Imagine that you are finishing a novel and realize that it is one of the best novels you have ever read. Then someone tells you that the novel was written by a robot. Will you believe them?
Today, the world of linguistics and artificial intelligence is at its earliest, pioneering stage in the development of bot writers. Currently, at least two of the Internet’s most significant content producers, Wikipedia and the Associated Press, are using robots to write articles on the Internet.
At first glance, this seems like a shocking evolution in the art of writing. Most people believe that there are certain human tasks or jobs. that robots can never replace — and such a creative and complex activity as writing is one of them. Either this?
The virtual droid that has received the most press lately is a Wikipedia bot named Lsjbot. This is the creation of Sverker Johansson from Sweden, who wrote a code to collect information from a number of reliable sources in order to combine short articles called «stubs» on topics related to animal taxonomy.
Lsjbot reportedly pumps out 10,000 articles a day and has written over 2.7 million articles to date, all of which are human-readable and understandable. According to the media in Popular Science, this accounts for «8.5% of Wikipedia articles». However, as explained on the Wikimedia blog, these Swedish-language articles make up a very large proportion of the articles on the Swedish Wikipedia, but none of them make up the much more popular and voluminous English Wikipedia.
With that said, this does not mean that the English Wikipedia is free from bot intrusion. The real invasion began back in 2002, when Wikipedia curator «Ram-Man» created artificial intelligence. The program he called «rambot» was essentially a script that would scrape the US Census and publish thousands of articles daily, covering virtually any small town, city, or county throughout the United States, and even some municipalities in other countries.
In almost any field you search on Wikipedia, there is probably a first draft Wiki created by Rambot. Even the small, small town of 800 people where I grew up has its own Wikipedia page, created in 2002!
Other Wikipedia article bots over the years have included:
- Robbot — robot, which was originally used to resolve cross-language links and eventually to resolve links to disambiguation pages.
- asteroids This bot has been hoarding NASA data and has written thousands of wiki articles about asteroids.
Today, about a thousand wiki bots roam Wikipedia, constantly making changes to existing pages as errors or omissions are found. The most active is Cydebot, which has made over 4.5 million edits to Wikipedia pages to date.
Other content created by bots
In July of this year, the Associated Press announced that it would be producing automated, robot-written business articles. Forbes is reportedly using bots to publish short articles about companies that are doing well in the market.
The most impressive use of bot technology for article creation was that of journalist/programmer Ken Schwench of the Los Angeles Times, who wrote a program called Quakebot to automatically write articles about earthquakes only minutes after they occur. The data for the articles comes directly from USGS alerts. In an interview with Slate, Ken revealed that it was this year, thanks to Quakebot, that LAT became the first media outlet to report a morning tremor within three minutes of the actual event.
The post consisted of only four short paragraphs and was crafted by weaving relevant data with a pre-written template that Schwenke had created beforehand.
Just like Forbes stock reports and AP business articles, the reports are fast, efficient, and get the job done, but do they represent the future. where more complex and creative articles can be written by bots? Should human writers be worried?
Writing about difficult stories
Of course, linguistics has been a part of artificial intelligence for a very long time. In the article «Artificial Intelligence», published in » Guide to Pragmatics» the authors wrote:
Extended discourse generation requires some careful planning. This complex task can be conveniently divided into two subtasks: deciding what to say and deciding how to say it.
In other words, AI scientists, in trying to find a machine to create discourse that feels authentic to humans, not only have to piece together the right words, but the “bot” also needs to understand how to pronounce those words in the context of the subject. It’s hard enough for the human mind, where understanding of context is ingrained in children from a very young age. For machines, it’s a completely different ball game.
Discourse formation is a process with many constraints, which must take into account various sources of knowledge: knowledge of the subject of discourse, situational context and discourse of the past, as well as knowledge of the interlocutor or reader.
Understanding the subject matter, having a knowledge base of existing information and data, and most importantly, actually understanding what the reader wants are all critical elements in bringing together not only informational text but also creating more abstract writing like creative fiction.
Authors—even very young authors—learn to do this intuitively. For programmers to create artificial intelligence that can do the same, it requires a level of algorithm generation (and self-learning) that is still far more advanced than what the data-cleaning robots of Wiki, Associated Press and others are still capable of. , However, these authors have described how this is not impossible.
First, new symbols and structures can be created dynamically during program execution. Second, structures can be recursively defined and thus can represent a potentially infinite number of real structures. And third, programs are also symbolic structures and thus can be created or controlled by other programs.
If you look at Ken Schwenche’s attempt to use Quakebot to create fast and accurate articles about earthquakes, you’ll see that some people play simple games of formulating patterns that the program can use to simply insert data into the right place. and the article «sounds» like it was created by a human, but only because the actually she is was created by man ahead of time.
However, there are some people, like Narrative Science, who are taking this concept to a whole new level and actually applying a crude form of intelligent AI. about the content they produce for companies like Forbes and government intelligence agency In-Q-Tel.
Programs that write like people
Narrative Science programmers take complex data—whether it’s a scoring chart and player statistics in a professional football game, or stock indexes and business data about companies—and use the data itself to pinpoint needs. to be said and how to say it.
For example, in 2011, The New York Times provided an excerpt from the sports report Narrative Science, which shows what this technology is capable of.
Wisconsin appears to be in the driver’s seat on its way to victory as it leads 51-10 after the third quarter. Wisconsin added to their lead when Russell Wilson found Jacob Pedersen for an eight-yard touchdown to make it 44-3.
As you can see, Narrative Science creates an algorithm that uses both context (sports) and data (player scores and stats) to formulate a report that sounds exactly like what sports fans expect people to read. writing about sports. ,
Where do the bots go from here
Even this impressive use of data analysis and AI linguistics is very limited in scope and scope. Company founder Chris Hammond said that in 20 years, the company’s own computer program could win the Pulitzer Prize for journalism.
While the enthusiasm is commendable, the reality is that it will likely take more than twenty years to accomplish this feat.
Example: Just this year, researchers at the University of New South Wales in Australia created a computer program they called the Moral Storytelling System. The goal was for the system to create a fable based on the user’s preferences.