By Gillian R. Hayes, Vice Provost, UC Irvine

At a recent Marconi Society Executive Institutes convening, we were asked to consider, what would your organization do if it had access to only 5% of its normal bandwidth for two weeks. The exercise was about resilience: what still works when the connection is nearly gone? I left that discussion a bit terrified but committed to making low-bandwidth website alternatives going forward. This challenge only grows when we need that bandwidth not just for communications, but for access to our AI systems. Expertise and judgment must be protected for those moments. We need to ask, now, how long could our organizations function without AI and prepare for those possibilities. 

Organizations all over the world, from non-profits to corporate to government agencies, are putting AI into operations faster than ever. Short-term gains from these choices no doubt provide us benefits, but we can be certain that the bill for the costs to human judgment and capacity will only surface fully years from now.

I have lately been thinking about AI’s impact on critical thinking, expertise, and human judgment. When I get either overly optimistic or worried, I find it useful to consider other places we have seen automation come in and the lessons from those experiences.

Automation in Aviation

I am one of those million-mile frequent-flyer types, and I also used to run a company focused on the supply chain for business aviation. So, I take a lot of comfort in knowing that modern commercial aviation is one of the safest forms of transportation. On a large commercial flight, auto-pilot is doing much of the flying. Highly trained and certified pilots are there to take over when we need them or when automation fails. It wasn’t always this way though, and we could have gone a different path. Understanding aviation’s history with automation offers useful lessons for the age of AI.

As cockpit automation advanced, some predicted a future with fewer pilots, or none at all. Instead, new challenges emerged. Regulators urged airlines to schedule regular manual flying (FAA SAFO 13002, 2013), but research revealed a subtler problem. Pilots retained their hands-on flying skills, but the cognitive skills of tracking the aircraft’s position and anticipating what comes next deteriorated when they disengaged mentally (Casner et al., Human Factors, 2014). A pilot who can still fly the plane but has lost situational awareness is a serious risk.

The same dynamic applies to AI. As we hand off parts of our work to AI tools, the greatest risk comes from losing the habit of thinking it through. Supervising an AI requires staying oriented to the task, engaging deeply even when things seem “good enough,” and retaining the human judgment necessary for success. Even now, as I write this article with help from both human and AI editors, I am struck by the need to attend to every word choice to make sure the points are the ones I want to make, not the ones the statistical model thinks are the most likely. Aviation’s answer was not to abandon automation but to pair it with deliberate practice. Pilots train extensively for emergencies they may never encounter. That redundancy is what makes the system safe. AI calls for the same discipline. We must keep human skills sharp enough to step in when the tools fail or are just unavailable.

Protecting Human Judgment

This is not just an aviation story. In one study, the rate of detecting precancerous growths without assistance among experienced endoscopists fell from 28.4% to 22.4% in the months after their clinics introduced AI assistance (Budzyń et al., Lancet Gastroenterology & Hepatology, 2025). The pattern is familiar: AI assistance makes work faster, easier, and arguably better while we use it, but we still need to perform without it, and the skills required to do so can quietly erode. While productivity gains arrive quickly, the long-term consequences of skill loss arrive more slowly. Notably, the physicians in this study were experienced specialists. As a scientist, I worry about how experts can lose judgment and how the same is likely happening to us. As an educator, I want us all to think deeply about how the next generation builds that expertise in the first place.

In my own research, I study teens and families. The teenagers in our studies will be the engineers and operators of these systems within a decade, and their parents and grandparents are already using them. In a national survey my colleagues and I ran, teens were twice as likely as their caregivers to have heard of AI companion apps (50% vs. 25%), and about one in five caregivers did not know whether or how often their child used common AI tools (Rubin et al., 2025). So, teens are pretty savvy. They are also not naïve about the risks: about two in three worried AI might stop them from solving problems on their own. What this generation practices now will shape the judgment and expertise they bring into university and into the workforce. In my lab, we have spent decades co-designing technologies with families living with learning differences and chronic health conditions. Recently, we have been working on family co-regulation: parents helping their kids to manage their deadlines and tasks alongside kids helping each other and their parents with their own self-regulation.

The families in our studies trust the tools we build in part because they build them with us and in part because of how we build them. This is not true for much of their experiences with AI. In a nationally representative study of more than 3,000 parents, teachers, and teens (Min et al., 2026), more than half of the people shown a tutoring app explicitly staffed by human tutors assumed it used AI anyway, and people who believed an app used AI trusted it less. And in our focus groups, more exposure to AI tended to make people more cautious. Trust in these systems is already low, and if the emails in my inbox after recent proclamations of doom from AI CEOs are any indicator, greater awareness is unlikely to make families trust AI more.

In our co-regulation studies, we help teens and families build their own internal sensors and management techniques to help themselves and each other. The technology, no matter how sophisticated, is secondary to humanity, and its use either fades into the background or disappears over time. Families learn to practice co-regulation in ways we hope help them even without the nudges from our system (Silva et al. 2026). In short, they are building judgment and learning new skills. Use of AI can be the same in education and the workforce. If done well, it can be used to build new skills, to prompt creative thinking, and to grow our capacity.

What Organizations Can Do

Just like in aviation, knowledge experts need regular practice with the automation switched off. Outages happen, and our processes need to keep working when they do. Judgment must be protected for those moments. To do this, we must ask of every AI tool whether people are more capable than they were before using it, even when it is unavailable. Tools that strip capacity from your employees are likely a short-term payoff with long-term problems. It is far better to invest in those tools with a little less short-term benefit if you have a better change of long-term positive effects. And finally, as someone who studies teens and is surrounded by 20-somethings on campus, I recommend pairing your early-career staff with senior experts so each learns from the other. Youth tend to be more open and adaptable to AI approaches, and the senior experts have the judgment, and dare I say wisdom, that you need to pass on and retain in your organization.

The Need for Collaboration Across a Long Horizon

The Marconi Society brings industry, academia, and policymakers to the same table. Universities bring a far horizon to this collaborative work. We are in the business of long-term planning, and our products are research that can take decades to come to fruition and students who take years to grow into the graduates they become. Collaboration between partners may need to move quickly in industry with those of us who can afford to be slower and more deliberate can produce the research and policies the future requires.

I see a positive path forward that is AI-enabled and human-led. Getting there will require sustained effort and serious introspection. That work starts with an honest answer to the question from our recent convening: not about systems, but asked about people. Human judgment is our most valuable resource and its already headed for short supply. If the systems went quiet tomorrow, who would still know what to do, and how do we empower and protect them?