Every couple weeks lately it seems like another artificial intelligence (AI) milestone is achieved or some impressive AI-driven stunt makes the news. Self-driving cars, computers beating humans at the most complex games of strategy, imitating the style of Rembrandt paintings, even predicting the outcome of a popular TV show are a few AI stories of the past few months. If this keeps up we may soon face AI fatigue.
But don’t stop paying attention. This is going to get really interesting.
A lot has happened since Deep Blue, IBM’s chess-playing computer, beat Garry Kasparov in 1997 and the company’s Watson system competed against two former “Jeopardy!” champions in 2011. Watson’s AI algorithms studied 200 million pages of mostly unstructured web pages to cram for the competition and pulled off a handy win on the show. It was a stunning demonstration of the power of machine learning.
In March of this year, AlphaGo, a computer program developed by Google’s DeepMind division to play the board game Go, competed against a master Go player, Lee Sedol. In what Go masters have since studied and widely praised as a series of “brilliant” moves, AlphaGo defeated Sedol, considered by some to be the best Go player in the world. Go is orders of magnitude more complex than chess, so this is a remarkable accomplishment. AlphaGo gained its chops using a technique called “reinforcement learning”. It played a second copy of itself over and over and along the way learned how to discover new strategies.
Also in March, an AI program in Japan wrote a short-form novel titled “The Day a Computer Writes a Novel” which passed the first-round screening of a national literary competition. The judges acknowledge weaknesses, but said they were unaware it was “written” by a computer program.
But it’s not literature or games or even popular TV show analysis that makes AI interesting in the Digital Asset Management (DAM) realm. It’s all about unlocking more value in your assets.
MerlinOne was a pioneer in this realm as perhaps the first DAM provider to employ AI-powered speech recognition to extract searchable text from audio and video assets. That was an early toe-dip in the AI pond.
A few weeks ago, MerlinOne announced the new Automated Metadata Enhancement (AME) feature being piloted by a handful of MerlinOne customers. It’s pretty cool and likely to get way better as the technology evolves. In its current incarnation incoming images (just photos at this time) are analyzed to extract identifiable features. Those features are turned into searchable terms. Of course, a human adding metadata can do much better, but for images with no metadata AME scores remarkably well at making images searchable on likely terms where none would otherwise exist.
AME performs a couple other neat tricks. It identifies foreground and background color information which can be searched. It also gives a ranking of “memorability”. This computes a value of how likely the algorithm determines the image is to be memorable to humans. It’s kind of spooky how well it does.
These are really just the first baby steps of how AI can enhance the utility of DAM, but it’s pretty exciting. Stay tuned.
MerlinX Development Manager
If you would like to learn more about MerlinOne’s products and services please…