How to stop agentic AI from eroding human agency - University Business
The greatest risk associated with agentic AI—artificial intelligence systems that are designed to autonomously make decisions and act with limited supervision—may not be the disappearance of any single entry-level or mid-management role. The far bigger risk may be the gradual erosion of human decision ownership itself—the decline of human agency. - Expertise is not downloaded. - Judgment is not automated. - Leadership is not produced by a machine that sounds confident. As AI-agent systems increasingly plan our schedules, generate our communications, make recommendations and initiate actions on our behalf, we may begin to see ourselves less as decision makers and more as supervisors of machines. Perhaps more importantly, this challenge reaches well beyond whether the college students we teach and graduate are prepared to get a job in a new AI-intensive world. That has never been our primary obligation. It matters, of course. But it currently does not matter enough. What we do as educators What we really strive to do as professors is help young people become more fully formed individuals at the beginning of their adult lives and hopefully, well beyond. Simply improving entry-level employment options falls short of our responsibilities as educators. Yes, we are charged with helping students build the abilities, judgments, and the prospects needed to succeed as they enter the so-called “real world.” But we are also responsible for laying a foundation that allows them to learn how to think beyond the typical what to think. To be successful in this new AI age, students must keep learning how to learn as technologies change around them. And we all know, technologies will continue to advance and evolve at an escalating pace. Dearth, or death, of expertise? Another question that is getting too little attention: What happens to expertise in the age of agentic AI? Expertise is learned agency. It is the individual capacity to search, assess, distinguish, and make uniquely valuable judgments about the critical issues affecting us all. All too often, the shortcut to a decision formed through agentic AI degrades the long path to expertise. Expertise is not simply having access to information. It is the disciplined acquisition of relevant knowledge, the slow development of deep understanding, and the cultivated capacity to separate what is merely available from what is actually valuable. That is how extraordinary individuals grow into leaders who help shape business customs, culture, laws, and public policy. While agentic AI may appear to push a general understanding of nearly any topic down to a level where almost anybody willing to do a little reading can grasp the basics, that is mere commonality. It is what everyone out there is already saying. It is an amalgamation of data based largely on frequency and pattern, and often not much more. In one sense, it can become a race to the bottom where no one really knows what they are saying, doing or why. Agentic AI may produce a serious debilitating learning effect when it comes to being able to learn, self-reflect, form judgements and lead. It may get users to an answer while quietly reducing their future ability to develop and sustain real expertise. It may create a comparable dependency in problem definition, synthesis, and ownership. Cognitive skill erosion The biggest concern involves the erosion of cognitive capabilities that are developed through repeated engagement with complex decisions and information. - Problem definition myopia: Effective problem-solving begins with defining the right problem. As Albert Einstein famously said: “A problem defined, is a problem half solved.” Agentic AI may reduce opportunities to practice asking difficult questions, identifying root causes, and challenging underlying assumptions critical to this first step in problem solving. - Shallow synthesis: Much of professional expertise emerges from reading broadly, comparing sources, and integrating disparate information into coherent frameworks. Agentic AI can limit the breadth and depth of data and experiences critical to developing cognitive skills. - Fragile knowledge structures: Expertise is not simply the accumulation of facts. It is the development of mental models that connect concepts across domains. When agentive AI becomes the primary repository of knowledge, individuals may be at risk of losing the ability to form conceptual frameworks of their own. - Weaker independent judgment: Judgment develops through repeated exposure to uncertainty. Individuals learn to weigh tradeoffs, evaluate evidence, identify contradictions and develop intuition through experience. Agentive AI threatens to reduce those opportunities. There are other damages as well, including the erosion of developmental skills and relational agency capabilities, each of which is central to real learning. The central concern from all of these is that agentic AI delivers answers while users of these tools risk becoming less capable of generating, evaluating and organizing knowledge independently. When agentic AI increasingly performs these functions, people may become less practiced at making decisions for themselves. Earning faster answers The long-term challenge, therefore, is not whether humans will remain technically capable of performing these activities in an agentic AI world. Many will. The challenge is whether we will continue to practice them often enough, deeply enough, and independently enough to remain worthy stewards of our own decisions. If we remove too much of that burden in the name of efficiency, we may discover too late that we did not simply outsource the boring tasks—we outsourced the apprenticeship of becoming capable adults, professionals, citizens and leaders. Agentic AI may help us reach answers faster. But the future will depend on whether we still know how to earn them.