Digital Ventures

Back to blog

Please Do Not Disturb: When AI learns to leave you alone




You are interrupted. You inevitably pick up the phone. The screen shows an unimportant notification. It says that there is a new episode of a podcast. You are quite annoyed. Damn, I was just in a middle of something, now I’ll have to start focusing all over again. You open the settings and turn of this application’s notification, so it never bothers you again.

Turns out, you only find out about a podcast that you are actually interested in after it’s been broadcasted five days later. You begin to think, is there a “middle pathway”? How can an application know when to notify you and when not to? Can AI learn to do this without us having to tell them “do not disturb”?


Such request is quite human-centric, yet it is a question that many researchers are seeking answers. This is because, from statistics, 60% of users opt-out the push notification. Moreover, after an application’s first download, users have a 90% chance of discontinuing usage of the application within a month.

Think of yourself, when we first download an application, we may use it once in a while. However, if we don’t use it as a habit, we tend to easily forget about it and even forget that it sits on our phone.

Last year, researchers from Rutgers University published a study titled How Busy Are You? Predicting the Interruptibility Intensity of Mobile Users. The study explored a two-stage hierarchical model to predict people’s interruptibility intensity. First, it predicts whether the user is available or unavailable. If available, it will enter the second stage which is the intensity of whether users are not interruptible, highly not interruptible, highly interruptible, interruptible, a total of four stages.

Researchers collected more than 5,000 smartphone records from 22 volunteers over four weeks, to evaluate people’s level of busyness (busy = not interruptible). Interestingly, researchers look at how personalities aligned with the Big Five personality trait to evaluate how they feel about being interrupted. They hope for mobile phones or electronic gadgets to become “an excellent human secretary who knows when you want to be interrupted such as a call from the kids should always be pass through while if a certain relative call, maybe it can be notified at a later time”

Researchers added “ideally, the smartphone would learn automatically. As it is today, the notification management system is not smart or only depends on a user’s setting.  Our model is different because it collects users’ activity data and preferences. This allows the system to learn automatically.”

Aside from Rutgers University, this year, a university in Taiwan also published a paper regarding the same topic. Their paper’s adorable name titled “C-3PO: Click-sequence-aware DeeP Neural Network (DNN)-based Pop-uPs RecOmmendation” (a great attempt to use the abbreviation C-3PO). They mentioned that “today, mobile applications face problems in that it overly notifies users which can result in annoyance and users may delete the application. Yet fewer notifications may also make users forget and discontinue using the application. We are trying to use deep learning network to study when and how often we should notify users for it to be most efficient.”

Taiwanese researchers train AI with user data such a web browsing history, finance data, and online purchasing data. This helps to create models that learn each consumer’s behavior which helps to predict the most appropriate interruption rate. Results after application testing in Taiwan reveals that users are less interrupted (lower push notification) while each notification has a higher click-through rate (which is predictable, when we are less interrupted, and we know that each notification must be important, we tend to click more.)


In the future, is it possible that we can focus better…


on what we are currently doing…


and be able to focus longer…

Beep-beep! as AI help mobile phones Beep-beep! learn to Beep-beep! leave us alone!



C-3PO: Click-sequence-aware DeeP Neural Network (DNN)-based Pop-uPs RecOmmendation