roadmap for adopting AI, in detail. Exactly which tools and AI concepts you choose depends on your organization. For example, if your business is image-centric – perhaps you manage a huge, constantly evolving photo library – then Facial and Object Recognition are ideas you might want to fast-track.


Ask: What are your people reliably good at? What parts of your workflow could AI be trusted with? Evaluate what strengths both humans and AI bring to the table to get the best of both.


Don’t think “automate and replace human labor.” Think “hybrid approach.” Embrace AI tools as a way to speed up and add value to existing human workflows, not as a replacement for human expertise and knowledge. Think human intelligence with a super-boost of AI augmentation.


Define the new responsibilities of managing AI and who will be responsible for them. And based on the role(s) you want AI to play (defined in your strategy), choose what resources you’ll need for that to succeed.


Integrate the new tools into existing workflows, adapting them as needed to create the “new normal.”


Be prepared for – and build into your DAM governance plan – frequent, ongoing check-ins on the efficacy of your AI/DAM tools. AI is still evolving and improving. Be willing to adapt. It’s never “done.”



Create a powerful media library