Say a company adopts a big-data framework and puts all of its information in a data lake. Then it has it’s resident data scientists go in and carefully cleanse, wrangle and label the data. Think this company is doing data science, the kind that leads to artificial intelligence? Think again.

“None of that is data science,” said Paul Zikopoulos (pictured), vice president of big data cognitive systems at IBM Corp. Unfortunately, that type of grunt work consumes about 80 percent of a data scientist’s workday, he added. That’s why companies serious about AI must find ways to lower that percentage.

An acceptable amount of time spent preparing data is about 20 percent, according to Zikopoulos. A data scientist should spend most of their time on stage four and five, “exploring the hyper-parameter space” for fine tuning and inferencing, he added. This is AI prime time — where models actually make decisions like humans.

Another key to making AI an operational reality is democratizing it for the whole enterprise. They can reach these goals with help from the Watson Machine Learning Accelerator, according to Zikopoulos, who spoke with Dave Vellante (@dvellante) and Lisa Martin (@LisaMartinTV), co-hosts of theCUBE, SiliconANGLE Media’s mobile livestreaming studio, during the IBM Think event in San Francisco, California. They discussed how the WML Accelerator can make everyday AI a reality for both pros and non-pros. (* Disclosure below.)

No data scientist was pestered in the making of this model

WML Accelerator combines advanced hardware — including Nvidia Corp. graphics processing units and software. It leverages IBM’s SnapML machine learning library running on IBM Power Systems servers. The synergy allows users to train ML models quickly with no data-science expertise.

For example, users could train a model to analyze video with minimal coding. They can click their way to an accurate model with  “Auto Label” or “Augment Data” actions, for example. 

Zikopoulos demonstrated this with a video-analyzing model he helped build. With no data scientist and a single developer who wrote a couple of lines of code in a half hour, they achieved 96-percent accuracy. A line-of-business person could expose it with a restful application program interface with one click and pass it into workflow, according to Zikopoulos.

“I’ll pull the data scientists in when I need to [for hyper-parameter tuning]. But 96-percent accuracy without a data scientist present is pretty good,” he concluded. 

Watch the complete video interview below, and be sure to check out more of SiliconANGLE’s and theCUBE’s coverage of the IBM Think event. (* Disclosure: IBM Corp. sponsored this segment of theCUBE. Neither IBM nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE