HR AI is one of the most hyped technologies as companies look for a way to gain an edge in the war for talent. Using it well is another matter entirely.
Over the past year, AI became ubiquitous in the workplace -- in Microsoft Office, Google and other browser-based applications -- and in HR software, said John Sumser, founder of HRExaminer. He discussed his latest report, "The 2020 Index of Intelligent Tools in HR Technology," in a video at the 2019 HR Technology Conference, where he moderated a panel with four HR AI software vendors highlighted in the report.
Vendors are working on embedding machine learning throughout their products, Sumser said.
"You may not have it in every piece of HR software that you've got, but it's coming down the pipe," he said.
Today's HR AI market is like the early days of aviation when there were 800 aircraft companies before the Wright brothers made the first successful flight, he said. It took several more decades to develop manufacturing methods and mass transportation.
One problem with HR AI is that vendors have focused on automating HR processes such as recruiting, which has a 50% failure rate.
"The question is, do you really want to do something that fails 50% of the time faster and cheaper?" Sumser said, adding HR AI vendors and users are hitting those types of questions "really hard."
HR AI technology has a very high failure rate, Sumser said. For example, companies have found chatbot management to be complicated, although Sumser predicts improvements coming in HR chatbot technology.
"Something like 70% of all chatbot implementations get turned off," he said. "While there's a big failure rate now, it's early."
The issue of ethics comes up often in discussions about HR AI.
Vendors aren't doing something unethical or making untrue claims, Sumser said. What's really happening is people are still learning how to think about AI and figure out what works, which is typical of the early evolution of any brand-new technology.
Bias -- another concern about AI -- is inherent in human inventions, and often in the data, Sumser said.
"When you get an output from a machine, you can't ever believe that it's the truth. It's the machine's opinion," he said. "Never let the machine make the decision."