GE thinks artificial intelligence could use a dose of humility where algorithms will have enough awareness to know what they don’t know.
GE Research said it has been piloting “humble AI” with its digital twin deployments for wind farms and gas turbines. The plan is to build humble AI into GE’s digital twin products and services.
Humble AI refers to the ability to have the capaciity to default to a known safe operation mode when there’s a situation the algorithms don’t recognize.
While operating in a safe mode, humble AI can then learn new situations via gathering more operational data, input from human engineers, correlation with other assets and running additional simulations and scenarios.
Colin Parris, vice president of software research for GE Global Research, said in an interview that the company’s humble AI approach is being piloted in wind farms. “We wanted to introduce the notion of business risk to algorithms and give you early warnings and predictions of model failures,” said Parris. “What business risk am I introducing to these assets.”
Parris also outlined the humble AI concept in a LinkedIn post.
The primary idea behind AI is that models will know their competency and areas where they have a high degree of confidence, explained Parris. If the model sees something outside of its confidence zone, say a temperature range, the model would be humble enough to know it’s not confident.
“AI should be humble to know that it is outside of its region of competency and hand off to standard operation,” explained Parris.
Ultimately, the promise of humble AI is that it can offer speed to value quicker with less business risk, said Parris.