CNETAnalysis: It might be concerned with aliens, outer space and dimensional jumping for now, but ‘sci-fi’ might have to be redefined if the latest advances in automated writing continue apace. Software that can construct sentences, analyse data and even put a ‘spin’ on a news story are threatening to make the newsdesk and the author’s writing room very different places. The end for journalists? Journalism isn’t complicated. The popularity of online news stories can be tracked – and therefore the importance of news easily ranked – while almost everything is written using the inverted pyramid structure . Since automated writing software can already do most of that, are we looking at the last generation of human journalists? Narrative Science’s Quill is the leading automated writing software title. It transforms structured data into readable, plain English stories that are identical to those written by humans, though at far greater speeds. “Quill’s power lies in the fact that it is a synthesis of data analytics, artificial intelligence and editorial expertise,” says Kris Hammond, Chief Technology Operator at Narrative Science. Quill has artificial intelligence in the form of a natural language engine, as well as an ability to describe, predict, and advise based purely on data. “From a data perspective, the system incorporates state of the art analytics to extract insight, describe situations, predict outcomes, and generate prescriptions based on the core data and the goals of the vertical in which we are working,” says Hammond. “From an artificial intelligence perspective, the system uses the results of these analytics to drive a heuristically based inference engine and the central natural language generation.” Quill’s success hinges on its ability to reproduce journalistic ‘angles’ on a story – the ‘spin’ �! �� and it’s that part which is protected by patents. The rest is just algorithmic analysis of the type that all software uses. Obviously it can’t conduct interviews with politicians, celebrities and football managers, but give it data and Quill can reproduce a bona fide news story in seconds. Some see other weaknesses in humans that automated writing software overcomes. “Some journalists don’t understand what they are writing about and occasionally get the story wrong, or have an irrational bias one way or the other,” says Peter Cochrane , former CTO of BT and now an independent technology analyst based in Suffolk. I can’t fundamentally disagree with that statement without displaying bias, though the concept of a machine being able to review consumer electronics products, apps and software, for instance, from a human user perspective – the only perspective that’s relevant – is a concept that doesn’t make any sense. TechRadar is safe, and investigative journalism is surely beyond the reach of robots, too. Avatars that interview? Perhaps – and there’s no denying that machines do have advantages when compared to humans. “They never sleep, never miss important information or the latest facts, and they deliver better results than humans,” continues Cochrane, insisting that automated writing software has already taken over in the financial industry. “The only limit on the technology is that Quill can only generate language where there is data to begin with,” says Hammond. “While this could, in theory, include textbooks and novels, it is doubtful we will see the technology applied to those problems in the near term. It does, however already send out Tweets during sporting events.” A writer’s best friend? That sounds like a bit of a time-saver for the sports desk. Hang on, could automated software be the journalist’s best friend? It writes news stories and the financial section, leaving human journalists ! to do the! interviews and attend the boozy press functions. If automated software is a time-saver, bring it on. “No one should be worried about automated writing systems,” says Hammond. “As with our technology they are designed for writing into spaces where no one else is writing, and working in coordination with other writers and analysts. That idea of collaboration and is already popular in the wider publishing industry. Aging author Wilbur Smith recently signed a £15 million six-book deal with HarperCollins, though in doing so he announced that he will use a team of co-writers so that he can publish two books each year. The author of Elephant Song and Time To Die hasn’t said he’ll be using robots to write his expansive African novels, but it does suggest that notions of authenticity are becoming dated. Whether they’re becoming dated enough for robots to take-over is another matter. “How many novels do you know that do not include a human character somewhere?” asks Andrew Philpott, Product Manager at NewNovelist , software that helps budding authors structure and write their novel. “Homers Iliad and Odyssey were written thousands of years ago and although the world has changed, they still provide great entertainment today because they are human,” he says, insisting that when someone reads a novel they create a relationship with the author. “If I had the chance, I would love to spend some time with Homer, but would I really want to spend hours with a soulless box of diodes while it tells me a story?” Could a robot write a novel? The short answer is ‘almost’, though not a good one. “Most automated writing software works by imitating samples from human writing,” says Dr John Lee, Assistant Professor in Chinese, Translation and Linguistics at the City University of Hong Kong, whose research group is developing an app that mines word usage and sentence structure patterns from a database of poems from an anthology of Clas! sical Chi! nese poems written during the Tang Dynasty. “Based on these patterns, it suggests the next words to the user as he or she composes a poem,” says Dr Lee, who insists that short poems are one thing, and long novels something else entirely. For anything longer than a few pages software must move on a peg or two and be able to learn not only word usage, but also plot development and characters’ personalities and behaviours. “It would be most difficult for robots to create texts that require both creative content and high-quality language,” says Dr Lee. “Computers may one day be able to write decent novels, albeit ones that sound somewhat like others, but they are unlikely to create true masterpieces.” ‘True’ science fiction The truth is that automated software produces language that’s hard to tolerate, let alone enjoy. “Perhaps one day technology and artificial intelligence will develop sufficiently to allow robots to write a meaningful, original, fictional works,” says Philpott. “However, I think these novels will only be worth reading if robots themselves play a role in our lives as living beings and we share experiences with them.” Sci-fi authors have been writing about artificially intelligent robots for decades. Maybe soon it will be their turn to write about us.
