MTA’s Artificial Intelligence Achieves 100% Accuracy on First Day, Rendering Itself Philosophically Redundant
NEW YORK, NEW YORK — The Metropolitan Transportation Authority unveiled a new artificial intelligence delay prediction system last Tuesday at a press conference held, appropriately, in a subway station that had experienced six service disruptions in the preceding 48 hours. The system, which uses machine learning to analyze historical delay patterns and real-time track conditions to predict service disruptions up to 45 minutes in advance, achieved 100 percent accuracy on its first operational day by predicting that all trains on all lines would experience some form of delay, which they did.
The Technology: Impressive in Its Accuracy
The AI system, developed in partnership with a technology firm whose name the MTA’s press materials spell three different ways across six pages, ingests real-time data from track sensors, signal systems, platform cameras, and the agency’s legacy SCADA infrastructure, which was installed during the Reagan administration and which the technology firm’s engineers describe as “characterful” in a way that suggests they have seen things.
The system’s first day of operation produced 847 delay predictions, of which 847 were confirmed as accurate within the predicted windows. This achievement was celebrated by MTA leadership as “a transformative milestone in rider experience” and by subway riders as “so you made a computer that knows what we already know, congratulations, can it fix the signals at DeKalb?”
The answer, the MTA confirmed, is that the system predicts delays but does not fix them, which is the technological equivalent of installing a very sophisticated thermometer in a house where the furnace is broken.
The DeKalb Avenue Question
The intersection of the B, D, N, Q, R, and W trains at DeKalb Avenue in Brooklyn has been the subject of MTA delay reporting since approximately the invention of the telegraph, and the new AI system’s prediction accuracy at DeKalb is described by system engineers as “extraordinary, in the sense that a very smart system can now tell you, with precision, exactly how late the train you are already waiting for is going to be.”
This is an improvement over the previous system, which told you nothing, but it is not an improvement over the ideal system, which would involve the train arriving. The gap between “knowing how late” and “not being late” remains, as MTA Board Member Claudia Pereira noted at the press conference, “the central challenge of New York City transit, which is not primarily an information problem.”
Rider Reaction: Divided Along Familiar Lines
Morning commuters at Jay Street-MetroTech were shown a demonstration of the app interface that will display AI delay predictions starting next month. Reactions ranged from cautious optimism to the specific brand of New York fatalism that develops after approximately eighteen months of regular subway use.
A man named Terrence, who works in Midtown and commutes from Crown Heights, said he would use the app “if it actually saves me time, which nothing has ever done on the F train, but I’m open to being wrong.” A woman named Patricia, who commutes from Astoria and has been doing so since 1998, said she already has a personal prediction system that is “100 percent accurate and costs zero dollars,” which consists of leaving 35 minutes earlier than necessary for any journey and accepting that this is the deal she has made with the city in exchange for the rest of what it provides.
“If the AI can beat leaving 35 minutes early,” Patricia said, “then we’re in business. If it can’t, it’s just a very expensive clock.”
The MTA’s Ambitious Roadmap
The delay prediction system is Phase 1 of what the MTA calls its Digital Transformation Initiative, a multi-year program that also includes real-time car crowding displays, contactless payment expansion, and a Phase 4 component described in the roadmap as “predictive maintenance optimization,” which means the system will eventually predict not only when trains will be late but why, which subway riders note will add satisfying specificity to information they receive anyway.
Phase 5, which the MTA’s technology roadmap lists without a timeline, is described as “integrated rider experience enhancement,” which appears to mean that the trains will run better, a goal that riders note has been on the roadmap in various forms since 1978 and which they support enthusiastically and await patiently, which is to say they await it the way all New Yorkers await things: loudly, on a platform, slightly too close to the edge.
The MTA Chairman’s Response: Careful
MTA Chairman Janno Lieber, who has managed public expectations about subway performance with the precision of someone who knows the difference between what is achievable and what is promised, said the AI system represents “a meaningful improvement in the rider information experience” rather than a solution to the underlying infrastructure challenges that produce delays. This is an accurate statement. The MTA’s signal infrastructure, large portions of which date to the 1930s, requires capital investment measured in the tens of billions of dollars to fully modernize, a program that is underway on several lines and incomplete on many others. The AI system operates on top of this infrastructure. It sees its conditions clearly. It predicts their outcomes accurately. It does not build new signals, because it is software, and software cannot weld.
For MTA service updates and delay information, see New York Post. For transit policy and city infrastructure reporting, visit The City. For New York commuter news and analysis, see New York Daily News.