AI is now more than a buzzword for all the Martech stack. Artificial intelligence (#AI) and machine learning are dominating the Martech right now. We just have to see how much #Google, #Microsoft and #Amazon are investing in these technologies to understand how big the role they will play in our future lives both professionally and personally. But there’s always a lot of hype surrounding new technologies and it’s sometimes difficult to know where we really stand.
A selection of quotes I think you already heard but a clear representation of #customers thoughts :
“ By 2020, customers will manage 85% of their relationship with the entreprise without interacting with a human”
“61% of marketers say artificial intelligence is the most important aspect of their data strategy”
“80% of business and tech leaders say AI already boosts productivity”
One of my favorite technology evangelist is Luc Julia, co-creator of Siri and VP of Innovation at Samsung who has just published early this year a book titled “L’intelligence artificielle n’existe pas”, meaning “Artificial intelligence doesn’t exist”. Luc Julia attempts to make things clear and helps readers understanding the difference between “Hollywood’s Artificial Intelligence” and real “Artificial Intelligence”, which he rather wants to call “Augmented intelligence”. He is giving us one very basic example… AI needs hundred of images to be able to recognize a cat when a 2 years old baby can do the same job after 2 images. Just like the fact and I do agree with him, the autonomous car will never exist because you need to reach level 5/5 for that not 3 or 4.
Just 2 days before the DAM CHI 2019 organised by Henry Stewart Event we can look at the implementation of AI technologies into #DAM systems and marketing plans as a whole. Since the last years #AI is more and more one of the main question for customers during the conference. Which benefits I can get for my metadata, workflows or experience ?
Today most of marketing technology platforms make use of AI and ML algorithms to bring more efficiency, to analyze large data sets and automate many of the processes which used to be done manually. But we don’t have to confuse automation and AI which is not obvious for both customers and vendors. When AI is able to analyze data and apply actions based on new informations, automation is just following rules or programmed code and can not react to a new situation. This is the job of vendors to enable AI within automation in order to bring new smart predictive workflows able to learn from the user behaviour inside the system.
The first impact of AI by customers is image-recognition tools inside DAM systems. Products like Google Vision, Azure Vision, AWS Rekognition, Clarifai or Imagga can help surface relevant metadata and tags, which can streamline asset ingestion and improves metadata governance.
Many vendors provide a set of features over stable APIs. Tagging works but is generally unspecific and often customer context is missing. You can also find On-Premise solutions to enable securely process for sensitive data. Similarity drives use cases like deduplication, metadata context without extensive training, facial recognition ( GDPR/DSGVO personal data identification and protection).
A lot of vendors are now proving such AI auto tagging but most of them are only providing english keywords. Stamp your vendors to get the translation because yes they can.
But I said at the early beginning even if when you ar using these technologies, #AI is driven and trained by human. Do you know the company figure eight ?
In fact these guys are working for some of the best retailers, brands and Google using task force with people behind their laptop to identity images..for just a couple of $ per hour, not sure everybody would like to do this dream job 🙂
The same for Amazon with Amazon Mechanical Turk…
2- Image retouching technology
AI is also more and more integrated in retouching software. Now you can make dozens of adjustments using one slider. New “human-aware” technology recognizes people and applies adjustments selectively for more realistic results. You can quickly delete unwanted objects, fix skin imperfections fast and restore old photos in just few clicks. All these technologies are used live by your last smartphone processing 1 trillion of operations in just one shoot. Next-generation Smart HDR uses advanced algorithms to finesse highlight and shadow detail in your image. And now it leverages machine learning to recognize faces in your shot and intelligently relight them. That means smartphones can automatically fine-tune detail in both the subject and the background. Even some DSLR cameras can’t do that.
Maybe the state of the art is with these phone vendors today. As for example Deep Fusion, this new software feature takes advantage of Apple’s progress in machine learning to allow people to take better photos with the iPhone. Like Google’s camera on the Pixel the feature uses machine learning to better decipher images and produce better-looking shots. it is using machine learning to do “pixel-by-pixel processing of photos, optimizing for texture, details and noise in every part of the photo.”
Today most of DAM vendors are using basic ImageMagick libraries to process their files. Some vendors (Bynder, Cloudinary, Wedia, Celum are doing a great job ) started to used microservices by Microsoft or Amazone to enhance the experience with a nice user experience.
Of course one of the leader is Adobe and his technology Sensei. It is now part in all products from Photoshop too AEM. From retouching to predictive analytics that you can also fin in Lightroom CC Software and CC Cloud.
Video is one more challenge within AI, some vendor like OpenText or Wedia are providing the best video viewer. But you can also imagine smart implementation like the advanced technology provided by VideoMenthe with speech to text, recognition and much more.
The AI tools offered via Eolementhe© unveiled during IBC 2019 are now combined with a conditional workflow feature, for automated sorting and other actions in accordance with criteria set by the user.
For example, if the AI detects images involving nudity in the content, the workflow is interrupted and follows a specific path (notifying the user, providing the option to re-teach the AI about this content, delivering it in separate folder). If not, the workflow continues.
3- AI Metadata is not working !
In the mean time and based on my experience of so many DAM systems, when you are looking at the results of image recognition tools used in DAM systems the context is far away form what customers are waiting for. You have a AI box of metadata but it should be better if vendors could understand this flaw of words to adresse a better solution. Reason why it is so important to capture the value of a digital asset before and after the recognition. It will add the missing context and will enhance the quality of the metadata. One solution is as for example to interface AI tools with the metadata generated by the assets, like text analysis or the #metadata of a project created at the early stage without the assets or coming from the ERP. At the other hand all the metadata used by workflows, asset usage or any actions by the users can also add value to the metadata.
Even if image recognition provides a source of metadata, for the end users are searching for relevant content in context. AI needs to get a smart implementation by the vendor in order to deliver the right content and not only playing with the APIs.
Please spend the time with your librarians to work your vocabulary, taxonomies, facets the successful metadata that you can customise with your business as never AI will be able to do.
A good example of implementation of #AI is the ECM Nuxeo. trying to do smart business, from generic AI to Contextual AI. And in the mean time AI can finally help for #GDRP compliance as well described by Uri Kogan in this post. Smart Business.
As a conclusion AI and ML are one new great source of #metadata which can bring a lot of benefits for the customer experience as soon as the implementation is smart done by the vendors. It’s important to keep your AI-generated metadata separate from your human-generated metadata.
DAM PIM CMS Consultant