It's kind of like the bastard child of Mad Libs and the stock market tables. Called Wordsmith, it's a product from a company called Automated Insights that generates short articles base on financial data. Feed it a bunch of info like company names, net income and earnings per share, and it generates a readable narrative suitable for publishing. For instance, if the table has the name Apple, Q1 net income of $78,400,000,000, EPS of $3.36, you get something like "Apple today announced financial results for its fiscal 2017 first quarter ended December 31, 2016. The Company posted all-time record quarterly revenue of $78.4 billion, and all-time record quarterly earnings per diluted share of $3.36."
Content-generating engines like this one from AI and other firms such as Narrative Science, Arria and Yseop are used by companies from the Associated Press to Forbes to Yahoo to generate publishable pieces quickly and cost efficiently. Automated Insights says that its software is used to create over a billion stories a year, so odds are you have consumed it without even knowing it. There are even specialized versions of similar programs focused on specific areas. For instance in 2014, the first published account of a California earthquake, hitting the pages within 3 minutes from when the ground started shaking, wasn't written by a person, but generated by a computer. That program got its data from the US Geological Survey data stream and "wrote" an article about the trembler. The software, appropriately enough, was called Quakebot.
But surely you could tell the difference between an article written by a machine and one by a human. I mean, a computer-written article would be clunky and formulaic and boring, whereas one by a human would be engaging and pithy and interesting. Right? Well, actually, not necessarily. Mind you, we're not talking Shakespeare here, but rather your basic everyday journalism. And in a study by Christer Clerwall of Karlstad University in Sweden, the data shows that the differences between workaday writing by a person and software were virtually indistinguishable. In "Enter the Robot Journalist" Clerwall writes "we can say that the text written by a journalist is assessed as being more coherent, well written, clear, less boring, and more pleasant to read. On the other hand, the text generated by software is perceived as more descriptive, more informative, more boring, but also more accurate, trustworthy, and objective. But are these differences significant? The short answer is, no they are not."
This all came to mind because you might have noticed a new feature on the bottom of your Gmail window on your phone. Called Smart Reply, it's a context sensitive set of suggestions that you can use to answer a given missive, saving you from having to create a response. It's grown in popularity after being introduced and tested in 2015 in Inbox, Google's own email system. There, 12% of all email replies sent currently are Smart Replies.
It works like this. Unprompted, every inbound message is scanned, and three appropriate answers are suggested. So a note from a client with an updated project schedule arrived with three buttons on the bottom for me to click: "Got it, thanks!" "Thanks!" and "I'll be there." Meanwhile, the next message was a link my wife sent me with some weekend activities, and the buttons said "Thank you," "Let's go!" and "Do you want to go?" Tap one, and the person on the receiving end will think you've actually cared enough to read and respond. Little do they know that some Big Data computer in a server farm Montana is doing the thinking for you.
At this point it's all pretty tame and boilerplate. To be pithy or smartassed still requires an actual human thumb-typing a response. But it's not hard to imagine as the system gets better, it will not only read the incoming mail, but learn your own personal tone from your responses. And then the three buttons will be more than just formulaic responses, but short answers than will really seem to come from you. Then machine generated snark will be possible, and the buttons are more likely to offer up options such as "What a waste of time!" or "That sounds boring!" or "You've got to be kidding!"
Progress. There's no stopping it.
-END-
Marc Wollin of Bedford uses GMail, but keeps his AOL account for sentimental reasons. His column appears regularly in The Record-Review, The Scarsdale Inquirer and online at http://www.glancingaskance.blogspot.com/, as well as via Facebook, LinkedIn and Twitter.
Content-generating engines like this one from AI and other firms such as Narrative Science, Arria and Yseop are used by companies from the Associated Press to Forbes to Yahoo to generate publishable pieces quickly and cost efficiently. Automated Insights says that its software is used to create over a billion stories a year, so odds are you have consumed it without even knowing it. There are even specialized versions of similar programs focused on specific areas. For instance in 2014, the first published account of a California earthquake, hitting the pages within 3 minutes from when the ground started shaking, wasn't written by a person, but generated by a computer. That program got its data from the US Geological Survey data stream and "wrote" an article about the trembler. The software, appropriately enough, was called Quakebot.
But surely you could tell the difference between an article written by a machine and one by a human. I mean, a computer-written article would be clunky and formulaic and boring, whereas one by a human would be engaging and pithy and interesting. Right? Well, actually, not necessarily. Mind you, we're not talking Shakespeare here, but rather your basic everyday journalism. And in a study by Christer Clerwall of Karlstad University in Sweden, the data shows that the differences between workaday writing by a person and software were virtually indistinguishable. In "Enter the Robot Journalist" Clerwall writes "we can say that the text written by a journalist is assessed as being more coherent, well written, clear, less boring, and more pleasant to read. On the other hand, the text generated by software is perceived as more descriptive, more informative, more boring, but also more accurate, trustworthy, and objective. But are these differences significant? The short answer is, no they are not."
This all came to mind because you might have noticed a new feature on the bottom of your Gmail window on your phone. Called Smart Reply, it's a context sensitive set of suggestions that you can use to answer a given missive, saving you from having to create a response. It's grown in popularity after being introduced and tested in 2015 in Inbox, Google's own email system. There, 12% of all email replies sent currently are Smart Replies.
It works like this. Unprompted, every inbound message is scanned, and three appropriate answers are suggested. So a note from a client with an updated project schedule arrived with three buttons on the bottom for me to click: "Got it, thanks!" "Thanks!" and "I'll be there." Meanwhile, the next message was a link my wife sent me with some weekend activities, and the buttons said "Thank you," "Let's go!" and "Do you want to go?" Tap one, and the person on the receiving end will think you've actually cared enough to read and respond. Little do they know that some Big Data computer in a server farm Montana is doing the thinking for you.
At this point it's all pretty tame and boilerplate. To be pithy or smartassed still requires an actual human thumb-typing a response. But it's not hard to imagine as the system gets better, it will not only read the incoming mail, but learn your own personal tone from your responses. And then the three buttons will be more than just formulaic responses, but short answers than will really seem to come from you. Then machine generated snark will be possible, and the buttons are more likely to offer up options such as "What a waste of time!" or "That sounds boring!" or "You've got to be kidding!"
Progress. There's no stopping it.
-END-
Marc Wollin of Bedford uses GMail, but keeps his AOL account for sentimental reasons. His column appears regularly in The Record-Review, The Scarsdale Inquirer and online at http://www.glancingaskance.blogspot.com/, as well as via Facebook, LinkedIn and Twitter.
No comments:
Post a Comment