Table of Contents
Every generation gets promised jetpacks and moon colonies, and most of those forecasts quietly expire. A handful, though, were so specific and so ambitious that when they finally arrived it felt less like progress and more like someone had been reading from a script written decades earlier.
These seven are the best of that bunch. Each one started life as a sketch, a lab demo, or a pavilion exhibit. Each one now sits in your pocket, your driveway, or your browser history.
The Picturephone at the 1964 World’s Fair
Visitors to the New York World’s Fair queued for up to an hour to make a three-minute video call to Disneyland. AT&T’s Picturephone Mod I looked like a prop from a spaceship: a squat unit with a small screen, a camera, and a speaker grille. The company sank an estimated $500 million into the project.
The technology worked. The business model did not. A 1970 trial in Pittsburgh charged $16 a month plus per-minute fees at a time when a normal phone bill ran a few dollars. Most subscribers dropped it within months.
Sixty years on, the same idea runs on networks Bell Labs could not have imagined, and it is free and slightly boring. Zoom went from roughly 10 million daily meeting participants in December 2019 to more than 300 million by April 2020. Video calling has joined the ranks of everyday technologies that were once pure science fiction, which is the usual fate of a prediction that lands.
A computer in every pocket
Apple’s 1987 “Knowledge Navigator” concept film showed a folding tablet with a conversational assistant, video conferencing, and clickable documents. It was never a product. It was a sketch of how computing should feel in roughly 20 years, and it landed close enough to be unsettling.
Six years later AT&T ran its “You Will” ads. In each 30-second spot the narrator listed things you would eventually do: watch a movie on demand, get directions from a dashboard map, pay a road toll without stopping, take a class from a professor on the other side of the country. In 2013 the company re-cut the footage with one change. “You will” became “and you did.”
More than 4.5 billion people now carry a smartphone. The networks underneath them keep shifting too, and the move to 5G has changed how mobile apps are built and what they can do once latency stops being the bottleneck.
The car that drives itself
General Motors’ Futurama pavilion at the 1939 World’s Fair seated 500 people at a time for an 18-minute ride over a 35,738-square-foot model of 1960s America. The designer, Norman Bel Geddes, imagined automated highways where vehicles held their spacing using radio signals and magnetic strips. Radio-controlled cars had already been demonstrated in the 1920s, so the idea was not as wild as it sounded.
What took so long was the software. The 2004 DARPA Grand Challenge sent 15 teams into the Mojave desert over a 142-mile course and none of them finished; Carnegie Mellon’s entry managed about seven miles before veering off. Fifteen months later Stanford’s Stanley completed 132 miles in just under seven hours. Waymo now runs roughly 150,000 paid rides a week across Phoenix, San Francisco, Los Angeles and Austin.
The machine that beats the world champion
In 1968 the AI pioneer John McCarthy bet the chess player David Levy that within a decade a computer would beat him over the board. Levy won that bet in 1978 and collected the money.
He just lost the argument slightly later than expected. In May 1997, IBM’s Deep Blue beat Garry Kasparov 3.5 games to 2.5, and the deciding game lasted only 19 moves. Nineteen years after that, Google DeepMind’s AlphaGo beat Lee Sedol 4-1 in Seoul. The defining moment was a move nobody in the commentary booth understood: move 37 in game two, a shoulder hit on the fifth line that most professionals rated as a mistake until they saw how it worked.
The capability that fulfilled the prediction is also the one now causing headaches, with AI-driven cyberattacks forcing a rethink of how secure code gets written in the first place.
A voice that answers back
HAL 9000 held a conversation, read lips, and played a decent game of chess in Stanley Kubrick’s 1968 film. At the time, Bell Labs’ Audrey machine could recognise a single spoken digit from a single speaker, and only after training.
Progress was slow enough to be embarrassing for decades. Then it wasn’t. Siri shipped in 2011, Alexa in 2014, and the underlying accuracy crossed a threshold around the same time: on the standard Switchboard benchmark, word error rates fell from roughly 43% in 1995 to about 5.1% in 2017, which is roughly human parity for transcribing conversational speech. That single number explains why voice assistants suddenly stopped feeling like a party trick.
Working from the kitchen table
Jack Nilles coined the word “telecommuting” in 1973 while at the University of Southern California, and he spent the decade arguing that a substantial slice of white-collar work could move home, cutting fuel use and office costs. Roughly 47 years later the experiment finally ran at scale, whether employers wanted it or not.
About 5% of full US workdays were remote before the pandemic. By May 2020 the figure passed 60%, and it has since settled near 28% according to the WFH Research team’s ongoing survey. That is roughly six times the pre-2020 baseline, which suggests Nilles was not wrong so much as early by two generations.
A wallet with nothing in it
Diners Club launched in 1950 with 200 cardholders and a paper list of restaurants, and the cashless society has been announced roughly every decade since. Kenya got there first in a meaningful way: M-Pesa launched in 2007 and within eight years its transaction value was equal to around 60% of the country’s GDP.
Elsewhere the shift has been slower and quieter. Apple Pay arrived in 2014, contactless cards went mainstream, and Sweden became the clearest test case. The Riksbank’s own surveys found that only about 8% of Swedes used cash for their most recent purchase. The prediction held. It just arrived without the fanfare, folded into a tap on a terminal.
Why the date is almost always wrong
Scan the list again and a pattern shows up. The idea was right nearly every time, and the timeline was optimistic by 10 to 40 years. The Picturephone was a working product three decades before the network existed to make it cheap. Automated highways were on a 1939 drawing board and are only now a commercial service in a handful of sunbelt cities.
So when someone tells you we will all be wearing AR glasses or buying lab-grown steak, the boring question is not whether. It is which constraint has to break first. For video calling it was bandwidth per dollar. For self-driving it was sensor cost and the long tail of weird edge cases. For voice it was a training dataset and enough compute to use it.
Forecasts that name a date are usually wrong. Forecasts that name a constraint, and describe what happens when it snaps, have a habit of coming true.
