Building The Cognitive Organization: AI-Ready Information (Part One)

The great promise of automation and Artificial Intelligence (AI) machines is that they not only empower productivity in a single person, but they also empower teams and organizations to operate at an unprecedented speed and scale. However, we are just beginning to explore what this means today and for our future, and the best practices for how to get this done. Most are just now testing the waters with automation and AI (AI solutions), begging the question, “what should I know before I get started?”

Based on our work with clients and recent research, we have developed a methodology and process for becoming AI-Ready across key business areas. Today we will look at a foundational area of AI readiness, AI ready information, which in this case includes data, content, and metadata.

Information First

Regardless of where organizations start with an AI solution, information is the critical foundation. Think of "information" as food for AI: the better the food you feed your AI, the better it can perform. Conversely, junk food in, junk food out. You need information to support automation and to build both types of AI machines: 1) machine learning algorithms and 2) virtual agents. Building machine learning algorithms requires data that is ready for the machine to use. Building virtual agents that work with humans, like chatbots, requires content (i.e., text, images, speech) that machines can understand. Whether you’re using data or content to build your AI solutions, humans need to translate it for machines by describing it, or adding metadata. Once you've added the metadata to the data and content, you’ve created information.

The Attributes of AI-Ready Information

Attribute One: Described Metadata

Metadata helps humans and machines understand how this particular data or content can be used to solve business problems. Data is raw facts presented without context. Metadata provides the context for the data and the meaning for facts. Examples of metadata include the data source; when it was collected; how it's structured; its relationship to other data; how people search or browse for the content; and who, where, when, and how can this content be used.

The image below is an example of how content (an article) is described by a few variables (metadata) that provide meaning, intent, context and understanding of the content to provide higher value.

Attribute Two: Accessible Information

The next step is to make the information accessible to machines. AI can't use the information it cannot see. Even if describing the data and content is done to perfection, the information must be accessible to the AI.

Machine learning algorithms and virtual agents have different information accessibility needs. Machine learning algorithms need lots of information to recognize patterns. This pattern detection can support making predictions about the future. To get the best predictions, you need to have the best information available.

When virtual agents work with humans to get work done, they access sophisticated cognitive services like computer vision, natural language processing, and speech to text through APIs. These services help machines identify objects in the real world and identify words, but not the meaning of the objects or words. Humans need to provide context and meaning to these objects and words so that machines can use this information, along with processing rules, to help us get work done.

Attribute Three: Connected Information

After you’ve created information and made it accessible to machines, the information needs to be connected so it can solve complex problems. For example, personalizing experiences means collecting, describing, and making accessible all we know about past and present interactions with a customer across transactional systems, digital applications, and product and sales systems—a network of connections to a variety of information.

Each AI solution requires slightly different kinds of connections. Machine learning algorithms require abundant and diverse information sets that are accessible in real-time or near real-time. Virtual agents need information at a smaller scale but require sophisticated metadata for machine-recognizable objects and words. They also require rules for identifying intents and performing actions to help humans accomplish tasks.

Connecting information to solve complicated problems has a multiplier effect; the more information sets you can connect, the more capable your AI solution is, the more workflow it can take on, resulting in a better partner for getting work done.

The visual above shows the connected demographic, operational, and competitive information all in one place so a human user can make cause and effect interpretations of the information. Systems like these help humans act upon this connected information and apply their knowledge to create better business outcomes.

Dynamic Adaptation

Information changes as the business changes and the new norm is a constant adaptation. A business adapts when it creates a new data source and retires another data source; the cycle drives the need for new content. We often talk about being "Day Two Ready” with technology solutions or creating solutions that can change with the business needs, instead of a point in time solutions that lose relevancy.

A critical feature for being Day Two ready is enabling humans to manage the changing information that supports AI solutions. Without the ability to manage information, AI solutions will get stuck in the past. For example, we recently created a big data analytics system with multiple bots orchestrating an audit process to move, transform, and store 2MM data elements every 15 seconds. The process that exists today has rules that use metadata to perform 2MM calculations over 80MM individual data elements. However, that process and the information (data and metadata) will inevitably change over time as new data sources come online. You create new metadata and processes are adapted accordingly. A dynamic information management approach, not just the ability to execute a process, will ensure this solution is evergreen.

These three steps – (1) describing information, (2) making it accessible, and (3) connecting the information – are fundamental to AI-readiness. Additionally, embedding information management as a core feature of AI solutions enables them to adapt to changing business needs. While we’ve described plenty of work to do, AI-ready information is just one component of making AI work for you. Stay tuned to our blog series and the next element of AI-readiness.