What the Latest Generation of AI Means for HR

The latest artificial intelligence (AI) tools, called large language models, are causing a stir. Consider the following chat:

Q: Are remote workers less productive?
A: It is difficult to say whether or not remote workers are less productive, as productivity can be affected by many different factors. Some studies have suggested that remote workers may be more productive, while others have found no difference in productivity between remote and in-office workers. It is important to keep in mind that everyone is different and what works for one person may not work for another. It is also important to have a clear understanding of the goals and expectations for a given task, and to provide the necessary support and resources to help remote workers be successful.

That answer is not from an HR manager — it’s from an experimental chat application. For a slightly longer conversation with a large language model, check out my brief interview with Lex.

What HR Leaders Should Do

HR leaders should take an interest in these AI tools because, in time, they will inevitably have an impact on HR and, more importantly, will have an impact on many tasks throughout the organization. A leader doesn’t need to have deep expertise in AI, just a reasonable intuition of what AI can and cannot do.

Developing an intuition for what these tools can do is challenging. The example above might lead you to think these tools have near-human levels of intelligence. Yet, these tools can be astonishingly stupid. There is a famous case of an AI large language model waxing eloquently about why fusilli is useful in heart surgery. It makes these stupid mistakes because, while it’s good with words, it has no real-world understanding of fusilli, or hearts, or surgery.

If you ask how intelligent an AI application is on, say, the scale of an average four-year-old to a certified genius, then you are on the wrong track. The sort of intelligence found in AI is of a fundamentally different nature than that found in humans. It will outperform the genius on some queries, while falling far below the capability of a young child on others.

The best way to think about AI is in terms of its capabilities. The question isn’t whether AI is intelligent, but what AI can do for us.

The Useful Capabilities of Large Language Models

You can think of a large language model as having the capability to read through all the material on the internet and summarize relevant material in a (usually) coherent way. In today’s world, you search for articles on a topic, read through a few that look relevant, then summarize the relevant material yourself. The new AI applications can save time by taking a step out of that manual process. Occasionally, the AI will spout nonsense, but that’s no different from having to weed out poor articles from the ones you find with a manual search.

An example of this process is a tool called Elicit.org. It scans the academic literature in response to your question and summarizes the relevant abstracts. You could do the same thing “manually” with Google Scholar, though Elicit.org is faster. It can even give you a list of studies that critique the findings of the one you are focusing on. It’s good to check before you implement a program based on research that has since been found to be flawed.

What to Do Next

Given the obvious power of these new AI tools, you need to keep abreast of them. Perhaps more than anything, you want to have a good feel for their limitations, so that you aren’t fooled by some slick presentation spouting the magical properties of AI.

You don’t need to do a deep dive. Just take a little time here and there to play with the tools with an eye to understanding an AI platform’s capabilities. Do get vendors in to show off their AI tools, and then be sure to push them hard enough that you understand their abilities and limitations.

The capabilities of these new large language models have surprised even AI specialists. No doubt more surprises lie ahead. Be prepared.