Putting Knowledge Graphs to Work: Q&A with Joe Hilger

<< back Page 2 of 2

Since you mentioned AI, how widely used is AI? And how do knowledge graphs relate to AI and support it?
JH: When I talk about AI I also want to talk about machine learning, and put those two together. Knowledge graphs are really based on both concepts and enable it. Machine learning is the theory that the machine or system is getting smarter. What happens with knowledge graphs is we're doing things like auto-tagging information, or spotting trends and patterns, that allow us to make predictive estimations as to what the next piece of information should look like and how it should be categorized and organized. That's a key part of taking these giant datasets and data lakes, and actually taming them—pulling them all together and mapping them. You can't do that manually. You need some level of automation and machine learning. That's active and happening every day.

And AI?
JH: I always struggle with that term AI. I know it is so hot right now. It isn't necessarily artificial intelligence. It's exposing information you wouldn't have thought to get on your own. And where knowledge graphs really excel at that is in a term called inference. And what inference means is because object A has a relationship to object B and object B has a relationship object C, then I know A and C have a relationship. Let me give you a real example: If I'm building bridges in Colombia, and I've got all the data and information about the bridge building in Colombia,  if I'm also going to build a bridge in Venezuela, the computer doesn't see a direct relationship between Venezuela and Colombia. They are two different countries.

In a knowledge graph, they're both given attributes that say that these countries are the same distance from the equator. They have a very similar kind of environment. And, as a result, learnings from Colombia can be, from an environmental standpoint, related to learnings from Venezuela. And that's the kind of leap or inference that a knowledge graph excels at. It looks like AI in some ways, but the reality is it's just doing a better job of jumping across information sets.

'Any technologies that make it easier to find, share, and collaborate remotely are incredibly valuable.'

And then, of course, if you talk to your phone, if you talk to Amazon Alexa, behind the scenes is a knowledge graph. They are very good at dealing with voice instruction and giving you back information. So those are some of the ways in which AI is exposed, but it's not—it's real. Now what I would say is if your vendor can't explain how it works and just says, "Oh, it magically gives you the answer," run away. There is no magic black box. You do need to understand how and why these answers come up.

Another thing that people tend to say about AI, or any kind of automation, is that you really need to keep a human in the loop. Is that something that you've also found?
JH: Absolutely. What you do is you start with human intervention, you identify patterns, use machine learning to start to replicate those patterns, and then periodically check in to see if it makes sense and that there is governance in place that confirms that the patterns that the system thinks it understands are still correct over time. It is not an all-manual process, but again, it's not a black box. It has to have governance and human intervention.

Are there technologies that are related to knowledge graphs that you are finding are useful? Is there anything on the horizon or anything that's been used in the past that are complementary?
JH: Knowledge graphs are more than just the graph database that it exists in. That may be the foundational element, these graph databases that store this, but to do it right, you need that machine learning component. On the ingestion side, when we're pulling information in, having tools that do auto-classification, auto categorization, products like PoolParty by Semantic Web Company, are very important to help the system learn, and capture and catalog information.

And for getting information out?
JH: We do a lot of custom development, but there are more and more tools that are meant for reporting and analytics on top of your knowledge graph so that you can get information out of it. And then as we talk about things like voice interaction, using tools such as Amazon Lex or LUIS [Language Understanding] by Microsoft on their Azure stack—these are products that make it much easier to incorporate understanding of intent, which is one of the components of building a voice-operated chatbot. There are a lot of tools that sit around the edges for this.

In addition to knowledge graphs, what are some of the new information management technologies gaining traction?
JH: Our customers are finding that all-in-one content management systems are too limiting. We need to manage content where it is. We need to be able to better integrate across a suite of solutions. And so things like headless content management systems allow you to get some of the core pieces you need— workflow publishing, editing, and management—that human interaction—and drop that on top of other systems that might be a knowledge graph, or could be an old-school content management system, or even a database or a data lake. We are seeing that some of these more service-based solutions allow you to assemble information together.

Anything else?
And then the other thing, I'm sure this is not a surprise, is that everything is going cloud. If you can afford to go this way, and security and other issues allow it, you can start to assemble pieces together that make up a full enterprise system—and I think that's a game-changer.

In closing, what is your view on the future? It seems there is no end in sight right now for the COVID-19 pandemic. How is that affecting what people are doing?
JH: Well, we're all working remotely. You can't walk down the hall. You're not in the office. You can't turn to the person next to you and say, "What do you know?" So the question is, how do we keep people effective in an environment when they are siloed in their homes? Any technologies that make it easier to find, share, and collaborate remotely are incredibly valuable. I think we're all seeing that it's going to be awhile before we're all back in the office together in any normal fashion. Tools and technologies that make it easier to interact with the information across the enterprise are going to continue to grow in importance.

Interview conducted, edited, and condensed by Joyce Wells.

<< back Page 2 of 2