Mature businessman or a scientist with gray hair with a robot/artificial intelligence (AI)
PHOTO: Shutterstock

As CTO of an AI-based startup in New York, Jason Wisdom of Aponia Data can attest that the company uses AI in many communication-related tasks, such report-generating. But he would not want to hand over the performance evaluation process entirely to AI. “I tell my clients, and audiences, that AI is good for automating 80% of the work. The remaining 20% needs to be done by a senior, human professional,” he said.

Ditto for performance evaluations, he says. “Today’s performance evaluation is more than just qualitative hunches and impressions. Yes, emails regarding a person can be ingested, and graded on sentiment analysis. But a human who is knowledgeable about the employee as well as the business, is critical to review, and either agree, revise, or decide the results are garbage and start from scratch.”

Should Performance Reviews be Automated by Artificial Intelligence?

AI is scoring higher marks in reading comprehension than humans these days. It has been known to write a novel or two. So why not apply it to preparing and writing a performance evaluation? Let’s face it after all, performance evaluations are widely disliked by both companies and employees for a variety of reasons and AI, so the theory goes, could weed out the bias and subjectivity and be truly objective in rating an employee. Plus, it would take a burdensome task off of a manager’s plate. What’s not to like?

Plenty, as Wisdom indicated. Performance reviews are too subjective, too tense and, frankly, have too much riding on them. He does makes clear, however, that business intelligence methods can greatly assist performance reviews. “Many quantitative performance evaluation metrics can be gathered and reported BI-style. AI can take BI results and make the process interactive/intuitive,” he said. A more realistic scenario, he advises, is a hybrid BI/AI approach.

Wisdom is not alone in his circumspection. Many of the AI experts contacted for this story were leery of AI being in control of this process. Those that liked the idea had their caveats. Here is what we learned when we asked, “could or should AI ever write a performance evaluation?

Related Article: Why the Benefits of Artificial Intelligence Outweigh the Risks

Algorithm Bias is Real

There are a lot of problems with relying on AI for a performance evaluation — chief among them the algorithm bias that is inherent in automated learning, says David Lavenda, VP of Product for harmon.ie. There is always some level of bias because the goals are ultimately defined by the developers, and it is difficult to understand and neutralize those biases. Therefore, the goal that the AI machines are trying to reach may be flawed to begin with,” he says.

Also, he adds, how can the company justify the decisions that an AI systems has made. “There is no real way to explain why the engine arrived at a particular decision. So, when an AI system reaches conclusions about hiring someone, a performance review or a promotion, it’s impossible to explain why that decision is a good one. Is that what we want from our HR systems? Is that fair?” asked Lavenda.

Improved Controls on the Data

There is a great opportunity for assisting employees to continuously improve through the use of AI, says Andrew Gilman, Immuta Chief Customer Officer. “AI based reviews can provide meaningful guidance directly to team-members on a more frequent basis than any individual manager has the capacity to complete.” Gilman believes that AI will get to the point where it will autonomously write, or at least provide a framework for a performance review — the challenge will be, he adds, in making sure that access is properly controlled and that there are policy controls on the sensitive HR data required to train the AI models. There are several questions, according to Gilman, that organizations will need to ask about the data being fed into these models before they can use algorithms to evaluate an employee’s performance.

  • Is the data being used to train the performance review secure?
  • Are the integrations being built between the model and HR records done so in a way that prevent the compromise of PII, confidential information and privacy?
  • Are we removing, masking or redacting any information or training data — social, economic, race, religion, gender, etc. — that may lead to bias?

“If these questions can’t be answered in the affirmative, then the validity of the AI-based review is called into question,” said Gilman, and added that at some point, we will have AI driven performance reviews.

Related Article: 8 Examples of Artificial Intelligence (AI) in the Workplace

Creating a Performance Review Framework

Sayeedul Islam, an Assistant Professor at Farmingdale State College and a human capital consultant for Talent Metrics doesn’t see AI writing performance reviews in the near future — but he does believe that AI would be well suited in parsing together the many different aspects of performance from task management to helping behaviors to social media citations by customers to generate a report. A manager could use this first draft to finalize a performance appraisal. “I don't see humans ever being removed from the performance appraisal process but I can see AI being used as a tool to complement the process,” he said.

Robots Taking Over

Marshall Barnes, R&D Engineer who among other things authored a paper on simulation hypothesis, "A Participatory Universe Does Not Equal a Simulated One and Why We Live in the Former" says that not only will AI be able to write our performance evaluations, but they'll eventually replace the people they're doing the performance review for, “...and it will just be another data report on the operations of other AI doing other tasks.” According to him all writing — reports, journalism, literature, and criticism of all kinds could potentially be done by AI. The only issue is the programming," he said.