Advanced Computing in the Age of AI | Thursday, February 29, 2024

Harqen.AI Helping Healthcare Sector and Grad Schools Screen for the Best Job, College Applicants 

via Shutterstock

Choosing candidates to hire or admit to a competitive university graduate program can often be a subjective exercise. Resumes, recommendations and test scores aside, evaluators often act upon gut instinct.

Given that reality, the evaluation technology company Harqen.AI is rolling out an AI-enabled automation service which the company says makes better decisions about hiring and college admissions than humans.

Focused for now primarily on the healthcare sector, Harqen specializes in AI-based evaluation technologies for corporate hiring and graduate school candidates. The Milwaukee-based company enlisted an oft-published industrial psychologist to evaluate its evaluation technology. Professor Michael Campion of Purdue University’s Krannert School of Management found in his research that the AI-based service often outperformed human recruiters and admission officers.

"It is not a low bar to reproduce human decisions" in applications like hiring, Campion said. Machine learning models have made it possible to simulate the decision-making processes for corporate hiring and college admissions, including making predictions about outcomes. Campion said greater application of text analysis frameworks such as Harqen's has been critical. In applications like hiring, "The big area that was untapped was narrative data," he added in an interview.

Harqen claims its service closely correlates with human screeners. For example, if human interviewers agree on a candidate 90 percent of the time, Harqen’s evaluation service agrees at the same of higher percentage.

Mark Unak of Harqen.AI

CTO Mark Unak said Harqen’s training data set consists of about 5 million interviews transcribed into text files. When using the evaluation technology, candidates remain anonymous to avoid ethical and AI bias issues. “Candidates are just a number” to the system, Unak said in an interview.

Ensuring anonymity is critical in an industry that is highly regulated, with “a lot of ethical AI scrutiny,” Unak stressed.

Using transcribed candidate interviews, the service performs linguistic analysis on the text files to measure qualification metrics ranging from skills and experience to ephemeral qualities like enthusiasm. The analytics algorithm uses “word clusters” specific to a job description or industry sector. For example, psychological analysis is used to scan essays to look for traits such confidence, enthusiasm and willingness to accept risk.

Along with text analytics, Harqen’s service uses decision tree and other machine learning algorithms to gauge interview responses, educational background, work experience and test scores. The tool automates the labor-intensive process of screening applicants, helping mostly healthcare providers and university graduate programs determine which candidates are most likely to commit, persevere and succeed.

The company said its service can model, or statistically replicate, the human decision-making process using the same data sets as human evaluators. The AI service also considers the decisions or scores generated by recruiters and admissions officers, accounting for large samples of previous candidates.

Unlike traditional tools like linear regression algorithms, the decision tree framework used for classification problems provides evaluators with fine-grained insights. “With decision tree algorithms, we can be very specific” in detecting desired traits, Unak said.

Those tools replace early and controversial screening tools like facial recognition and voice analysis that have largely been abandoned to avoid bias and promote diversity.

Harqen’s machine learning models are used to pre-qualify, then analyze candidates. For now, most of its business is centered on the healthcare sector, filling high-stress, high-turn-over positions like ICU traveling nurses and phlebotomists—the latter position in great demand as the population ages and veins get harder to find on the first jab.

The service can also graphically display job stress that the model characterizes as “neuroticism," or feelings of anxiety, depression and self-doubt, said Unak. For COVID nurses, “you can see it graphically” on the company’s dashboard, he added.

On the academic side, the tool is also being used to screen candidates for graduate physician assistant programs. The AI service scans for candidates displaying grace-under-pressure traits like the ability to analyze data and the willingness to make split-second treatment decisions.

Another metric is determining whether or not a candidate will wash out before completing a PhD program. Those that survive have little trouble finding a job—enabling them to pay off hefty student loans, according to the research.

“Can we predict success for a candidate two years down the road?” Unak asked rhetorically. “Yes, we can.”

The evaluation service also illustrates the lowering of barriers to market entry for entrepreneurial firms like Harqen. AI Pre-trained models are now readily available in the mid-hundred-thousands of dollars, Unak said.

Purdue's Campion said Harqen's service is an example of statistics meeting data science. Using machine learning for hiring "is still a specialized skill [but] AI will become commoditized." For HR specialists, healthcare providers and a growing list of industries and disciplines, "This is a trend you cannot ignore."

--Editor's note: This story has been updated.

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).