“Talent acquisition” has become the most dominant task in human resources, as we have pivoted away from developing talent from within. Hiring is the focus not only within HR but also in the organization as a whole because it makes “talent management” so obvious.
AI was supposed to help make better hiring decisions, but it might be fair to say that, so far, the biggest change has been that AI has expanded the opportunities for applicants to cheat. Hiring is one of those situations where we expect the seller (in this case, the applicant) to not tell us the whole truth (e.g., “What’s the minimum salary you will accept?”) and to exaggerate their value. Lying about things that are material is not acceptable, but it is widespread, with estimates that half of people lie on their resume, which is remarkably blatant because those lies can be checked.
What constitutes cheating in hiring, though? If someone posts the questions an employer asks in interviews (which is widespread by the way), is it cheating if you see them before the interview? The lines may be fine, but some things are clearly are over the line. Cheating, defined as breaking established rules with respect to assessments of your attributes, is pretty clearly wrong. It’s a set of actions not dissimilar to what we would see in college, with students getting unauthorized help answering questions and plagiarizing written assessments being the most common.
See also: Most HR leaders are asking the wrong AI questions
AI makes cheating in hiring ‘easier’
Aiseptor, a company that examines cheating, published an assessment of the extent of cheating in hiring, and it is a stunner. Attempts to cheat doubled this past year, and over 60% got through undetected. In assessments of technical skills, cheating happens almost half the time. They report that scores on tests that are unproctored, which means not being watched for cheating, were four times higher than those that were proctored. Cheating is greatest in campus recruiting. I’d like to think that graduating students are more tech-savvy and not that they got more practice cheating in college.
AI makes cheating easier because it can come up with most answers immediately and can write virtually anything in a credible manner, including an essay about yourself (!). What facilitates that, however, are remote interviews and screening.
Think of any take-home assessments where we give candidates a case to analyze or a report to write about some problem we have. At a minimum, it will be assisted by an LLM agent. When we moved to remote interviews, we found that the candidate was often getting help to answer the questions from an AI agent on the screen that can listen to the questions and reply in real time on a screen from which the candidate can read. Let’s not forget the worst cases, where employers find that the candidate they hired job had someone else do the virtual interview pretending to be them.
On the college campus, most every class has given up on take-home assignments and term papers, anything that can be done outside of class. We can go back to proctored exams in a classroom, but we can’t do them on computers because students can access AI tools to answer the questions. Even if we use “lockdown” browsers where it is not possible to connect with the internet, it is still possible to get the same information by phone. So, it is blue books once again (also it turns out that many students don’t know how to write cursive)—but even then we have to make sure they are not using their phones.
Back to hiring: I am told by credible sources that all the attempts to catch cheating of the kind described above can be overcome with other tools or preventative measures. So, we likely only catch the least sophisticated (like one of our students who turned in the prompt he had given ChatGPT to write the paper rather than the paper it produced). We could prevent all of this by conducting our interviews and assessments in person in a context where they can be proctored. We don’t do this because we measure hiring by “cost per hire,” and we aren’t factoring in the big costs of making a mistake.
I discovered something last year you might try in hiring. I asked students to write something about their life experience, knowing that even then it is possible to get ChatGPT to do that for them (“Write an essay about a 24-year-old working at XYZ company in the following job illustrating these concepts.”). But I told them they were going to have an oral exam about their paper. We had a video link that asked questions at random about their paper that would not have been in the final product, such as, “What is an additional concept you would have used had there been more space?” The idea is not that this would be a great assessment but that it would discourage them from not writing the paper themselves. I think it worked—at least the oral responses weren’t bad.
I saw today that Caltech is checking student applications with an AI tool called VIVA, which basically does an unstructured oral exam with questions based on the content they submitted. To see how complicated this world is, the vendor that created the tool, Initial View, also coaches potential applicants on how to get their story across in interviews. To see how “all in” we are about AI, some schools use AI tools like these to grade applicants.
Back to hiring again: The best thing we can do to prevent cheating is to do assessments in person. Yes, it’s more expensive, but hiring someone who is not what we thought they were is even more expensive.
Credit: Source link
