Do you remember VCRs? I realize I may have just lost half the readers …
Related: The ‘enormous possibilities’ for ChatGPT, according to Josh Bersin
They were the first way in which you could tape TV shows and watch them later. More importantly, when you watched those shows later, you could blow through the commercials. Advertisers would no longer be willing to pay for ads (we were told), which would undermine the economic model behind commercial television—and we would be doomed to watch only PBS.
That never happened—at least in part because VCRs were so difficult to program and so cumbersome that virtually no one used them for extensive recording, although their invention did create another industry: movie rentals.
That takes us to artificial intelligence—the force that we were told would revolutionize society, business and also human resources. That hasn’t happened so far, and the reason is just like what happened with VCRs.
We in HR already had the ability to build machine learning models, which could produce algorithms useful for most any predictions. Hiring was one of the biggest industries where AI was expected to hit. To develop a hiring algorithm that worked, though, you first had to have lots and lots of data—the attributes of thousands of employees and their performance—that the “machine” (really, software) could use to “learn” (really, identify patterns) from the attributes of past candidates and their job performance. Then, we could look at current candidates and assess: Do they have the attributes that in the past were associated with successful hires? Very few employers are big enough to have that data, and few have the resources to build the model if they had the data. Models built on someone else’s data might not work for your hires, and so far, vendors haven’t sold many employers on taking the plunge.
ChatGPT’s algorithms are already built, making it easier to use
And that takes us to large language models like ChatGPT and its competitors. Here’s what is different about them: It’s machine learning, but the algorithms are already built. Because they are based on written language, they work anywhere we need to use written language, which is a lot of places. Unlike the VCR, using them is super easy. Once you write them a prompt, they predict what words make sense next in sentences based on the enormous canon of the written word. They are already really good at writing poems, essays, business letters, and yes, college essays. One of my Wharton colleagues took his mid-term with ChatGPT and got a B+.
What does this mean in practice? Well for us professors, term papers are dead because these LLM packages are already as good as the average sober student. Essay tests are done as well unless we want to have them written by hand. No, there is no software that can tell reliably what was written by LLMs or by a student. So my classes stopped doing term papers and take-home essays recently.
See also: Loving ChatGPT for HR? Early adopters should move with caution
In business, cover letters, “Why do I want this job?” questions on job applications and, in fact, almost anything written can be done very competently by LLMs today. When these LLM programs are trained on specific sets of domain knowledge—such as every legal case ever written in the U.S.—they will be able to prepare specialized documents like legal briefs. I suspect this is already happening somewhere. They can prepare customized responses to questions from employees about your HR policies. They can prepare performance appraisals about individuals. You name it. It won’t be original but, as Picasso said, all art is theft: Something pretty good and likely very appropriate has been written already for almost every situation.
Most of the work in HR involves other things besides writing documents, of course, so only parts of our jobs can be automated by ChatGPT. But some jobs mainly involve writing. Consider computer programming—what we used to think of as one of the hottest and safest jobs for the future. These LLM programs can look over billions of pieces of computer code and in a second write code that will solve a programmer’s problem or complete a program where they are stuck. At a minimum, programmers will be able to get their work done much faster, and that means we will need fewer of them. We will need fewer paralegals, fewer client reps, fewer HR staff members, fewer of a lot of employees. That will not happen overnight, but it is inevitable.
As long as the change is slow enough, we might say that this is a good thing for productivity, which has been lagging terribly, especially in service and white-collar jobs, where it has barely budged in decades. Here’s the big worry, though: If you’re not writing, you’re not thinking, at least in the same way.
The reason we in academia have students write essays is not to give us something to grade; it is because doing so forces them to organize their thinking, confront the evidence and figure out what they really believe. If ChatGPT writes your proposal, you haven’t thought about it. If you get questions about a performance appraisal report written by software, it will be hard to answer them because you didn’t wrestle with what to say.
That has implications we haven’t considered.