This data scientist wants HR leaders to ask these three questions when shopping around for AI tools

New AI tools are emerging every day. Businesses are actively learning about new technologies and figuring out how to deploy them, and many business leaders are eager to reap the productivity gains AI tools promise.

Cassie Kozyrkov, CEO of Data Scientific, joined HR leaders and executives in Silicon Valley for the Society of Human Resources Management’s (SHRM) AI + HI Project event. She told people pros researchers have found that the AI applications bursting onto the professional landscape are really good at “the copy-paste work, digitized and repetitive.”

“[Digitized and repetitive] means lots of examples [exist], all available electronically. What does [AI] want? Examples from which to draw patterns that get turned into code. So, that is why the impact is hitting there,” she said.

Kozyrkov said that currently a disproportionate amount of the type of work that can be replaced are the tasks commonly performed by women. But they’re not the only group at risk. Last month, Slack’s Workforce Lab found that AI tools are poised to transform 41% of day-to-day work for desk workers, specifically by helping them tackle those types of tasks.

“Who else is doing repetitive, digitized work?” she asked. “The answer is people who are freshly beginning their careers. Why? Because nobody trusts them…You [should] give them work where you know what the right answer is, and then you can check how well they're making judgment calls.”

“Those [early career tasks] are exactly your perfect targets for this kind of automation, which is why we have this phenomenon where we’re seeing that people who are more senior in organizations seem to be cannibalizing the work of the people underneath them,” she said. “But what is our plan for training graduates? What is our plan for creating those senior positions?”

Kozyrkov said that AI development is at the stage where the technology is really good at assisting “human-in-the-loop” work. Humans need to work alongside the technology and ensure its outputs are what was asked and that its outputs are correct.

“Enterprise automation is a completely different game,” she said. “This is where…automatically, no one checks it. At scale, off it goes. [Remember] that these systems do make mistakes—maybe not often, but they do.”

Kozyrkov warned HR professionals and executives procuring the technology to get into the practice of asking questions about what’s under the hood.

“We love to talk about technology as if it’s faceless,” she said. “‘The AI system decided. The AI system did it.’ But that might be one of the most dangerous moves for society. We need to see the person behind the curtain. Every AI system comes from those three big decisions.”

When choosing new AI tools, Kozyrkov suggested HR professionals ask three questions:

1. The “objective” question: What is it for? How is success defined and scored?
2. The “data” question: What data is it using? How was it collected and processed?
3. The “testing” question: How do we know it works? How was this tested?

“Not all experts will agree, but in my opinion, the safety nets of your system are more important than the algorithm’s quality. So, focus a lot on safety nets, then at least you’ll throttle it before it does something bad if the quality is not so good.”