Apr 03, 2017 Artificial Intelligence: Industrializing Data Science
PARTNERS 2016 – Summary of Session 548 – Using Artificial Intelligence to Accelerate Data Science Initiatives.
By: Mike Schmidt, Founder and CTO (Nutonian), Matt Mazzarell Data Scientist (Teradata)
The following information is from 2016 session attendee notes.
As the volume of data has and continues to expand to incomprehensible levels, the problem of how to grab value from runaway train of data gets more complicated. That’s a problem because corporations thrive on data. As Fortune is quoted in the presentation, “Data analysis is the lifeblood of a 21st century corporation.”
While the speed of data entering the system has accelerated, the speed of the data analysis has not kept pace. Great analysis is the result of accuracy plus interpretability. With the number of connected devices set to dramatically increase, producing even more data, that goal becomes even more challenging. Artificial intelligence (AI) might prove to be the answer that provides the power of data across the enterprise.
The Data Problem
At its roots, the data problem has three issues.
First, there is no understanding or interpretation of the data. Everyone in the organization inherently knows there is value in the data, but can’t realize it.
Second, there’s not enough time or people to address the data. Adding more data scientists is problematic. As a skill, data scientists are in great demand. When you can find them, they’re still hindered by existing tools and processes.
Third, data and analysis in its current state does not reach its potential capabilities and impact for the organization. In short, because the number of users is limited, fewer questions get asked and answered, preventing the ability to scale it broadly and utilize it a variety of applications.
Things have to change, and the answer lies in artificial intelligence (AI).
Data Science Must be Industrialized
“Industrializing analytics is not an exploratory project in big data usage,” according to Justin Kershaw, CVP and CIO at Cargill, Inc. “Rather, it is a broadly scoped, strategically grounded effort to continue transforming the way we collect, manage and analyze data—and to use the insights we glean from this analysis to improve our business outcomes.”
Industrializing analytics transitions the practice from a series of individual custom projects to a broad and extendable solution. An AI-powered modeling engine can analyze vast amounts of structured data for the most accurate and interpretable models. The results are accessible to analysts, not just the data scientist. That, in turn, empowers analysts to extract meaning from what was previously chaos. The results can and have been applied across a variety of use cases. Consider these examples:
- Optimize processes to drive higher-quality materials
- Predict when components will fail and why they’re failing
- Maximize natural resource extraction while minimizing waste and side-effects
- Automatically sift through chaotic data sets to identify the most meaningful patterns and relationships
- Determine when a cyber security threat is looming, versus standard “alerts” or performance fluctuations
- Predict failure rates along with application impacts for large, distributed environments
- Accurately estimate volume demand to optimize entire supply chain
- Precisely staff stores based on anticipated daily traffic flows
- Predict the outcome of a clinical drug trial early on to wisely invest R&D dollars
- Predict asset volatility far more accurately and transparently
- Pinpoint distressed loans before they occur by understanding the drivers of individual loan defaults
- Calculate true portfolio risk, based on thousands of contributing variables and complex relationships
Bringing Data Science for the Masses
The application of AI to data science makes the power of the data accessible to analysts throughout the enterprise. They can ask questions and get answers faster, and act upon those results faster. One company found better results in 5 minutes than they were previously able to find in > 30 man-months.
Visit the 2016 PARTNERS archives to view this presentation in its entirety.