Does your predictive text feature on your device predict your writing?

Hang on for a minute...we're trying to find some more stories you might like.


Email This Story






Kenneth C. Arnold asserts that the predict text feature on most of our electronic devices perpetrates certain biases while typing. Arnold,  a graduate of Harvard University involved in the field of Human-Computer Interaction and Artificial Intelligence (HCI + AI), noticed this bias when one day he decided to tell his wife about his day. Instead, he ended up trading one word for another other because he assumed that the suggested word better described what he intended to say.

He affirms that this technology pushes us in a different direction than we usually intend to go. Thus, Arnold conducted an experiment which involved image reviews on two restaurants online to find out if his initial hypothesis was accurate. To control the task, one of the restaurants did not have any recommendations whereas the other did.

He noticed that with predictive suggestions about the restaurant on the website, people wrote what the predictions made easier to type. Arnold also observed that most of the positive reviews for the restaurants were positive because most restaurants had some positive reviews already. Statistically analyzing the results, he further noticed that the positive predictions generated while typing increased by an additional 75 percent towards their star rating as a result of the positive recommendations.

As a Christian computer scientist, Arnold contends that communication is an inherent attribute we as humans have since we were made in the image of God. To try to solve the problem of biased typing due to predictive text features on our mobile devices,  Arnold is developing a new suggestive mechanism which doesn’t necessarily display cue phrases as well as nudge the user to implicitly use them. Rather, he asserts that his new suggestive feature on mobile devices will involve posing suggestive questions based on the user’s recent typing activity.

For instance, if a user types “This is one of my favorite spots!”, Arnold says this feature might pose a follow-up question like “Is it cheap?” One of the challenges he faces with his solution is that many of the critiques he has received, including from students present at the event, was that his suggestion seemed similar to what predictive text features on mobile devices do.

Arnold emphasizes the fact that the cue phrases that are currently in use allow the user to rather unconsciously choose a phrase which he or she may not intend to use. Though Arnold’s solution is still in the working phase, he believes it will help make communication a more human interaction rather than a machine interaction.  

In the end, Arnold encouraged students who want to get involved in this field to not only consider the field of computer science but also be explore fields like psychology as well as statistics as these are useful tools needed.