Is the Google company taking a step forward in automatic text creation? In an article published by the Cornell University Library, Google artificial intelligence researchers explain how they use software to generate Wikipedia-style articles.
The process works in particular by basing itself on the first ten web pages that emerge during a search on a given subject, by excluding those coming from Wikipedia, or by using links from the “References” category of Wikipedia, as explained. an article of theregister.co.uk. Mohammad Saleh, co-author of the article and engineer in Google’s artificial intelligence team, told The Register that the extraction phase is one of the sensitive points. It is at this stage that the data are chosen which will pass to the abstraction phase. Whereas ideally engineers would want all data from reference documents to pass.
The question is whether the results are convincing. In their article, the researchers publish an example. The text is quite readable. When we compare it to that of Wikipedia, therefore created by humans, we realize that the artificial version is ultimately longer. The formatting is not perfect. For example, capital letters are missing at the start of sentences. The question of verifying information inevitably arises. Since the system automatically takes the first results on a given subject and there does not seem to be any other criteria for the moment.