NicoElNino - Fotolia

How GE uses a 'Humble AI' approach to manufacturing

GE executive Colin Parris explains why a deliberate approach to deploying AI is needed when dealing with products that cost hundreds of millions of dollars to make.

HALF MOON BAY, Calif. -- Colin Parris has a challenging job. As the vice president of software and analytics research at General Electric, Parris must evaluate new technology and applications that can benefit the manufacturing giant. All of which must work within the framework that GE employs when assessing safety and efficiency known as Humble AI. But even after his own rigorous evaluation and approval process there is no easy way to get buy-in from the rest of the company for new ways of doing business.

With huge investments in aviation systems, energy and healthcare, GE is always looking for ways to use technology to improve operations, deliver products faster and better anticipate problems along the way -- all areas where AI can potentially be of use. But the challenge isn't always technological, it can also be cultural.

"When you talk about using AI on an asset that costs two to three hundred million dollars, engineers don't believe you," Parris said at the recent Techonomy conference here. "You have to verify your findings for months to show what you predicted really works before you get a cultural acceptance."

But Parris and his team have been able to show the value of AI, especially the advantages of digital twin technology. A Digital twin is essentially a virtual representation of a physical product with data linking the two. It is through this link that the virtual version is updated.

In the case of GE, a digital twin of an aircraft engine updates itself every time the plane lands, instantly giving the company crucial information.

"It's an AI that delivers insights," Parris said. "Insights that let's say [are] based upon the way it was flown from which I can predict a part will fail in three months and take an action -- that's an insight. Or I can predict fuel efficiency is falling and that in the next 20 flights you won't make as much money because you'll be paying more for fuel."

The possibilities of this technology have resulted in GE investing significantly in digital twins. Parris said that between 2016 and 2017 the company created 1.2 million digital twins that helped to generate roughly $600 billion in value.

GE knows that the data provided can be crucial in maintaining profits, as well as safety. "It's an AI dealing with trillions of dollars in assets that has to be safe and you have to be sure is used in the right conditions."

Guided by Humble AI

To be sure digital twins are used safely and efficiently, GE has developed a framework it calls Humble AI.

There are three tenets of Humble AI. The first is that the AI knows the zone of competency it is ideally suited to run. Most of today's AI applications can only train on narrow tasks and don't generalize to new scenarios well. "We can figure this out based on the data we've collected," Parris said.

The second tenet is once a digital twin is outside its region of competency, GE will not allow it to run. "We don't use the twin, we go back and we use a legacy algorithm that's been proven for years," Parris said.

The third aspect of Humble AI is to give the digital twin more data as it becomes available so that it can expand its region of competency.

Why you should T.R.U.S.T. AI

Adam Burden, chief software engineer and global lead of intelligent software engineering services at Accenture, agreed it is important that companies have a strategy for deploying AI and a framework that establishes limits as to how it can be used.

"I agree with Colin from GE that there needs to be boundaries and guardrails. You don't want an AI to run on things it was never intended to do," Burden, another speaker at the Techonomy conference, said in an interview.

Burden said that developers and consultants at Accenture are guided by the acronym TRUST.

T for Teachable. "Teachable in the sense that there is a feedback loop that lets the system continuously improve itself," Burden said. "Examples are facial recognition and voice response systems and the ability to learn from intents and mapping."

R is for Responsible, i.e. that the system is designed responsibly.

U is for Understanding. "Being able to understand what the system is doing is a key characteristic," Burden said.

S is for Secure. Ensuring the privacy and security of company and customer data is critical.

T is for Transparency. "There has to be transparency in the AI model," Burden said. "If you can't explain why decisions are being made, we would question the output."

Whether it be for GE, Accenture or any other company working with similar technologies there seems to be a drive for clarity and responsibility when working with AI.

Dig Deeper on AI infrastructure

Business Analytics
Data Management