loading . . . GenAI is pushing federal leaders to define the human core of every job
Interview transcript:
Terry Gerton You are the lead author on a new IBM report that talks about not whether, but how AI will change government work. I mean, it’s no secret that federal agencies are undergoing massive upheaval in their workforce, and you throw AI into that and it’s just like a food processor going around and around. Tell us what makes this moment so pivotal for federal workforce planning.
Bill Resh Well, I think we’re in the midst of a peculiar political economy, to say the least, where we see fairly drastic change in how our president and his aligned party is approaching the management of the executive branch. But this is simultaneous to an incredibly fraught time in our economy in which we have this massive kind of technological revolution going on, this this punctuation in the equilibrium of our economy, if you will, of what they’re calling the fourth industrial revolution. And that’s the introduction of these emerging technologies around artificial intelligence and machine learning. And with that, there have been several conjectures about how our federal government and its workforce could change or is changing and will change as a function of the introduction of these technologies. And so we were with the great team that I have of several people — Nisa Gurbuz from University of California, Santa Barbara, Ye Meng, who is here at Georgia State University’s Andrew Young School of Policy Studies as a doctoral student, Brandon De Bruhl at the Rand Corporation, Michael Overton at University of Idaho and Andy Xia, a great data scientist from USC — all together put these talents together so that we could identify, exactly what are these impacts? How will they manifest? But let’s do so in a careful way. Let’s think of how jobs are really a portfolio, each particular job and occupation requires a portfolio of different competencies. And let’s not think of jobs as just some unitary thing, but that it really is this portfolio of difference types of talents that people bring into a given position and how these various competencies might be affected differently with the introduction of these technologies.
Terry Gerton As you went through that research process, what most surprised you? I mean, the federal government is not really good at workforce planning to start with and probably doesn’t have a great handle on what all of its current jobs have to do. But as you looked at the impact of AI, what did you find? And again, what really surprised you?
Bill Resh Well, there were a few things that surprised me. First is when we started talking to people who are chief technological officers, chief information officers, chief AI officers across the federal government — and this was, we started this work a couple of years ago — and so we were talking to folks who really were at the vanguard within the federal government of introducing these technologies, and we found that there were actually quite a lot of innovations that were happening across the federal government in the use of these technologies. And on top of that, we were also fairly surprised — and this is coming from a person who has studied federal workforce for my academic career — but the level of overlap that the federal civil service has with the general labor market in terms of the representation of different occupations. Well, we went into this enterprise with the idea of analyzing this specifically for the federal workforce. But we found that the level of transparency, the level of standardized documentation that OPM and OMB have produced over the years to describe the various positions and occupations that are in the federal government allowed for a very robust case study on how these emerging technologies can affect these occupations generally, not just in the federal government. And so our paper, while it’s focused on the federal government, actually provides a very good case study for occupational labor markets, generally speaking. And so those two things surprised us. But in terms of the results of our analysis, I think it might be more surprising to an external audience than it was to us. But to the external audience, we might be hearing conjectures of 45% of the labor market is going to be replaced or substituted by AI in the near future. Well, we were a lot more conservative with our estimates going into this because of the way we were measuring it. We were measuring it, again, by competency-by-competency unit of analysis, not by the whole job itself. And when you unpack a job by the various competencies and tasks that are involved with that job, you start to realize that, hey, there is, in fact, a lot of complementarity to these technologies, to what people are doing. In fact, there’s augmentation. It will change the very nature of some of these jobs. But there are places for humans in the loop, so to speak. That is, that there are competencies that are very difficult, if not impossible for these technologies to replace. And instead of thinking of the person being replaced, think of aspects of their job being complemented, being more efficient, being potentially replaced. But then other competencies in which we can more ably lean into as individuals and upskill our workforce towards more soft skills, more critical thinking oriented skills, emotional intelligence, et cetera, and really improve the production, the outputs, the outcomes that are associated with this workforce, but not having to replace them wholesale. And so what we found was our outcomes are a lot more conservative than some of the conjecture you might see in general media, in popular media, as to what the impacts of these technologies might have.
Terry Gerton I’m speaking with Bill Resh. He’s the chair and professor in the Andrew Young School of Policy Studies at Georgia State University. Well, Bill, you laid out a lot of findings there. Everything from the fact that AI is not likely to fully replace people, to the competencies that individuals have that really set their jobs apart from what AI can do. Let’s talk about one of the last points that you made there about training and upskilling and cross training. What kinds of competencies should agencies prioritize to prepare their workforce for this coming AI integration?
Bill Resh Yeah, so we looked at generative AI specifically, because that’s where capability leaps and adoption are really occurring. We did not look at robotics, many aspects of machine learning in our analysis. However, more and more, generative AI is being integrated into those systems. And so there was some overlap, if you will, in our analysis with those technologies as well. But the idea was that we can provide managers an idea of how every mission team could have, or should have, in terms of training, AI literate staff, that there are ways that we can retrain individuals to lean into those soft skills and keep those humans in the loop, so to speak. But to do so requires careful upfront analysis first on what these technologies, or how these technologies, might change the nature of different occupations. Again, at a very granular point of analysis, going down to the task level, assessing the competencies necessary to perform these jobs, and addressing where some competencies might be strengthened relative to others. So it’s useful when agency leaders can weight occupations according to the mix of staffing they have to see where these technologies might complement, or might change or augment the very nature of some of these occupations, or substitute aspects of the occupation. In the report, we show an example of the U.S. Department of Agriculture, where we show how heavy components for complementarity might be concentrated in particular offices, particular sub-agencies as well. Administrative-heavy components show relatively higher substitutivity, so where your more repetitive administrative tasks are being performed, then perhaps you’re looking at those positions and saying, maybe we don’t need to invest as much in future hires for those positions. And it would be useful for targeting upskilling, targeting different pilot programs, and targeting specific different hires in those agencies going forward, without making large cuts across the workforce generally.
Terry Gerton So if you’re a federal leader right now, or an HR manager, and you are trying to do workforce planning, understanding that you have this generative AI asset or tools, from everything that you’ve learned in this report, where should they start? What’s the most important thing that they need to be considering now as they look to strengthen and stabilize their workforce in an AI world?
Bill Resh Well, they need to realize that some jobs are going to resist, or just be resistant to, by the nature of the jobs, AI substitution. And so the jobs that hinge on contextual judgment, ethical trade-offs, more interpersonal skills, and also, importantly, for government field conditions, whether it’s a mediator, or all the way to a wildlife refuge manager, that AI can help, will still help, in those occupations, in surfacing information, bringing more efficient collection of information to help them make decisions. But it can’t credibly stand in for the embodied expertise or the trust that’s necessary from a human level. But that there are other positions in which these positions are going to change drastically. A lot of it has to do, ironically, with federal IT. And so, for instance, software roles. Software management roles are going to score more highly on relative augmentation. That is, how much the nature of the job itself is going to change. And you’re going to gain a lot of efficiencies in terms of coding, in terms of testing out various software applications and refactoring different cycles, while still a place for that human architect to review and to ensure security and integrity. And so the implications are, invest in these tools and make sure that there are humans in the loop for secure by design guardrails. You’re not allowing secure data to be exposed to these foundational models that are owned by private actors and might have interest in training their own proprietary models on very, let’s say, scarce data. Government-protected data has a lot of value to a proprietary model, such as ChatGPT, Grok, go down the line. Because the better data that these models are exposed to, these models are advantaged as a function of the exposure to that data because that data helps train those models. But it’s also, from an ethical standpoint, compromising privacy considerations, compromising national security considerations, when these models are exposed to that data. And so having the ethical decision-making structures in place and the accountability from a human standpoint in place is going to remain important. And so when approaching these things, it’s important to think about, at a granular level about the, again, portfolio of responsibilities that individuals bring in these positions and not thinking from a whole job, a unitary kind of job approach as to, oh, AI is going to have a huge impact on software developers, so we can just cut software development by 50%. That’s not the point here. The point is how you approach the integration of these technologies on a job-by-job basis without putting the cart before the horse, if you will.The post GenAI is pushing federal leaders to define the human core of every job first appeared on Federal News Network. https://federalnewsnetwork.com/artificial-intelligence/2025/09/genai-is-pushing-federal-leaders-to-define-the-human-core-of-every-job/
