When people worry about American bridges, they usually picture the worst ones: rusted girders, crumbling concrete, a sign announcing a weight limit. Those bridges exist, and they deserve attention. But the most important number in the latest national assessment of bridge condition is not about the worst bridges at all. It is about the ones in the middle.

The Report Card

Every four years, the American Society of Civil Engineers publishes a Report Card for America’s Infrastructure, grading eighteen categories from aviation to wastewater on a school-style scale. The 2025 edition was, by its own standards, good news. The overall grade rose from C-minus to C, the highest the nation has received since the report card began. Recent federal investment, much of it from the 2021 infrastructure law, was starting to show up in the data.

Bridges received a C, unchanged from 2021. The United States has about 623,000 of them. According to the report, 6.8 percent are in poor condition, down from 7.3 percent a few years earlier. That decline is real progress, and it reflects years of deliberate work by state transportation departments to replace or rehabilitate their most deteriorated structures.

But the same report contained a less comfortable finding. Only 44.1 percent of the nation’s bridges are now rated good. A larger share, 49.1 percent, are rated fair. Fair bridges now outnumber good ones, and the report’s authors flagged that as a trend that will catch up with us.

We are getting better at rescuing bridges at the bottom. We are not keeping enough of them from sliding down.

What the Ratings Mean

The words “good,” “fair” and “poor” are easy to misread, so it helps to know where they come from. Under federal rules, most highway bridges are inspected at least every 24 months. Inspectors rate the major parts of each bridge, typically the deck, the superstructure that carries it, and the substructure of piers and abutments that holds everything up, on a scale from 0 to 9. A bridge’s overall classification is driven by its lowest-rated major component.

A bridge rated “poor” is not necessarily unsafe. It means at least one major component shows significant deterioration, and the bridge will need major rehabilitation or replacement. Unsafe bridges are closed or restricted. A “fair” bridge is structurally sound but showing its age: cracking, some section loss, joints beginning to leak, bearings starting to wear. It is doing its job. It is also on its way somewhere.

Why the Middle Matters

Bridge deterioration is not linear. A new bridge stays in good condition for years, sometimes decades. Then the small problems begin: a deck joint fails and lets salty water drain onto the steel or concrete below; a crack in the deck opens a path for chlorides to reach the reinforcing bar; paint breaks down and rust begins. Each problem, left alone, speeds up the others. A bridge that drifts into fair condition can fall into poor condition much faster than it fell out of good.

The economics follow the same curve. Preservation work on a fair bridge, such as sealing a deck, replacing joints, repainting steel or patching concrete before the damage spreads, is relatively cheap. Rehabilitating a poor bridge costs far more. Replacing one costs more still, and it means detours, closures and years of disruption for the people who depend on it. The ASCE has estimated the backlog of bridge investment need at around $191 billion.

For decades, many agencies funded bridges using what engineers sometimes call a “worst first” approach: fix the bridges in the worst shape, because those are the ones most visibly at risk. It is an understandable instinct, and politically it is easy to defend. But it means money flows to the most expensive problems while the cheaper, preventable ones are left to grow. Federal law now requires states to maintain risk-based transportation asset management plans that look ahead over ten years, partly to push agencies toward preservation. Writing a plan is easier than funding it.

A dollar spent on a fair bridge buys years. A dollar spent on a poor bridge mostly buys time.

What Failure Looks Like

Bridge collapses in the United States are rare, and when they happen, the causes vary. On August 1, 2007, the I-35W bridge over the Mississippi River in Minneapolis fell during the evening rush hour, killing 13 people. Investigators traced the collapse to undersized gusset plates, a design error compounded by added weight over the years, rather than ordinary deterioration. In March 2024, the Francis Scott Key Bridge in Baltimore collapsed after being struck by a container ship, a failure of protection against vessel impact rather than of condition.

The Fern Hollow Bridge in Pittsburgh was different. On the morning of January 28, 2022, the steel frame bridge collapsed into a ravine, sending a bus and several cars down with it. Remarkably, no one was killed, though several people were injured. The bridge had been rated in poor condition for years. Federal investigators later pointed to corrosion and section loss in key structural members, and to a long failure to act on repeated inspection findings.

Fern Hollow is the uncomfortable lesson in this story. The deterioration was documented. The inspections were done. What failed was the step between knowing and fixing.

The Quiet Work

Very little of what keeps bridges standing makes the news. It is the inspector climbing under a deck in February with a hammer and a flashlight. It is the maintenance crew sealing joints before winter. It is the engineer arguing, in a budget meeting, for repainting a steel bridge that looks fine to everyone else in the room. None of it is dramatic. That is the point.

New tools are starting to help. Drones can reach parts of bridges that once required lane closures and snooper trucks. Sensors can monitor strain, vibration and corrosion continuously. Data from decades of inspections can be used to forecast which bridges are likely to slip from fair to poor next, so that money reaches them first. These technologies do not replace inspection or judgment, but they can make both more targeted.

The Real Question

The national bridge story of the past decade is a genuine success, and the people who produced it deserve credit. The share of bridges rated poor has been falling steadily.

But a bridge inventory is not a snapshot. It is a moving population, aging every year. The bridges built in the great highway expansion of the 1950s, ‘60s and ‘70s are now old enough to need serious attention, and many of them sit in that large middle group. Whether today’s fair bridges become tomorrow’s poor ones depends on decisions being made now, in budgets that rarely make headlines.

The worst bridges are the ones we see. The ones that will decide the next report card are the ones that still look fine.

❦

A 2025 survey by the Digital Education Council found that 86 percent of students worldwide use AI in their studies, with more than half using it weekly. A separate UK study put the figure at 92 percent, up from 66 percent a year earlier. In the United States, Inside Higher Ed reported in mid-2025 that 65 percent of students reach for a chatbot every week. Another survey found that 88 percent of students had used AI on a test or assignment in 2025, compared with just over half a year before. The exact number depends on who is counting, but the trend line is vertical.

The technology has moved faster than the institutions trying to absorb it. As of spring 2025, only 28 percent of American colleges and universities had a formal policy on student use of AI. Another third said they were still developing one. Nearly a third of students said they did not know when they were allowed to use the tools in class. Some professors prohibit them, some require them, and some say nothing at all, leaving students to read the room.

For supporters, the picture is one of overdue modernization. AI tutors do not get tired, do not run out of office hours, and can rephrase a concept until a struggling student finally hears it. Adaptive platforms can spot a student falling behind weeks before a midterm reveals the problem. Universities including Arizona State, the University of Texas at Austin, Columbia, and the entire California State University system have struck deals with OpenAI to put ChatGPT in the hands of every student. Three out of four college-aged Americans say they want to use AI in their education and careers, and roughly half of recent graduates report they do not feel they have enough AI skill for the jobs they are applying to. By this reading, a college that refuses to engage with AI is a college failing to prepare its students for the economy they are graduating into.

For skeptics, the picture is darker. The same surveys that show booming adoption also show that 60 percent of college students believe AI is diminishing the value of a degree. More than half describe using it on assignments as a form of cheating, even as they keep using it. Researchers studying students who lean heavily on the tools without guidance report weaker writing skills, less original thinking, and a growing dependence that resembles the way calculators once eroded mental arithmetic. The American Association of University Professors has warned that AI is being absorbed into teaching and research with minimal oversight. The worry is not only that students will cheat. It is that students who never wrestled with a difficult paragraph, who never sat in front of a blank page and worked their way out, will leave college without the kind of mind a college education was supposed to build.

A strange middle ground has emerged in the meantime. Even students who do not use AI describe a kind of unease, what some have called a constant surveillance over their writing. They edit their drafts to sound less polished, less coherent, more identifiably human, because they know their professors are watching for the opposite. Faculty, in turn, hesitate to accuse anyone, because detection software is unreliable and the cost of a false accusation is high. Trust on both sides of the seminar table is thinner than it used to be.

What the technology actually does to learning may depend less on the technology and more on how it is used. AI can help a student think harder, or it can spare them from thinking at all. The evidence suggests both are happening, often inside the same classroom, sometimes inside the same student. The institutions that have responded most thoughtfully have not banned the tools and have not embraced them uncritically. They have tried to teach students to argue with AI, to fact-check it, to treat it as a sparring partner rather than a ghostwriter. That is harder than writing a policy.

The deeper question the technology forces is one universities have spent a long time not asking. What is a college education actually meant to produce? If the answer is a credential, AI is mostly a shortcut, and the credential will eventually mean less. If the answer is a mind capable of doing the work the credential implies, then the question of how, and how much, to lean on the machine is the most important one being argued on American campuses right now.

Most campuses are not yet ready to argue it. The students, in the meantime, have already decided.