For some time, sui social network the game of complete sentences with smartphone automatic suggestions to then compare the results: those that are published are sometimes comical and surreal, but on as many occasions they take on a complete meaning that demonstrates how much the artificial intelligence systems (to be precise, natural language processing algorithms) which help complete sentences, often providing a choice between 3 words that might be suitable.
Second as explained by Paul Lambert, responsible for Google’s Smart Compose feature (the automatic suggestion system used on Gmail), the idea of a program able to help the writer to conclude the sentence also comes from the experience of those who deal with software. Since codes can contain long strings of identical sequences, computer engineers often rely on shortcuts called code completion, which save time.
This is where the idea of creating a similar system could have come from, but designed to help the many workers who, according to a study by McKinsey, they spend over a quarter of their time replying to emails and messages.
Google, the Accademia della Crusca and the challenge of what it is
by Emanuele Capone
How automatic prompts work
But how do these systems often accurately predict what the next word you intend to type is? The logic is the same used by the Google Docs correction system: By analyzing billions of emails or text messages, these artificial intelligence systems learn to understand which words are most likely to follow the ones you just typed. In more advanced cases, by dint of independently predicting the succession of words, these algorithms (notably OpenAI’s GPT-3) have also managed to write short essays of complete sense.
Autocomplete software relies in part on people’s general behavior – for example, if you’re typing a message on a Friday it’s more likely that, after typing the word “buon”, it is suggested to complete the sentence with “weekend”. Other suggestions are based on the analysis of one’s personal use: when I write down the shopping list on my smartphone, after having written “baby food” the prompter immediately advises me to add “Johnny”, which is the name of my dog. Which “baby food” is actually intended for.
Does all of this actually make you faster? On the subject, the studies are divided: according to some researchers, autocomplete could have the paradoxical effect of slowing down writing, especially since it forces us to parse between various options while we are already busy typing; others indicate how it can improve the speed and the accuracy of people who are less familiar with the smartphone keyboard.
the interview
Because we need more women who deal with artificial intelligence
by Andrea Daniele Signorelli
Joys and sorrows of prompters
However, other studies have not focused on speed and accuracy, but on the way in which the use of predictive text programs can change the way you write: “Suggestions are offered several times per second in the middle of the planning process of what we are writing, consequently they can shape the content we write,” it reads in the studio signed by Kc Arnold of the MIT in Boston.
The results of this research, for which the texts of hundreds of people were analyzed, written with and without the aid of automatic suggestions, show how those who use predictive writing systems significantly increase the use of predictable words. This means that people select certain words, often simpler and more vague, through the prompter, which they would otherwise have replaced with less common and more specific terms. Not only that: in the same research it was highlighted how those who use the prompter tend to write shorter messages. According to the researchers, this happens because the prompter encourages us to skip a few words, immediately offering the next one as a suggestion and thus shortening our text. An example related to the context: while I was going to type “our” (in the row above), the software could have prompted me to write “text” immediately, skipping the possessive adjective and thus shortening the message.
More predictable words and shorter sentences: put this way, the predictive writing system doesn’t seem to incentivize you to formulate more original or better texts. On the contrary: as happens in other systems in which the selection algorithms play a fundamental role (the music on Spotify, the tv series on Netflix, information on Facebook), it is possible that this system will lock us up in a sort of linguistic filter bubble, in which we are encouraged to always use the same words, limiting our expressiveness.
When the smartphone writes for us
There is still another element, because on some occasions the automatic suggestions even make us use expressions that do not reflect our character. Recently, my sister wrote to me to give me good news regarding her e I was about to answer her: “Brava! I’m happy for you”. When I arrived at the “f” of “happy”, the automatic completion also showed me the alternative “proud”. And without thinking too much about it I wrote: “I’m proud of you”. A way of expressing myself that is not mine at all and that sounded unnatural to me the moment I clicked Send. And in fact it was not me who wrote: at least in part, it was the auto-completion of the smartphone that answered my sister.
But then is it really the phone that thinks for me? Apart from some exceptional cases, the day when artificial intelligence software will replace our cognitive faculties is still a long way off and may not even come. Sometimes, however, it is difficult to escape the feeling that this path has already been taken.
.