One of the paradoxes of artificial intelligence is that the companies poised to make the most of it are those with the most data: enterprises; yet, they have the most institutional and organizational difficulties to overcome to do so. In this episode of the O’Reilly podcast, I had a chance to discuss the challenges with someone whose job it is to tackle these issues—Ron Bodkin, VP and general manager of artificial intelligence at Teradata.
Here are some highlights from our conversation:
“Connected AI” as a strategy to build on existing data sources
Connected AI is an approach to deliver strategic value from AI by connecting it with high-quality data that's already curated and integrated in the enterprise. This typically starts with high-value transactional data, such as information about past purchases, customers, and accounts, as well as data about devices and configurations. It also includes high-volume data, such as clickstreams or sensor readings. It’s connecting that data with an analytic ecosystem that allows data scientists to use it in a way that's robust and well understood. And finally it means connecting it to applications that allow customers to interact with artificial intelligence in a way that really sits naturally with the business in an AI-first manner.
We see connected AI as a way of connecting artificial intelligence with the data, analytics, and interactions of the enterprise to deliver strategic value, but leveraging past investments.
The organizational changes needed to make the most of AI
The change to AI-first processes is a big change. Many organizations start by thinking about applying AI in a very narrow way to automate something that's being done by a person, but when you think about the changes in roles in a world where computers can make more decisions, there's an opportunity to drive much better outcomes and to leverage the creativity and ideas of people in a more extensive way. An example of that opportunity to better leverage people is, instead of having humans sitting and making training data sets, looking at a lot of images to annotate them, now we have the ability to have humans provide higher-level guidance to AI algorithms that let the algorithms perform more effectively, thus taking better advantage of the intuition of humans.
Teradata’s collaboration with Danske Bank on fraud detection
We worked with Danske Bank starting from data engineering, creating a foundation to organize data from across the enterprise around transactions and accounts, so our models could have the right set of information for detecting fraud in real time. We integrated this data into their mainframe application to be able to call out within milliseconds to a set of model programs that would run algorithms and detect fraud, and return the response of suspected fraud or not. Also, one thing that was key is that we worked to apply recent research to interpret the model—it wasn't enough just to say, ‘Hey, we think there's fraud in this banking transaction,’ but to explain why.
This post is a collaboration between O'Reilly and Teradata. See our statement of editorial independence.