Project Cortex – Knowledge Management powered by Search, Graph, and AI

At Microsoft Ignite 2019, Project Cortex was announced. It’s Microsoft’s leap into better enabling customers to use information by leveraging their substantial expertise in search, social network analysis, and artificial intelligence. While Microsoft bills Project Cortex as a knowledge management platform, I stop short of this. I believe that, while Project Cortex has the power to reduce the friction to getting to knowledge, I’m not sure what they’re accomplishing is really knowledge management.

Why You Should Care

Before explaining the pieces that drive Project Cortex, it’s important to understand the interactions that Microsoft is already speaking of. First, there’s the idea of topics. The system determines that there’s a topic in the organization, and it assembles a page that collects what it knows about the topic, including descriptions, related topics and resources, and the people who seem to know the most about the topic. This topic page is something that you can curate and revise if the system doesn’t get it exactly right.

Having topics is interesting – but it’s not where the power is. The power is that, when you’re reading anywhere – in email, in a document, or on the web – you see topics get underlined in your messages. If you hover over that topic, you’ll see a topic card pop up with a short summary of the topic. If you click on the topic card, you’re taken to the topic page for the topic.

This substantially lowers the friction for someone to learn about the various acronyms and topics in the organization. If you’ve ever read on a Kindle and didn’t know what a word is, so you tapped it and got a definition, you know how much this frictionless approach leads to a better reading experience. The author doesn’t need to write information about a topic if the person doesn’t know about it. They can just trust that the system will flag the topic, and the reader can research if they need to.

How Does It Work?

Microsoft hasn’t explained the details yet, but let’s look at some of their investments and infer some of what is going on behind the scenes. The power seems to be coming from search, social network graphs, and artificial intelligence.

Search

Microsoft’s been making significant investments to take what they’ve learned with Bing and SharePoint’s search engine. SharePoint’s search engine itself is largely an enhanced version of the search engine from their FAST acquisition several years ago. The short version is that Microsoft has a deep expertise in search across several platforms. Microsoft search unifies the experience across the board. It brings searching one set of information to every platform and application from a corporate intranet, to Bing and to the Office applications. It is one set of search results displayed in the context you run search from.

On the other end, Microsoft is enabling more search connectors, so that you can get information from all sorts of new sources. SharePoint Search has had the capacity to index Exchange and file shares for some time. New indexing support was announced for Salesforce, Box, MediaWiki, ServiceNow, and other providers. This means that search has an even greater capacity to look across applications to present users with a single search view no matter where the information is stored – in other words, the enterprise search that we’ve been looking for.

Social Network Graphs

The next component that helps to make Project Cortex possible is the work that has grown from the Yammer acquisition. Microsoft got a community building platform, but it also acquired nascent technology for building social network graphs over lightweight signals, which eventually found its way into Microsoft Graph and was surfaced via Delve. Social network graphs are a representation of the relationships that connect users to one another.

Where LinkedIn encourages you to explicitly identify people in your network in an active and intentional way, Microsoft Graph looks at the actions you’re already taking, and it infers who you’re working with. It looks at the files you open or modify, the meetings you attend, and about a dozen other things. These signals are converted into edges – or relationships – between people. The beauty of this is that it happens completely transparently.

Users just do their work, and the system watches what they do to see whom they’re working with and therefore have a relationship with. When you turn this model loose to content and not just people, you get an interesting opportunity to identify relationships between not just people but content as well.

Artificial Intelligence

Artificial intelligence (AI) is a powerful thing. When most people are talking about AI, most of the time they mean machine learning (ML). That simplifies down into an intelligent implementation of Bayes theorem, which is an application of statistics. In short, most AI is about predicting what is and isn’t something based on continuous learning and correction.

In addition to ML, there is natural language processing that attempts to extract meaning from our written language. You see this in action as Word and PowerPoint try to help you correct your grammar. The system begins to recognize the way sentences should be structured and coaches you when you don’t get it right.

In the context of Project Cortex, AI is necessary to determine what should be topics. In a social network graph, you know that people are the objects. In content, you don’t know which things are important and which aren’t. If you can identify what the likely topics are, then you can start to evaluate their connections and start to build a graph of the topics and how they relate.

Knowledge Management or Not?

Historically, knowledge management has been focused on two things. First, capturing knowledge before it leaves the organization for good. Second, enabling people to connect with one another to share knowledge. That’s built on the understanding that there are three types of knowledge (or two types with one having two sub-types) someone can have. Explicit knowledge can be written down – or has already been written down.

Implicit knowledge is knowledge that can’t be easily articulated. Sometimes called tacit knowledge, it frequently has two sub-categories. The first is for that knowledge that can be articulated but for which no investment has yet been made to do so. The second is the category of knowledge where it’s not believed to be possible to convey the information no matter how much work is put into trying to convert the knowledge.

For the tacit knowledge that can be converted, consider cooking a dish that’s never had a recipe. It’s possible to write it into a recipe, but it just hasn’t been done yet. Also consider the idea of riding a bike; it’s hard to put into words exactly how to do this. That’s a kind of implicit knowledge that may never be able to be accurately conveyed – or at least is very difficult.

Knowledge managers often look at explicit knowledge as the tip of the iceberg. It’s the explicit knowledge that Project Cortex has access to. Often knowledge managers are trying to use the explicit knowledge as an indicator that there’s a wealth of implicit knowledge beneath the surface. That’s why communities of practice are useful and why retiring executives are encouraged to record videos that are later transcribed. The belief is that if there’s something interesting in what the executive said, someone in the organization should be able to reach out to them and get access to their knowledge – even after they’ve left the organization.

The problem with calling Project Cortex a knowledge management solution is that knowledge management is much more about building communities and enabling people to talk to other people.

If you want to learn more about how the knowledge management industry thinks about things, check out my white paper, The Road Ahead: Knowledge Management and Records Management Converge with Office 365.

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.