Can an Algorithm Write a Better News Story Than a Human Reporter?
m.wired.com
m.wired.com
Here's Ben Welsh of the L.A. Times talking about "Human Assisted Reporting" (which itself is a play on "Computer Assisted Reporting"): http://www.youtube.com/watch?v=iP-On8PzEy8
Oh, and here's the slide deck: https://docs.google.com/a/knowtheory.net/presentation/embed?...
First, publishing (whether it's papers, broadcast, whatever) is a big lumbering business. There are entrenched interests, people who fear or loathe technology, egos, and all of the other perils which modulate large scale human endeavors.
There are people within the news world who have been working on computer assisted reporting, doing all sorts of cool stuff for decades now. What hasn't happened is a uniform and widescale adoption of all of these practices, because much of the workflows involved in the business are tied to their legacy business, and those legacy components are really difficult to shed for a pile of reasons.
Ben may be talking about this stuff as a good idea, but it's not that nobody's been doing this stuff... it's that stuff like this isn't standard practice across the industry (yet).
Also, please remember that Ben is talking about automating these processes. Reporters have been filing FOIA requests manually, and getting info from the government essentially for as long as there has been a government and foia laws. That's where part of the tepidness about technology comes from. The government isn't particularly friendly to efforts to stream line processes that get more information to reporters.
It'll often be a CD of some 1970s mainframe database format with no information on what format it is, how to open it, etc. They'll do a little happy dance if you just give up and move on to another story.
I had never seen one in action before.
That said, it's a really neat idea and I'm curious to see this technology mature.
But also, once you get to massive robot production, then the amount of journalists will seem small. Think of content personalized to each user. Real time description of things around you in your software. In our startup we plan to have a robot write the plot for user generated games based on their custom settings. Hiring a writer for each use would certainly increase our costs.
The thing that frustrates me about the way that Narrative Science is discussed and some of the unhinged techno-futurism stuff that people write, is that it's never about making people or individuals more effective.
Automated processes and people have different strengths. Computers really aren't that great at pretending that they're people, and people are bad at accuracy on repetitive tasks.
That people keep on talking about "replacing" so-and-so misses the point. Change their market, change their workflows, make them better at accomplishing the function journalists or whomever play in society.
Endeavoring to make them obsolete is both unnecessarily inflammatory and also not terribly sensible.
For our company specific use. Hiring a writer to be assisted by our robots would just add unnecessary cost to the process and delay to the end user experience. Versus pretty much little to no gain in quality. We only need a simple transformation from structured data to narrated english, and we want lots of it to be delivered instantly to our users.
It just makes more sense for applications like ours to have robots replace humans.
He stayed two days [at Stanford] before dropping out to start Zip2, which provided online content publishing software for news organizations, with his brother Kimbal Musk. In 1999, Compaq's AltaVista division acquired Zip2 for US$307 million in cash and US$34 million in stock options.
Either: (a) stories will be more objective, and provably so; (b) editors will need to consciously decide how the algorithms should slant (and this itself would be discoverable, if only by leaks); or (c) we'll learn that the underlying datasets were assembled in a way that is more critical of one side or another.
I want to read details that can only come from being there in person. I can look at a box score to see Lebron James scored 27 points or missed a free throw at the end of a game. A journalist's job is to tell me that he appeared distracted by the crowd loudly booing a controversial call or that Andrew Bynum pouted on the bench after getting into a heated argument with his coach during a timeout. Those kinds of observations are what makes a story come alive and it will be a long time before software can replace this human element.
Not everyone does necessarily. There is a place where good writing and journalism overlap, but that's not the totality of what journalism is for.
Robots only do what they're told and are only as good and creative as the coder. If you can write an emotional heated narrative. Then you can code a robot that does the same. There's nothing intrinsic to programming that makes it impossible for good writers to coffee a robot that mimics their style.
Unless you have structured data that represents all human experience someone present at an event could muster (smell, sound, emotional feeling...) I don't see how a robot could replace it.
Congratulations you've replaced the absolute bottom of the barrel of reporting, that which could and probably should be replaced with a table or two.
Until the software actually attends the event somehow to gather the information and add additional context its just a clever trick. If you want to save money on this kind of reporting id probably stick to outputting some nice tables of data in the sports section.
I thought the whole point of HFT was to use small shreds of information to play the liquidity gap between sell and buy orders.