Knowledge management is experiencing a resurgence because of the growing awareness that not only can it solve current problems, but it can also help organizations prepare for the artificial intelligence (AI) solutions of the future. To find the best path forward, organizations need to understand the connection between knowledge management, knowledge engineering, knowledge architecture, and artificial intelligence.
New holy grail for KM
Fundamentally, knowledge management is about getting the right information to the right person at the right time. We hear this over and over, and it has always been the Holy Grail for knowledge management, but the ways in which it is being done are constantly changing. In the past, for example, knowledge management was about organizing information and using search engines so people could find what they needed more quickly and easily. Now it’s about anticipating what they need and getting it to them in an automated way.
Part of this process involves filtering out the information that is extraneous—content that is not relevant to a particular customer. I was with a client today whose organization is managing 800,000 pieces of content, and they need to be able to get people just the right piece of content. This is in the entertainment industry, so people’s choices are very subjective. They are developing was of using AI to achieve personalization in a context where there are more subtleties than there are in something like grocery shopping, where purchase patterns are easier to detect.
Make things easier for users
The goal is to make it easier for people to do things--use your website, or your application, or your services--by representing an individual’s mental model in your knowledge architecture. The system needs to be able to anticipate something the users need, and it can do that only if information is correctly mapped against the user’s model. This ability is an evolution of technology, which has always been about reducing the cognitive load. Technology should make it easier for people to accomplish their task, achieve their objectives, find a solution or find an answer.
Knowledge management is a way of sharing, enabling collaboration in order to solve problems. It's really about interacting with other humans, and gaining access to enterprise information. Knowledge architecture allows us to support knowledge management, because it reflects the mental model of the users, and knowledge engineering is the way to componentize, ingest and reuse that content. Knowledge engineering involves the mechanisms for deciding how to structure that information so it can be ingested by enterprise systems.
One of the challenges around enterprise knowledge is that many times it's unstructured and uncurated. There's a lot of inefficiency in using that type of information. When we put it into a knowledge architecture, however, we are helping to improve the reuse and reduce the costs of managing and using it. Knowledge is usually distributed across different departments and processes. With a correctly built architecture, aggregating and retrieve knowledge can be accomplished much more easily. It doesn't have to be put in one place. What’s needed is an architecture that will allow users to retrieve that content wherever it may live. Sometimes some harmonization of the different structures is required, and a way to reconcile the different terminology from those diverse applications.
A great deal of enterprise knowledge is also embedded in systems and in enterprise processes. And the expertise of humans--tacit knowledge--is the stuff that people walk around with in their heads. Explicit knowledge is documented in some way, but tacit knowledge is not really accessible until it becomes explicit. With respect to tacit knowledge, people always know more than they can say, and they can say more than they can write down. That's just the way knowledge works. I can answer lots of questions when we have a conversation, but I know more than what I'm telling you.
When we can componentize that knowledge we can actually break it up and drive it into various functions. It can be used for self service, or to solve day-to-day problems. We are making it more findable for humans and more digestible for the systems. So if you think about the types of applications that require question answering, such as a chatbot or intelligent virtual systems, the information you need to train that chatbot is the same information you need to train humans.
Training the systems
When people characterize algorithms as not needing any architecture, that's a bit of a misstatement. Even when the algorithm does not require any type of structure and can just operate on the content, we still have to give it some data. The particular pieces of data that will go into that training set need to be identified in some way. So in other words, the training data is really the knowledge and the content and the curated information that has to go into these algorithms. The content needs to be classified in some way for it to be useable.
If a system is being trained to recognize images of a cat, it must be shown many examples of cats, and examples of animals that are not cats. What is it that defines “cat-ness”? The cats have certain features, and in machine learning parlance, features are actually metadata. You look at cats from different angles, different lighting or conditions. The system can learn to distinguish them from other animals. But at some point you have to call it something, tell the machine what it is, and the classification is “cat.” Then you tell the machine to do off and find more images of cats. That is the process.
Hypothetically, you can train a system to do something like find images that show defects in manufactured parts without any metadata being applied. But then how do you do something with that information? What parts are they? What processes are affected? What departments make them? In order to remediate the defects and use the insights from that machine learning algorithms, the information must be contextualized and the insights apply to the right processes and departments, and it is the knowledge architecture that allows this to happen.
Where AI projects get stuck
One of the organizations we're working with is a large process manufacturing company that has hundreds of thousands of data points and hundreds of thousands of machines, equipment, and tech tools that produce data. They found they were not able to get to the next level of leveraging AI machine learning without having a knowledge architecture. The roadblock they ran into was what to do with the results. Who owns them? How do you change your procedures and feed that information back into the organization? People cannot act on the data without having a knowledge architecture that provides context.
To recap, knowledge management is really about collaboration of the culture, processes, and getting people together to leverage their expertise. Knowledge architecture is the mental model of the user, how they think about the data. What are they trying to do with the information—set up a marketing campaign, develop new products, or improve customer satisfaction? Do they think about the data by client, product, or chronologically? And finally, knowledge engineering provides the mechanisms that guide how we structure information, so that it can be ingested into a system, and ultimately inform the knowledge management process. All of these need to work together to support the rapidly growing applications of artificial intelligence, but in the meantime, they also help improve human performance.
Are your knowledge systems helping your users work more productively? Talk to us about how to get there.