CognitiveScale Product VP on The Maturation of AI Tech (PART 2)

If you look at what we do with AI tech, everything must be described with some kind of mathematics. Clients frequently talk about a specific use case and hope that by throwing in some data we’ll get a magical prediction out of it. Sometimes they don’t even have the data, but AI “can’t create something out of nothing.”

Getting Inferred Attributes 

Much of our focus at CognitiveScale is on financial services, healthcare and media. For those industries, we recommend first on how to collect a whole host of data around end-user profiles. Initially, the data includes a declared type of information, such as name, address and other attributes provided by the user. 

Next, we gather data on observed attributes and insights. AI tech analytics determine if users spend more time on certain insights and topics. Are they clicking on them? Are they taking action?  Most importantly, are they buying? Simply put, all of this information is gold. The trick is harnessing the value in the data. 

From declared and observed attributes, we can then extrapolate inferred attributes. This means we can infer what the user might like even when it’s not explicitly expressed. It gives us a sense of how to apply a recommendation to their profile.

The Effect of AI Tech on Product Lifecycle

The lessons learned from AI’s impact on product lifecycle allowed us to build our augmented intelligence platform, Cortex. The traditional product lifecycle typically builds functions one-by-one and releases them. With AI solutions, we use the DIAL loop (Data, Actions, Insights, Loop) instead. 

DIAL takes into account that AI tech is a living, breathing type of application. It may have some trouble at the start, but it improves as it progresses. Compare this to the traditional product approach where features are stacked on individually. There’s no real expectation of progressive, spontaneous improvement as can be seen in AI.

Bootstrapping AI applications requires deeper insight. For example, you might not have the data to train models for your ideal user experience. Instead, you intuitively lay down an initial experience for end users, collect data and learn. Later, you train AI solutions to lay down the experience you really want. 

 

Click here for Part 1

Click here for Part 3

Comments

Please to leave your comment

Load more comments
About the Speaker
Jon Richter
CognitiveScale VP Product
Jon Richter is the VP of Product at CognitiveScale - leading the product team in developing AI solutions for businesses who are looking to streamline operations with augmented reality solutions. Prior to joining CognitiveScale, Jon spent over a decade at IBM - where we worked as a program director in developing Waston (one of the first AI products at scale). Prior to working on Watson, Jon was the lead software developer for IBM's software services division. Jon holds an MBA from The University of Texas at Austin and currently lives in Austin.

Recent Posts