Understanding the ChatGPT Revolution in Employee Assessment and Hiring: Changes for I/O Psychology

[@MustafaAkben]

Great talk! I really enjoyed your discussion on crossing the bridges between IT and IO. I agree that IO is a very meticulous and careful field in the way it operates. Your discussion brings up a question for me: Could the rigor and careful work we do be the reason that IO and related fields such as OB are behind in the development and use of Generative AI?

[@richardlanders]

It absolutely contributes. A lot of that also comes from the heightened awareness IOs in the assessment/selection space have of legal risk/consequences. Startups and older tech-centered companies are generally much less concerned about that kind of risk (or more technically, they are more likely to assume it to be acceptable and manageable than an IO generally will).

[@MustafaAkben]

​@@richardlanders That makes much more sense! Great insight, indeed. Thanks again! By the way, I was pleased to hear that you were talking about an interview app. This semester, I designed an interview app for my HR students via OpenAI API and Azure platform. Students are taking interviews with the AI chatbot and receiving some feedback from it. Then, they analyze the script to see how AI formulates the questions, what kind of interview questions it asks, and so on. I hope to turn it into a research project and that we have a chance to talk about it in the future.

[@daviddubin3176]

@richardlanders Thanks for the great video! There seems to be a strong emphasis on learning how to prompt engineers and fine-tune, what is the best place to learn how to do this well within an I/O context?

[@richardlanders]

It depends a lot on your specific application domain. Prompt engineering is the more basic/accessible skill, so I'd start there. There are a lot of generic resources for learning how to prompt engineer in specific contexts - for example, Ethan Mollick made this video on prompt engineering for writing: kzbin.info/www/bejne/nJO1hoWdeLumnbs Fine tuning is a more technical skill, as it involves providing a dataset to the LLM via its API. I would only try that after you get a good handle on prompt engineering.

[@daviddubin3176]

@@richardlanders, thanks. I'll check that resource out! I've found many ways to make my workflows within I/O more efficient and effective with chatGPT, but I'm trying to find easy ways to pass this on to my current and future clients. Do you think this could work if I make specific GPTs with prompt engineering? Have you seen any examples of this?

[@richardlanders]

@@daviddubin3176I think this is why there are so many custom NLP AI products out there right now - a lot of companies are trying to figure out how to do exactly that. Most IO approaches I’ve seen so far are of the “replace what we’ve been doing with automated versions” flavor, like the use cases I describe here, but I’d guess that will start to change in the next year or so. For now, I’d say that if content generation (in whatever form) is a major service you already provide for your clients, reducing costs by speeding up those services is the most obvious starting point.

[@daviddubin3176]

@@richardlanders Good point. In the spirit of your talk - not hand over all control to the machine nor be a Luddite - I think taking an approach to the efficiency of AI mixed with the expertise of a Ph.D. in I/O psychology is the best I can do now. Thanks for the great info. I'll follow you for more great talks and hope to run into you at SIOP Chicago!