We get asked about Artificial Intelligence (AI) and machine learning quite a lot. People want to know if it is part of our DAM, or if we plan to implement it any time soon. At the moment, the answer is No. We believe that AI in its current state does not bring significant value to our clients. When it comes to adding metadata to your assets, we believe the human brain is still the best tool and digital asset management is an inherently human endeavor.
When people ask about AI, they are usually asking in reference to adding metadata. Metadata, the information that you add to give a file context, is fundamental to digital asset management. This kind of information can include asset creator, rights and usage, product line, description, details about what is in an image or video, tagged keywords relevant to your business, and much more. Metadata is what will help you find what you need in your search, and will also make sure that you use the file properly. The hope for AI is that it will make adding metadata quicker and easier, but we believe there are several fundamental flaws with it at this time.
Real Learning Vs. Image Recognition
AI in most instances is more pixel recognition than machine learning. An AI processor may be able to recognize “bridge” for example, but it won’t be able to tell you the information that is actually useful to your business including location, engineer, construction firm, etc. The same goes for products; it may recognize something as a bottle, but it can’t tell you the product line, what product it is, or any other defining details. Say you are a shoe company, how useful is it to have 10,000 images of your entire product line all tagged with “shoe”? Or for a university, tagging something as “building” when what you want to know is what images are of the Russell Library.
The information that these systems can recognize is very narrow, based solely on the pixels it recognizes and with none of the context that our brains might imbue. The metadata that is actually useful to your business is the information and context that your employees and colleagues have stored in their heads and this context is crucial. This is what will make for the best metadata and in return the best search results.
In a fascinating review of computer scientist Melanie Mitchell’s book Artificial Intelligence: A Guide for Thinking Humans, the author states that “appreciating that machine-learning systems are essential statistics workhorses makes it obvious why they are so vulnerable to inaccuracies, holes, and other shortcomings in the data they’re fed.” Without the ability to think beyond the narrow margins that they have been trained for, AI processors are a long way off from being able to understand the type of contextual metadata that is useful for humans to utilize. Put simply, AI can’t do judgment. Humans can, and are great at it. That judgment is how you add value to the assets your team needs to use.
AI and Your DAM
Many vendors state that AI will help improve metadata entry speed, but we have found this not to be the case. Any of the specific details that need to be added will have to be done by a human, so going back through all of the data that the AI added and amending it can be a big waste of time. In addition, much of the metadata that is added may be irrelevant to what your users are searching for and add extra clutter to your results.
Artificial Intelligence also needs to be trained. It doesn’t start with all of the correct information and implement automatically in your new system. If you have a dedicated staff member, and thousands of images to upload and tag each month, then perhaps you can start training your AI to work towards some of your goals. But If you are small to mid-size organization, like the bulk of Image Relay’s clients, then you probably do not have the time or the resources to fully train an AI processor.
We have seen other vendors demo AI (we won’t name names!), and the results were not good. Simple images such as animals, or a person smiling came up with good enough simple metadata tags, but once images become more complex, the metadata becomes random and irrelevant. This then needs to be removed from a file before it can be used.
In their book Rebooting A.I.: Building Artificial Intelligence We Can Trust, New York University researchers Gary Marcus and Ernest Davis state that people often think computers are smarter than they are due to a “gullibility gap.” Many DAM vendors will take advantage of this by trying to upsell products that have not been fully realized. Some organizations ask to have their AI tagging turned off after several months of use when the realize the true state of their AI processors.
The Future of AI
AI certainly has potential, and large businesses such as Google, Amazon, and Microsoft are working continually to improve it. If at some point we feel that the rewards from using AI are worth it and will greatly benefit our customers, we will be ready to implement. Image Relay has an open and flexible API, which means we can easily integrate a third-party AI processor.
We believe that our mission is to make the most user-friendly and useful DAM on the market. We choose to focus on implementing features that help meet this mission. When we feel that implementing AI will also serve this mission, we will implement it. But at the moment AI is still in development and ineffective at adding high-quality and useful metadata.