RALEIGH — Michael Capps, former CEO of Epic Games, is on a mission to “Put the humanity back in artificial intelligence.” His new startup, Diveplane, is developing “the world’s first understandable A.I. system.”

A.I. systems do remarkable things, such as beating world champions at the two toughest human board games, chess and GO. Yet, some A.I. researchers have been disturbed that they don’t actually understand how the “black box” A.I. systems come up with some of their results. Capps, in an exclusive interview with WRAL Techwire, repeated a remark he made to the Fortune Brainstorm Tech 2018 conference, sharing the stage with former Yahooo@ CEO Marissa Mayer and discussing the impact A.I. will have on humanity.

Diveplane’s whole mission, Capps said, “Is to build A.I. that is debuggable.” That means A.I. that exposes how it does what it does to its programmers and users.

Mike Capps, co-founder, CEO of Diveplane. Diveplane photo.

“With most traditional neural network algorithms, if results are not what you expect, you have no idea how it got there or how to fix it. With debuggable A.I., you watch its behavior as it’s performing to see if it is behaving as you expected.”

Building conventional machine learning systems is like raising children, Capps said. “You teach them right and wrong.” But that’s not an entirely acceptable method  if bias can creep in, even if unintentionally. Bias could affect A.I. uses in parole hearings, facial recognition, and many other potential applications.

The company is hiring

The idea for Diveplane was hatched at the Council for Entrepreneurial Development during a Carolina basketball game, said Capps, who led Epic Games for a decade. The name derives from the parts on a submarine that make it dive and surface. Capps once taught at a Naval post graduate school and Diveplane co-founder Chris Hazard also worked for the Department of Defense. “The name is something people can pronounce and spell and it fits us because of our experience,’ Capps said.

Hazard is chief technology officer of Diveplane and co-founder Mike Resnick is chief R&D officer. Capps said they conceived the company when Hazard outlined research he and Resnick had been working on in their tech consultancy working on defense and intelligence projects.

It’s also metaphorically significant. “A.I. is about searching up and down, high and low,” Capps said.

The company, already up to 18 people and hiring, is backed by “North of $3.5 million” in friends and family backing. It took no institutional money as yet. “We raised $2 million in a week and by week three, had to stop. It was really gratifying. Angel investors, philanthropists and friends see the value of what we’re doing. We think we can do a lot of good in the world and it still has excellent commercial potential.

Capps said the company plans to remain in Raleigh.

Commercial potential: NASCAR, insurance, healthcare

The company is already working with clients in healthcare, which Capps calls “A good field for us with its mix of a regulated environment and liabilities. Having a non-black box A.I. unlocks potential for healthcare.” It could be effective anytime someone has to make a tricky decision, he adds. “Anytime you need a human in the loop. We’re perfect for that. We provide the answer and how it got the answer.”

Another area they’re working on is “The NASCAR space and racing simulations.”  Many people don’t realize how technology oriented NASCAR is, he notes. “They do serious high end precision machining.” Other areas of commercial interest include venture capital decision support, agricultural science discovery, insurance providers, and drone training.

In the miniature A.I. cluster developing in the Research Triangle, opinions differ on whether it is something to be concerned about. Robbie Allen of Infinia ML is not overly worried and has said there is no sign in the field on how to get to the types of human-like A.I.s depicted in Westworld, Humans, or the Terminator movies.

“I don’t know that we’ll see human-like A.I., but I’m insanely worried about intelligent A.I.,” Capps admits. “I care that A.I. could in fact develop faster than humanity developed. How you build something is important. I wouldn’t want it built by the Windows team. You’d see a blue screen (of PC death) every five days. The idea of rushing to build A.I. without a debugger makes me nervous.”

There are, for instance, 200 million facial recognition system cameras in China and recent studies have show how unreliable that technology still is. “Do we want that here?” asks Capps.

“We think responsible A.I. development will not only give us better outcomes, but a faster path to better outcomes. It will make it easier to solve any problem and not worry about what lawyers might say. It will be dominantly better.”

Here’s Capps interacting with Mayer at the Brainstorm conference.