AI Tech + Humans = Improved Bottom Line #KMWorld

KMWlogo_Stacked_Session Focus:  Semantic AI — how assisted intelligence improves human productivity. Metadata drives insight.

Session Description:

The AI hype is rapidly exploding into C-suites of organizations around the world and with good reason; the promise is compelling. The convergence of AI, robotic process automation (RPA), machine learning and cognitive platforms allows employees to focus their attention on exception processing and higher-value work while digital labor manages low-value, repetitive tasks. While the debate as to whether digital labor will add or eliminate jobs is ongoing, what’s important in today’s enterprise is how digital and human labor can be integrated to improve efficiency and drive innovation. Using real-world examples, this session covers how machine processing, when guided by human knowledge, curation and control, provides assisted intelligence (AI) to organizations which want to streamline processes, reduce operating costs, and improve ROI.

Speaker:Jeremy Bentley, CEO & Founder, Smartlogic

See Also:Artificial Intelligence Hits the Barrier of Meaning” by Melanie Mitchell.

[These are my notes from the KMWorld 2018 Conference. I’m publishing them as soon as possible after the end of a session, so they may contain the occasional typographical or grammatical error. Please excuse those. To the extent I’ve made any editorial comments, I’ve shown those in brackets.]


  • How to Improve AI’s contribution to business value.  How semantic classification, harmonization, extraction, and enrichment makes AI work and deliver values.
  • What is real about AI?
    • the promise is real
    • the ability to automate higher-order work functions using digital rather than human labor at a fraction of the costs will disrupt global markets
  • What AI are we talking about?
    • Machine learning
      • Deep learning
      • Supervised/unsupervised
    • Natural language processing
    • Expert systems
    • Vision
    • Speech
    • Planning
    • Robotics
  • Myth #1. “You just give the machine the data; it learns and then delivers useful insight.”
    • If you are the machine, how would you look around the room and decide to classify/sort it?
      • a group of humans
      • organized by gender, hair color, handedness, whether people are wearing glasses
    • Whether you are a human or a machine, you need context in order to understand the data you are observing.
    • METADATA helps convey context. This is what machines need to gain and provide insight.
  • Myth #2. “We will  quickly assemble the training set, then pour it into the machine and it will deliver actionable intelligence.”
    • it’s actually not just one set:
      • a training set
      • a validation set = a control to compare the resulting output
      • a testing set = to test the output
    • You need to create this trio for each data set in which you are interested.
    • A one-dimensional classification is binary. It separates data into two groups.
      • if you ask the machine separate the humans into two groups: one group = people wearing black or white shirts, the second group = people wearing all other colors.
      • to do this exercise, the machine must first:
        • exclude everyone who is not wearing a shirt
        • learn the concept of color and how to differentiate among colors
        • learn the concept of shirt and how to deal with blouses and tops
    • It takes a lot ot time, effort, and specialist knowledge to assemble these training sets. The more data dimensions you include in the training set, the more likely you will need to include a mathematician in your team. This is another barrier between the business experts and the machine.
  • Myth #3. “Machines are going to be autonomous decision makers soon.”
    • Today, AI is not mature enough to provide full authonomous decision support
    • It can automate the next level of repetivive work, which is compelling from a cost and efficiency perspective
  • Takeaways.
    • the fewer dimensions, the better the quality of the machine’s output
    • subject matter expertise is paramount
    • humans provide knowledge and context
    • meaning is referenced from ontologies and then encoded in metadata
    • there should be an interplay between human and machine
    • we should play to respective strengths: humans do creative work, machines do repetitive work
    • metadata is how you encode context for the machine
  • AI Business Value Continuum.
    • Automated Intelligence — this is increasingly common
    • Assisted Intelligence — we have begun using assisted intelligence
    • Augmented Intelligence — we should have achieved this in the next 10 years
    • Artificial or Autonomous Intelligence — this won’t happen any time soon

Jeremy Bentley Keynote:The Value of Content Intelligence to Big Data #KMWorld

Jeremy Bently is CEO and Founder of Smartlogic.

[These are my notes from the KMWorld 2012 Conference. Since I’m publishing them as soon as possible after the end of a session, they may contain the occasional typographical or grammatical error. Please excuse those. To the extent I’ve made any editorial comments, I’ve shown those in brackets.]


  • Unstructured Content. Unstructured content is highly varied: it can range from a Twitter feed to a Word document or a scanned image. It can cover a range of subjects — as many subjects as arise in our lives.
  • Information Management, circa 1950. In the 1950s, the focus was on manual tagging, content management, indexing, search and distribution. What is the same today? The scope of infomration and the process (e.g., tagging, indexing, search, etc.) The great crime is the continuing reliance on manual tagging. What is different? The techhnology and variety of information has changed. Further, we’ve moved from “information overload” to what it is currently called: “Big Data.” (Calling it Big Data suggests that we can cope with it, in a way that we couldn’t cope with Information Overload.) Other changes are the velocity of information and the complexity of requirements. The complexity relates to different audiences interested in different aspects of the information we have. It also relates to the different uses of that information.
  • Flow. For the purposes of information management today, Flow = velocity X volume. (This is not entirely accurate according to fluid dynamics, but works for KM.) Being able to harness information in real time (in that flow), gives you a competive advantage and efficiencies. It is the relationships between data that present the opportunities.
  • Content Intelligence Allows You to “Enrich” Your Content. This is now the Holy Grail for organizations. Knowledge our content helps us find opportunities, gives us competitive advantage and helps us stop it from leaking out of the organization. At the heart of content intelligence is labeling = metadata. Another key element of content intelligence is extracting the key information. We also need to classify the content so that we can provide indicators as to what it’s about. Given how much content there is and how many topics are covered by that content, it is impossible to manually tag it all effectively (especially since you don’t know who is looking for content and what they are looking for). Therefore, deriving metadata should occur at the point of use, not at the point of archiving. This is a huge reason why manually tagging can’t work. Content needs to be integrated with the existing collection. Finally, it needs to found so we need tools to help surface the content relevant to the person looking for it.
  • Closing Questions. Can you afford the risks inherent in manual tagging? Can you afford to ignore customer feedback via huge unstructured data flows (e.g., via social media)? Do you have the means to track trends reliably?