The newspaper written by robots
Would you read a newspaper in which you worked an algorithm? Yes, the question is a bit tricky by two points: first, algorithms are already used in almost all newspapers to optimize traffic (although few at the level of writing); and, secondly, “work” may not be the key word.
The journalistic robots already exist and, although they do not stand out by elaborate parables or careful text, they are a trend towards a digitalized world. Maybe they do not completely substitute journalist writers, but they already have a foot (metal) inside the rotary presses, and they will not remove it.
What are the “journalists” robots doing today?
When we talk about journalism we have to mention several roles: local journalism to move to a point, investigate, interview people; and the journalist of edition or writing, more involved in the writing. It is in this last field where the robots have more court, given their difficulty to walk.
That said, the interviewer robot already exists. Her name is Jia Jia and her interviews, although somewhat absurd, follow a logical enough argumentative thread. It belongs to the Xinhua News Agency, the official news agency of the Chinese government and, although it is more a promotional façade than a functional tool for quality journalism, it is still interesting.
Jia Jia is able to speak with relative fluency (not as much as Sophia, of the Hanson Robotics, but almost) and to make a battery of questions by slightly modifying some of the statements. Surely it is more than any of the most denied humans, although there is a long way to go to reach the average.
Ana Gamazo, a journalist specializing in ICT economics, commented a few months ago for Nobbot that a “journalist robot cannot replace a professional nor in the most routine activities”, a statement that will be confirmed (or not) over time.
At the moment, there is already at least one Spanish company that, after analyzing some notes of sports journalism and its simplicity, pretends that from more or less dense headlines to very short news be written by algorithms. It’s Narrative, a local startup that already has an algorithm sufficiently refined to broadcast this type of news:
“Arsenal won 2-0, adding 8 wins in a row at home. Manchester had bad luck despite managing possession of the ball ”
Start with those easily automatable, sports results, economy, perhaps even political (results, not debates), and will go up the difficulty ladder as it is fed with more data; since it uses Big Data and AI techniques. The robots are not going to buy for us yet, but they are already writing a bit.
The agency L.A. Times has been experimenting for several years with automatic news regarding catastrophes such as earthquakes or tornadoes with data from different universities and organizations that handle this type of data in record time.
Please note: We just deleted an automated tweet saying there was a 6.8 earthquake in Isla Vista. That earthquake happened in 1925. – L.A. Times: L.A. Now (@LANow) June 22, 2017
In general, everything went well: the information is simple but true and serves as a first contact with the news. But from time to time a false news is leaked, like the warning of an earthquake for 2025 that actually happened in 1925. Someone had sent the data wrong and the robot journalist could not contrast them.
As a result, in June 2017 the news was published about an earthquake of 6.8 on the Richter scale for which there would be about eight years and that, of course, was not real.
Some difficulties for “journalists robots” to become a reality
Robots analyze structured information much better than humans but, unfortunately for them, most news is based on an unstructured data set. For example, like this article, which is not written in Excel with rows and columns, it has images in between, with hyperlinks that point to different sources, with format changes …
Analyzing structured data and extracting knowledge from it is really easy for an algorithm or an AI, but when we include text analysis, processing information requires more calculation capacity and much more optimized software. In another example, an algorithm discovered that Shakespeare had not written Henry VI, but not understanding the text, but counting words and structuring information (with the loss of context that entails).
Machines process, like many animals, but lack consciousness. It does not mean that they are not intelligent since non-conscious intelligence is perfectly valid as intelligence, but not to interpret texts, make interviews or summarize information for a human to enjoy it.
In a third example, we could mention the parables, the literary resource by which, mentioning one element, actually speaks of another. For example, in this well-known fragment by Garcilaso de la Vega it seems to be talking about seasonal change, harvest and snow, when in fact it talks about age:
Take from your cheerful spring the sweet fruit, before the angry weather covers the beautiful summit with snow;
In the same way that translation requires an understanding of language beyond language itself (culture, political tension, society …), robotic journalism will not be a reality until the algorithms are able to understand, even if it is an unconsciously, the text.
The reading between lines that we like so much to humans is a skill that, at the moment, do not have robotic journalists, so it is unlikely that in the near future we will be replaced. We are safe, at least for a few more years.
However, given the advances in convolutional neural networks and machine learning, it is very likely that we will not see the kick off the AI coming on our backs to the desk chairs. Maybe in a few years, we will have to change from one profession to another with non-automated skills.
Image: iStock / Ekaterina79, iStock / Zerbor, iStock / Bestjunior, iStock / Ieremy
Also published on Medium.