You open Google Maps. You punch in the address. You look at the route.
And you assume the obvious: Google is trying to get you there as quickly as possible.
Well, for six months, that assumption wasn’t always correct.
Google researchers conducted an experiment across 10 U.S. cities in which Maps deliberately altered routing recommendations for some drivers in an attempt to reduce congestion across the larger road network. In certain cases, that meant putting a slower route in front of a driver.
The cost to that individual?
About 30 seconds.
Not exactly a cross-country detour. But here’s the thing: The driver wasn’t told.
Google researchers began with roughly 100 recurring congestion points in each city. On experimental days, the system applied a hidden penalty to certain routes, effectively making them less attractive to the routing algorithm and shifting some drivers elsewhere. On control days, Maps operated normally.
Drivers didn’t know which version they were getting.
Google then measured what happened.
And according to Google’s numbers, it worked.
Los Angeles produced the largest improvement, with speeds on targeted roads increasing 4.56 percent. Atlanta recorded a 3.3 percent improvement after some traffic was shifted from Interstate 85 toward Interstate 285.
The median improvement across targeted roads was about two percent.
Google also says it checked whether the experiment merely picked up the traffic jam and dropped it somewhere else. Across all roads affected by the test, speeds reportedly increased 0.35 percent while trip times fell 0.69 percent.
So, problem solved?
Not quite.
All those measurements come from Google’s data. The researchers who conducted the experiment also wrote the paper, and every author works or previously worked for Google Research. The underlying trip and traffic information isn’t publicly available because Google considers it confidential business information.
Nature Cities peer-reviewed the study, and Google released statistical code connected to the research. But outside researchers cannot independently take the underlying traffic dataset and reproduce every result for themselves.
That’s an important asterisk.
Then there’s the philosophical part of this little traffic experiment.
Google described the approach as “proactively shaping traffic flow for the benefit of society.”
There’s a phrase for you.
Translated into ordinary commuter language: Sometimes the route that’s best for everybody isn’t the route that’s best for you.
And Google wanted to know whether it could exploit that difference.
Navigation apps have traditionally been built around a very simple bargain. Tell us where you’re going, and we’ll find a fast way to get you there.
But when millions of people are simultaneously receiving individualized shortcuts, strange things happen.
A 2019 analysis examining navigation applications such as Google Maps, Apple Maps and Waze described the problem directly: “The apps are typically optimized to keep an individual driver’s travel time as short as possible.”
Great for the individual driver.
Potentially lousy for the neighborhood suddenly receiving hundreds of cars because everybody’s phone discovered the same clever shortcut.
Los Angeles got an especially memorable taste of this problem on Baxter Street, a steep, narrow road that became a navigation-app shortcut. The city eventually converted part of it to one-way traffic after increased vehicle volumes created problems.
So Google’s new experiment flips the old formula around.
Instead of asking only, “How do we get this driver there fastest?” the algorithm can effectively ask, “What happens to everybody else if we send this driver there fastest?”
And sometimes the answer is: Give this person the slightly slower road.
Now, Google’s ambitions don’t stop with the route on your phone.
Project Green Light uses Google traffic information and artificial intelligence to recommend traffic-signal adjustments to cities.
And this is where the machines encounter that wonderful old-fashioned obstacle known as local reality.
Seattle reversed at least one signal change after it produced undesirable results. Engineers in Manchester reportedly rejected many recommendations because Google’s system lacked some of the context affecting those intersections, including bus movements and nearby schools.
Aleksandar Stevanovic, who studies traffic control at the University of Pittsburgh, explained why this is difficult.
“Traffic has so many uncertainties,” he said. “In one hour, you can have five different goals that you want to achieve.”
Exactly.
Maybe you want maximum traffic flow. Maybe you need buses moving efficiently. Maybe children are arriving at school. Maybe a neighborhood street technically can absorb more cars but absolutely shouldn’t become the morning alternative to an interstate.
An algorithm can crunch traffic data all day long. The argument begins when somebody has to decide which objective matters most.
And that brings us back to those drivers in Google’s six-month experiment.
The actual individual effects were remarkably small. According to the paper, the average time saving amounted to roughly 0.25 percent of a trip. Normal day-to-day travel-time variation was vastly larger.
In other words, most people probably never noticed anything.
That’s good news if your concern is inconvenience.
It’s also precisely what makes the experiment so interesting.
