Traffic Lights Learning to Think: Lessons for Public Schools

Aug 08, 2026

NOTE TO READERS:

This essay is ostensibly about traffic lights. I found the topic interesting to learn about and intend to mine it as a theoretical analogy in work I’m doing on anthropomorphism, and it can be read theoretically that way, but it can be read as a simple tale of two realities on the road to a destination, one coordinated and planned, one chaotic and dangerous, a case study in how one public system tried to meet a new technology for the benefit of society.

Traffic engineers did not wait for permission to use AI. They pointed cameras at intersections, directed models to count cars, and let the results argue for themselves, imperfectly, unevenly, but seriously with authentic intent to work with, not against, technological development. There was never any worry that AI would replace system engineers. When Google’s models got a timing change wrong, an engineer, the human in the loop, reversed it and moved on. The failure mode was ordinary and correctable.

Public education has taken the opposite posture. Where traffic departments have spent years building toward the capacity for full use AI’s affordances, however slowly and however underfunded, school policy has spent the past three or four years building against AI, banning, shaming, disclaiming, litigating, treating the technology itself as the threat when the real threat is lack of governance and resources.

Both traffic lights and public school systems suffer from the same weaknesses: fragmented governance, chronic underinvestment, and a federal government stepping back from the infrastructure it committed to just a few years ago. But only one is trying to solve its version of the AI problem through knowledge making. The other is avoiding the problem. That difference is the argument this essay is making.

»»»»»»»»

The traffic light has been a profoundly indifferent social expressive device for most of its history. For most of its existence, it didn’t know anyone in a vehicle was even there at the intersection, waiting to pass through to the next block while precious time ticked away. It couldn’t tell an ambulance from a pick-up truck.

How could it? It ran on a timer, after all, cycling through red and green, then red yellow green, ad infinitum. And participation was mandatory with quite strict potential consequences for unruly humans. After it acquired cameras, it could know you had been there, enough knowledge to convict you in a court of law.

At one time traffic signals were human beings, not sparkly lights on poles. Officers stood at intersections with flags, whistles, and, later, hand-cranked signs, directing horse-drawn carriages and then automobiles, using their own good sense and talents.

The first attempt at a machine-human hybrid wasn’t really automation at all, but a gas-lit signal installed in London outside the House of Parliament in 1868 at which a policeman was stationed to turn a crank which rotated the gas lamps and thereby the color of the signal.

There always seems to be a price associated with progress. The device exploded within a month and injured or killed the policeman operating it. While some historical popular accounts state he was killed instantly and others report severe facial disfigurement/injury, the tragedy halted gas-lit traffic signal experiments across Britain for over 50 years.

Imagine if the Wright brothers had crashed and burned.

Almost half a century later in 1914, Cleveland’s two-color system used an audible buzzer to alert drivers before that perilous moment when the light changed between red and green. The American Traffic Signal Company installed James Hoge’s electric signal system, including this buzzer to announce the changing of the colors, at Euclid Avenue and East 105th Street, perhaps the forerunner of the bells between classes in school.

In 1920, William Potts, a Detroit police officer, designed the first three-color four-way traffic signal. New York followed a different path. Starting February, 1920, Fifth Avenue got a string of towers where a traffic officer sat inside a booth and switched red, yellow, and green lights by hand cued by a whistle from officers stationed further down the avenue.

The first really important innovation was the inductive loop, a wire coil embedded in the pavement to detect the mass of a waiting car. During the 1960s and 70s vehicle-actuated signals gave intersections a crude capacity to sense presence, though not flow. By the 1980s, cities across the world began deploying centralized adaptive systems using networks of sensors and early computer models to adjust timing across whole corridors in close to real time.

These systems worked, but they were expensive to install and maintain, and most cities never adopted them, an underinvestment still visible today. Only a small fraction of American traffic signals, generally estimated at four to five percent, operate in any form of adaptive mode. The rest run on the same basic logic as Cleveland’s 1914 box, just with better bulbs.

What has changed developmentally in the last several years is not the ambition but the toolset, one with amazing potentials, which is facing ideological and political obstacles inherent in the manic separation of responsibility for shared projects between federal and state driving Washington DC now.

The toolset is enhanced. Cameras, radar, and thermal imaging can feed live video into processing hardware mounted at the intersection itself, and machine learning models trained to recognize vehicles, cyclists, and pedestrians can do far more than a human police officer could ever hope to do.

A modern smart intersection doesn’t wait for a car to roll on the buried scale and signal to be acknowledged. It observes the street, counts the queue, and decides, cycle by cycle, how long each phase needs to run. If a cross street is empty, the red light on the main road cuts short instead of holding a driver in an absurdly immobile state.

Google’s Project Green Light analyzes anonymized driving data from Maps across an entire city, looking for intersections where small adjustments to signal timing would reduce stop-and-go driving along a whole corridor. It needs no new hardware, no cameras, no buried sensors, just access to the data cities already generate through the phones in every dashboard mount.

Boston and Seattle have both successfully used it to retime dozens of intersections. Seattle’s transportation department had to reverse one AI-Green Light recommendation after it produced no real benefit, a reminder that these systems still function as advisors to human engineers rather than autonomous authorities.

Emergency and transit pre-emption of the light has layered intelligence onto infrastructure that already existed. Ambulances and fire trucks have been using optical and radio-frequency emitters to trigger green lights for decades. What is new is the coherence, the preconditioned coordination, the potential to prepare traffic for an emergency long before the emergency arrives in rearview mirrors.

Cloud-based systems track an emergency vehicle’s GPS position continuously as it moves, adjusting signal timing at every intersection along its projected route before it arrives, building what amounts to a rolling green corridor rather than a single locally controlled light. The gain is measured in minutes, and in an ambulance run, minutes often make the difference.

Human lives are saved.

Pilot programs are making their way into ordinary use. For example, weather-responsive signal timing, a category the Federal Highway Administration has studied for years, links ambient sensors to the signal controller so that a wet or icy road triggers longer yellow intervals and wider clearance gaps, holding one red light constant for a few seconds more, stretching a corresponding yellow light out for a few more seconds..

Vehicle-to-infrastructure communication extends to ordinary drivers, too, most visibly through Audi’s Traffic Light Information system, which has been broadcasting real-time signal data to equipped cars since 2016 and now offers speed recommendations calculated to let a driver catch a green light rather than stop for a red, riding what the company calls the green wave.

And AI vision systems are increasingly being used to track not just vehicles but the people crossing in front of them, extending a walk signal automatically when a pedestrian’s measured speed suggests they will not clear the crosswalk in the standard interval, a capability with obvious value for elderly or disabled pedestrians, the group least served by a fixed clock.

Yet dumb intersections still exist everywhere in America with little conscious, focused, societal support for changing them. We seem to inhabit a period when our government is more interested in destroying infrastructure in Iran than in addressing infrastructural development in this society.

Why? In addition to arrogance toward and disregard for ordinary citizens, this government is organized to satisfy the demands of one person: Donald Trump.

Outfitting a single intersection with high-resolution cameras, edge-computing hardware, and a reliable network connection costs anywhere from twenty thousand dollars for a modest camera-based upgrade to well over a hundred thousand for a full adaptive system with radar and dedicated processing. In contrast, one AGM-114 Hellfire Missile used in a drone or helicopter attack in Iran costs $150.000.

A mid-sized city may have hundreds or thousands of intersections. Nashville’s plan to modernize roughly six hundred signals citywide carries a price tag of a hundred and fifty-eight million dollars. Multiplied across the roughly three hundred thousand signalized intersections in the United States, it’s clear why most municipal budgets fall well short of full modernization, federal grants and vendor financing notwithstanding.

USDOT announced recently that under the Trump administration no new notices of funding opportunity (NOFOs) will be issued for future SMART grant cycles. While existing Stage 1 and Stage 2 grant agreements are being honored, cities that had not yet secured a SMART grant can no longer rely on this dedicated competitive federal pipeline for signal modernization.

Cities must now compete at the state level for formula dollars rather than receiving direct federal funding. State DOTs generally prioritize road resurfacing, bridge repair, and highway capacity expansions over municipal urban signal digitizations unless state-specific grant programs exist (such as Tennessee's state-funded Traffic Signal Modernization Program).

The last obstacle is the least ideological on the surface and the hardest to fix: ownership of the problem. A single arterial road can cross through the jurisdiction of a city traffic department, a county public works office, a state department of transportation, and a regional transit authority, each running its own signal timing plans on its own equipment on its own budget cycle.

Coordinating a green wave across that patchwork is less a technology problem than a structural problem, and no amount of good will solves a dispute between agencies over who pays scarce dollars for the fiber connection between their two halves of the same street.

Perhaps AI might lend a hand in rationalizing and humanizing how we as a society share our resources equitably and wisely.

Isn’t it ironic that traffic lights are becoming smarter and smarter while the humans who need them seem to be becoming dumber and dumber? The traffic light has been learning to see and hear and weigh to take in as well as express signal information to assist human beings driving in heavy traffic at rush hour.

The one space the traffic light has not learned to see? It has not yet learned to see across a jurisdiction line.

Share Learning to Read, Reading to Learn

Learning to Read, Reading to Learn

Discussion about this post

Ready for more?