Wikipedia will use artificial intelligence to increase the number of editors

Wikipedia is an online encyclopedia that anyone can make changes to. As for crowdsourcing, the creation of a non-commercial encyclopedia site forever changed the process of finding information by users. Wikipedia is among the top ten most visited sites on the Internet, replacing heavy volumes of huge encyclopedias. But she has her drawbacks. If everyone can make changes to Wikipedia, then anyone can mistakenly add incorrect information. And anyone can harm the site by adding erroneous data intentionally. Aaron Halfaker, an American computer scientist who works at the Wikimedia Foundation (this organization and owns Wikipedia), created his own AI system to detect such acts of vandalism.

"It turned out that the lion's share of vandalism is not something witty."

In a sense, this means facilitating work for volunteer editors who are checking out Wikipedia articles. And this may seem like another step towards refusing the services of these editors (another example of how AI can replace human work). But the Halfquecker project is rather an attempt to increase the human contribution to the construction of Wikipedia. And while some people predict that AI and robotics will replace 47% of human labor in the next 20 years, others are confident that AI will also create new specialties for servicing such systems. The same project is at least a small example of such a trend.
')
“This project is an attempt to re-engage a person’s work to engage a person’s attention where it is needed,” said Dario Taraborelli, head of research at Wikimedia.

Don't scare off newbies

In the past, if you tried to make changes to one of the important articles of Wikipedia, you often received an automatic response with a refusal. The system did not allow to take part in the creation of the encyclopedia, if users did not follow clear rules, and according to the results of the study of Halfaker and other scientists, this caused many people to abandon the idea of ​​editing Wikipedia (and they could become permanent editors). A 2009 study showed that eight years after the launch of the project, people began to take less and less part in it.

“This is because the newbies do not stay with us,” says Halfpacker. “In fact, on Wikipedia we exchanged the experience that the new editors of the encyclopedia could get for the effectiveness of the fight against vandals and unwanted people coming to the service.”

In the matter of this AI project Objective Revision Evaluation Service or ORES echoed Mr. Halffecker’s opinion, believing that his main goal is to make Wikipedia more pleasant for new editors and to get people to take an active part in Wikipedia’s life. Using a set of open source machine learning algorithms called SciKit Learn (a free code available to everyone), the service hopes to automatically recognize obvious vandalism and separate it from those changes that were made from good motives. With a more detailed acquaintance with the changes, these algorithms could identify vandals without scaring off potential active participants. This does not mean that Wikipedia needs automated tools to attract more editors. The bottom line is that Wikipedia needs the best automated tools.

“Our approach should be different depending on whether the changes were made for good or evil reasons,” says Halfaker, who used Wikipedia as the theme for his doctoral thesis at the Department of Computer Science at the University of Minnesota.

Globally, AI algorithms are just simple examples of machine learning. But they can be very effective. Their job is to recognize specific words or combinations of certain words, or specific key patterns. For example, they can fix unusually large blocks of characters. “Vandals tend to recruit everything in a row without spaces,” says Halfacker.

He admits that currently the service is not able to detect every act of vandalism on the site, but he hopes to find most of them. “With such strategies, we cannot track well-written nonsense. But it turns out that the lion’s share of acts of vandalism is not original. ”

Wikipedia articles that write themselves?

Meanwhile, giants such as Google, Facebook, Microsoft and others like them are developing new machine learning technologies known as deep learning. Using neural networks — networks of machines that are similar to those of the human brain — deep learning algorithms can recognize photos, “understand” words that are spoken aloud, and translate from one language to another. For example, such a network can be taught to recognize dogs by uploading photos of dogs.

Using the same algorithms, scientists begin to develop systems that recognize natural language — the way people speak and write every day. “Feeding” tons of dialogues to such networks can teach the machine to keep up the conversation. If you let them get acquainted with a huge number of news posts, you can teach the machine to write articles on its own (although it’s still far from this point). And this may be one of the factors determining the future in which machines will be able to edit Wikipedia themselves.

Halffecker is sure that we are still very far from such a future. And even if it comes, then, according to him, Wikipedia will still need people who could direct these networks. “I’m not sure that there will ever be a time when cars can outperform a person’s opinion - or they will not come so soon,” he says. “But even in this case, we still want human opinion to be part of this process.” And so he created an AI service that could increase the army of Wikipedia editors.

He and the Wikimedia Foundation do not implement these algorithms, offering them as an online service that can be used by the wider Wikipedia community. “We have simplified the ability to experiment and criticize algorithms,” says Halfaker. “We want to create a dialogue so that it contributes to our progress towards the future in which we cope with new content, using new methods and cooperating with new editors.” This is AI. But, again - this is fully consistent with the "human" principles.