If they’re watching a sporting event such as the PGA Championship, the summer afternoon isn’t totally restful for chief data officers. As the players chase the golf ball around the course, the IT pros at home must keep one eye on the leaderboard and one on the advertisements, and anticipate honing their chief data officer skills.
The ad spots often tout new technology. They use quick-cut imagery of futuristic cities and data centers and feature notables ranging from rapper Common to troubadour Bob Dylan. The technology for sale could be cognitive computing, blockchain technology, IoT or other trendy tech. The result is the exec in the C-suite who has a Monday morning question to test chief data officer (CDO) skills to the max.
These days that question is often, “What’s our plan for AI?”
Because AI can encompass almost anything magical, it can be a tough question for the chief data officer (CDO) to field. A look at a reporter’s notebook from last month’s MIT Chief Data Officer and Information Quality Symposium (MIT CDOIQ) in Cambridge, Mass., may provide a clue or two.
Kaizen and AI
At an MIT CDOIQ symposium panel sponsored by data platform vendor AtScale, the topic of BI on the data lake turned to a discussion of the imp called AI. Chris Crotts, group manager for enterprise data at Toyota North America, said business users tend to bring up questions on AI — questions that can test data strategy and chief data officer skills.
“Someone will call and say, ‘I need to do AI tomorrow.’ We look into it and find that what they are doing is reporting,” he said. In these cases, he said he asks the line-of-business user to describe the actual problem they are trying to solve. His teams then show them ways of analyzing the data to find answers.
“Part of going digital is to have data competency,” Crotts said. That means users have to be prepared to successfully employ something like AI. If people aren’t ready to analyze the data, Crotts said, it is not worthwhile to spin up a host of new tools.
So, his enterprise data group endeavors to prepare users to understand “how data consumption works.”
For their part, Crotts said, users become increasingly helpful in digging in and discovering issues in the data, such as the complex data that has begun to populate Toyota’s data lakes.
He said Toyota’s lineage in continuous improvement — the company is regarded as the birthplace of Kaizen, a work culture philosophy that focuses on understanding problems firsthand — infuses his and colleagues’ approaches to realizing the kind of change that AI can bring.
In a separate presentation at the MIT conference, database veteran and MIT professor Michael Stonebraker also touched on the interest AI is garnering these days.
The guiding technical founder behind such database companies as Ingres, Illustra and Vertica, Stonebraker spoke under the auspices of one of his more recent foundlings, Tamr, a maker of advanced data preparation software.
Stonebraker, like others of late, highlighted the issues influencing chief data officer skills that stand between big data and AI-style analytics. These include the difficulty involved in getting varied data ready to ply for AI insights.
Michael Stonebrakeradjunct professor at MIT and Tamr co-founder
“The hot button now is to talk about AI, machine learning and the data scientist,” Stonebraker said. “But if you are saying data scientists are going to save your butt, you are going to have this problem: They get 10 minutes a week for doing the job they were hired for.” Preparing data for the new engines, in short, is the first step toward AI.
On deep learning for the enterprise — the hallmark of what is new in AI today — Stonebraker was not optimistic. There, a lack of data volume, rather than a surplus of data, can become a determining issue.
“Getting training data is always a problem,” he lamented. For traditional business enterprises, as opposed to web juggernauts like Google and Facebook, “deep learning needs way too much training data,” he said.
Deep learning “works fine if you are doing image data, natural language [processing] or machine translation,” Stonebraker said.
It is not an entirely bleak outlook, however. He indicated that Tamr customers are seeing success with “conventional machine learning using random forest techniques at scale.”
The AI landscape
The admonitions of Stonebraker and Crotts suggest CDOs need to know their way around enterprise data. That is true whether the technology is AI or BI.
Sure, a good understanding of one’s data is a useful club to have in the golf bag of chief data officer skills. But things do change; an organization’s data must be seen in new contexts, as technology progresses and big data, AI or whatever comes next makes inroads.
A symposium takeaway: CDOs must focus on the people side of data and analytics, and be doubly sure to understand the nature of their data and how malleable it is for newer AI techniques.