It’s 1950 in Communist China, and you want to type a letter. You lean over a metal tray about the size of a sheet cake. Instead of keys, the tray is filled with tiny pieces of metal type. You manoeuvre a knob on a lever across the tray until it hovers over the piece you want. Press the lever, and with a thwunk the piece leaps up from the tray, strikes an ink pad, then pounds the paper mounted above the appliance. Now begins your hunt for the next piece. You have about 2,500 to choose from.

This was the Chinese typewriter, and it may not surprise you to learn that its top speed was a plodding 25 characters per minute. But that little fact, says Tom Mullaney, a professor of Chinese history at Stanford University, in the United States, led to a remarkably forward-thinking innovation: machines that could provide you with options of what to say next. Today, autocomplete is a global phenomenon – but its unexpected back story may offer some surprising insights into its future.

To live in China in the 1950s was to be deluged with words. The government issued propaganda campaigns about everything from marriage reform to steel production, and charged local typists with making copies for everyone to read. But it was slow going, not least because the 2,450 characters in the trays were organised according to a Qing-dynasty dictionary that grouped them by shape. The equivalent, says Mullaney, would be having “apple” and “aardvark” next to each other on a keyboard of English words.

“Those words will never appear next to each other in a human sentence,” he says.

To make things a bit easier, typists across the provin­ces began to personalise their trays. First you might move the most common propaganda phrases into the middle and next to each other, so you wouldn’t have to hunt around for the words “American” and “imperialist” every time. Then, maybe you’d put “grain” next to “prod­uc­tion”. Someone working in a farming town might type certain phrases more often than someone in an urban office, and vice versa, so typists’ trays grew increasingly diverse. Eventually, they became customised so that surrounding many characters were the eight or so others most likely, in the typist’s experience, to come next.

Chinese Typewriters Exist, and Tom Mullaney Wants to Save Them

By 1956, The People’s Daily was heaping praise on a typist who had more than tripled the previous speed record, simply by rearranging his keyboard. The inno­vation stuck: the Typewriter Museum, in Lausanne, Switzerland, features a Chinese typewriter from nearly 30 years later, still sporting a customised tray.

Computers did not make this early form of predictive text obsolete. Mullaney says that visitors to the People’s Republic in the 1970s reported seeing an early typing program on a series of cathode-ray-tube screens. The moment the user selected a character, an adjacent screen would change, repopulating itself with a new set of characters that the designers had decided were the next most likely. The expectation was already in place that typing – wherever you did it – would involve an easy way to select the characters you want next.

Book review: A Billion Voices asks: is there such a thing as ‘the Chinese language’?

Compare that with the West, where predictive text didn’t emerge until the mid-90s. T9, one of the first such programs, was originally developed to help people with physical impairments type more quickly, explains Cliff Kushler, its co-developer. Push a few keys, and T9 would tell you which word you were most likely to be looking for. Then, the software adapted itself to the user, picking up their preferences for certain words. Today T9 is loaded onto more than four billion phones.

This pales compared with predictive text’s prevalence in China. Today, to write a Chinese character, you type its pronunciation on a QWERTY keyboard and a box appears by your cursor, offering several characters that share that sound. Choose one, and the process begins anew. It’s not just single characters – the software can assemble entire phrases, and even suggest a relevant emoticon. Every­thing from e-mails to love letters to government announce­ments is written this way.

What happens when autocomplete dominates so, well, completely? It turns out that relying entirely on software’s predictions – being shown words, rather than creating them with your own hand – may affect your ability to write. Over the past two decades, a phenomenon called “character amnesia” has invaded Chinese popular culture – akin to putting pen to paper and suddenly realising that you cannot remember how to spell the word “cologne” or “enough”. In a 2010 survey, 83 per cent of respondents said this had happened to them – and many people blamed autocomplete. A Xinhua News headline blared: Is China Facing A Character Crisis?

Imagine what it’s like being dyslexic with thousands of Chinese characters to remember

There is some merit in the idea that visually recogni­sing a character is neurologically different from writing it. There is also evidence that something analogous is happening in the West as handwriting cedes to typing: children who learn to write by typing do not have the same understanding of character or letter forms as those who learned to write by hand. At Beijing Normal University, one professor now advises that elementary school students be banned from completing their homework electronically, and that university students complete half of their assignments by hand. He also suggests sponsored handwriting contests for students and government officials.

While that could address character amnesia, some see a deeper problem. As the early Chinese typists thwunking away on propaganda no doubt knew, predictive text lends itself to the formulaic. Presented with pre-digested options, will we grow less creative?

“The danger of getting very formulaic in how people express themselves is a real one,” says Kushler. Evan Selinger, a professor of philosophy at Rochester Institute of Technology, in New York, who writes about the effects of predictive text, notes that both Apple’s QuickType and Google’s Allo messaging service offer pre-packaged responses. Presented with a photo of a friend skydiving, Allo might offer you the choice of the following replies: “How exciting!”; “So brave!”; “How fun!”

Although this software can study your habits, letting the program choose a comment that sounds like you, it’s still you at your most generic. And while generic might appease the immediate needs of friendship, it’s not particularly satisfying. Studies on outsourcing suggest that putting less effort into a behaviour makes you feel less responsible for its effects, both good and bad. This kind of communication actually creates distance. And Selinger points out that prediction eliminates the possibility for whimsy that often makes human interaction rewarding: it mistakes the predictable for the inevitable.

Creating a predictive future with big data

That said, perhaps the problem is self-limiting. Being constantly presented with options – especially when they are not quite right – can be maddening. For typists in 1950s China, predictive text was a boon, a way to make typing far more efficient, and today it is built into how people express themselves on the screen or the page in Chinese. But in English, the first 10 Google search results for “predictive text”, as it happens, contain instructions for turning it off.

New Scientist

Tom Mullaney’s book, The Chinese Typewriter: A Global History of the Information Age, will be published this year. He is also running a Kickstarter campaign to fund the world tour of a Chinese typewriter exhibition, which the Chinese University of Hong Kong has expressed an interest in staging, in either 2019 or 2020.