Guillaume: At the heart of Ontotext solutions lies what we call a knowledge graph. Why do you think knowledge graphs are the best way to access knowledge? What makes them so efficient?
Doug: The best parallel I like to make when it comes to knowledge graphs and why they are so good at accessing, storing, and interacting with knowledge is that they work much like a human brain. As you bring new information into your brain, it automatically makes connections and applies the context of everything you already know to classify and organize the new information. And then the reverse happens when you try to retrieve that information – your brain automatically makes those connections.
In the same way, a knowledge graph takes all the data a company has, loads it into a graph database, and applies semantics to make connection points to the new data. In this way, I can access not only the existing data but also connect other data points to it and enable machines to understand how to use it.
Guillaume: Perfect. Now, let’s go a little deeper into this topic. When you talk about the brain, you refer to inference (or reasoning capabilities), is that right?
Doug: Right. Let’s give you an example. You haven’t told me you are French. I’m making that inference, that connection point, because I know the company you work for and I’ve heard the accent before. I’m making those connections in my brain and concluding that most likely you are French.
Machines can’t make such connections because they just understand one thing at a time. But with a knowledge graph and inferencing capabilities, machines can start to understand this data point and that data point and give you information, analytics, or insights that you may not have been aware of before.
Guillaume: It sounds like a game-changing advantage compared to classical databases. Can you share some examples of use cases that illustrate this added value?
Doug: One of the biggest challenges for companies is trying to understand the complex relationships of all the data they have. They have all this data, they have spent all this time building their information up. But it’s still not enough. So, there are many examples I can give. One company we worked with brought all their video, text, and various assets into a platform that enabled users to access information faster and in a more user-friendly environment. This also allowed them to save 30% of their data management costs.
Another example is our Target Discovery solution. It offers researchers a far more speedy, scalable, and efficient process of bringing a drug to market. It allows them to look at all the different information about interactions, medical details, and tests, and run all those test combinations through the knowledge graph. In some cases, they can bring drugs or formularies to market 10 times faster. So, they can make money back on their drug work faster and get valuable medicines into people’s hands much more quickly.
I also want to mention what we call a data fabric approach, where you bring together disparate systems and tie these different silos across companies. We are working with two companies to implement that, where the database becomes their central point of record of information. In this way, they don’t have to move or duplicate data. They don’t have to rip and replace anything and mess with their legacy systems. Everything gets fed in and out of the semantic knowledge graph. Users don’t care where the data is, they just want to ask questions.
Guillaume: So, maybe we can say that dealing with knowledge graphs is a matter of bringing things closer to each other. You have this heterogeneous data that lacks connections and thanks to the graph approach, you are connecting things and it can serve many use cases.
Doug: That’s right.
Guillaume: We talk a lot about AI and particularly generative AI. What is the relationship between AI and knowledge graphs?
Doug: Such a hot topic! I was just at the Gartner Conference not too long ago, and generative AI was on everybody’s minds. Let’s say artificial intelligence is exactly that: artificial intelligence. You have to provide machine-readable knowledge to that artificial intelligence. A knowledge graph is the best way to empower and train generative AI. And let’s emphasize that generative AI is more than just large language models.
There are also things like Graph RAG, which helps break down the information into the different searches you do and serves it in a more reliable, accurate, and traceable format. That’s one of the biggest powers knowledge graphs bring to the generative AI space. Because, if you aren’t careful, you can get wrong answers quickly across the organization. So, having a semantic knowledge graph that provides a trusted corpus of information is critical. It allows you to be more scalable and handle large language models more effectively, ensuring you can trust the answers.
Guillaume: So, you are saying that maybe knowledge graphs are the key to the structure of the training generative AI, large language models, and other tools need.
Doug: That’s a big piece of it. Because otherwise with a large language model, you are looking at a huge amount of data with no real way of knowing if this is the right information. Is it being fed properly? Are you able to eliminate or at least minimize hallucinations? When you try to ask a large language model about itself it doesn’t always give back great answers. Bard does better than Chat GPT, but you have to be able to train the large language model.
By the way, it’s not always large language models, sometimes it’s small language models. Often, a company just wants to get answers within their proprietary data and they don’t want to query public information. So, they also have to train their data using a knowledge graph to ensure that they have the right information pulled together.
And let’s also mention role-based access. If you query data within your organization, you don’t want everyone to be able to find, for example, salary information for top executives. You want to have information accessible but also narrowed down so that only people with proper access rights can see it. For this, the training part is extremely important as well.
Guillaume: I totally agree with you. And I think this is the key to the future of AI and the improved quality we can expect from this AI. Okay, my last question. Have you integrated generative AI tools within Ontotext?
Doug: We have. Actually, we have our Ontotext Knowledge Graph where we have ingested the existing content on our website into our graph database GraphDB. We have over 1,000 blogs and many other categories of marketing content. We are also planning to build a chatbot on top of our website. So, you can make a natural language query about, “What case study can I find in this industry?” Or you can say, “I want to learn more about how Ontotext does natural language processing.” And it will search our proprietary corpus of knowledge. It will bring answers using only our data but running it through the generative AI process. We hope to have that done soon. I’ve seen some initial pieces and it looks pretty cool.
Guillaume: Cool. That sounds great. And I’m sure you plan to integrate this into business use cases solutions. So, thank you, Doug!
Doug: Thank you!
Want to learn more about next-gen graph technology?
5
Share this article
