Peter Flood, CEO of AIDA Content Management, has recently published an article on LinkedIn entitled The Value of Accurate and Precise In-Frame Metadata or Dealing with the Current Impenetrable Fog of AI, in which he ponders the current state of AI within DAM and the possible negative effect that it might be having.
Peter’s article adds to the groundswell of DAM professionals that regard AI with a certain level of caution and scepticism:
“AI companies are following the same path promising a holy grail of reduction in cost and processing time with output in fact maybe only decreasing the already low value of metadata” [Read More]
He continues to expand his point by referencing a number of quotes from my co-contributor Ralph Windsor, specifically, his article What it Will Take for Artificial Intelligence to Become Useful for DAM, in which he presents the inherent shortcomings of digital asset definitions:
“Metadata is traditionally defined as “data about data.” This definition doesn’t go far enough in explaining its importance. I contend metadata is better described as “contextual data.” Context gives perspective or meaning, and that provides value to users. Objects stored inside DAM systems are called “digital assets” in part because of this — they have a value which is implicit in the description. Meaning and value are synonymous, so a reductionist approach to cataloguing asset metadata, such as AI image recognition tools use, will likely result in a corresponding diminution in the value of digital assets.” [Read More]
Peter goes on to outline a number of current issues with the process of metadata, tagging and keywording, albeit with some nods to his organisation’s VRmeta solution and its ability to provide “accurate, precise object and time-based in-frame meta tagging” using AI and “deep machine learning”.
“The fact is with current meta tagging processes, methods and solutions, digital assets have a value that is already diminutive which we attribute to the concept of keywording or tagging. Tagging will never rise above the capability to generate general and at times ambiguous, misleading and in fact inaccurate tags. And further, the process of keywording, or ‘tagging’ videos can be as exciting as filling out a tax return (as I have heard stated) which would tend to undermine the quality of output.” [Read More]
The first problem is not necessarily the laborious nature of metadata tagging, but rather the tendency for organisations to delegate what they perceive as low-value tasks to the least knowledgeable and qualified staff like interns or general data input operators who simply lack the skills to provide the contextual data required for the asset to achieve its optimum extrinsic value. The second issue is that AI is still in its infancy, a point that the article picks up on:
“All AI engine outcomes returned are only at the level of the current knowledge or learning level that exists. So, at this stage or point on the curve, utilization of an AI API whatever the company, is developing more value for that company then the outcomes it is delivering to clients. To be clear, the output from auto-tagging that meet a client’s requirements does supply some level of value but at this point that is only in a short linear range given these existing constraints.” [Read More]
I did find myself wishing for a more in-depth description of ‘in-frame datasets‘ and ‘in-frame metadata’ – phrases that are referenced throughout the piece when discussing machine learning – yet the tone of the article is optimistic, and aside from a few subtle plugs for VRmeta, mostly impartial.
“That is why precise, accurate in frame datasets for machine learning is the only method to move AI up the learning curve to return optimize value to an AI company’s clients. This is clearly consistent with the way multi-layer neural networks develop knowledge and accuracy and precision of inputs enhance this learning process exponentially.” [Read More]
According to Peter, the adoption of processes that can produce accurate metadata is the only way to realise the true value of digital assets. Whilst I couldn’t agree more, the issue of whether such processes are best suited to humans or machines is still wide open for debate; either way, there’s still plenty of legwork to be done to improve both the current status quo of manually entered metadata and the effectiveness and accuracy of emerging AI solutions.
{ 0 comments }
