These are the bank messages, alerts, one time passwords (OTPs), verification codes and more. And while all of these messages have various use cases, the important entity in all of them are usually a few codes or numbers. The smart messaging software on the new OnePlus 7T takes this information and highlights it, instead of having you read all the other text.
The software though comes from a company called Gupshup, pretty well known in the chatbot space. It’s an artificial intelligence (AI) that works on the device, thereby ensuring privacy and not sharing any data with OnePlus or Gupshup’s servers.
Long time OnePlus or Xiaomi users would know that the companies have developed software earlier to pick OTPs from texts and make them easy to copy, instead of having the user actually open the text. Gupshup’s software though uses AI, which enhances its reach, according to CEO, Beerud Sheth.
Sheth explained that traditional rules-based algorithms can be used for simplifying messages. On the front end, users really only see the cards, so the AI doesn’t make a big difference to them. However, the software recognises about ten types of texts right now and it’s difficult to write rules for all of them.
The software can recognise bank messages, OTPs, flight messages, bill reminders, courier and cab bookings, recharge notifications and more. Sheth says more types of messages will be enabled in future and the company, theoretically, could also integrate direct payments services to them, customer support bots and more. So, if you get a text to pay the electric bill, you could direct the app to pay the bill automatically.
To develop the AI model, Gupshup started with millions of sample messages. “The first goal is to determine what kind of message you’re getting. The human brain can classify messages easily, but it’s not that easy for software," explains Sheth.
Brand marketers phrase these messages differently. For instance, “your food is arriving" and “your package is arriving" would fall in the same category, but the phrasing is different. “Tomorrow, some new marketer will create a new kind of message that you have to read (and classify)," explains Sheth.
While classifying is one aspect of what the AI does, the other aspect is something called “entity extraction". For instance, if your text is about flight timings, the AI has to determine the important information from that, like PNR, flight timings etc. “It’s humanly impossible to write rules (for this all of these) because the variety of messages is so huge that we’ll have to write thousands of rules," Sheth said.
Gupshup’s AI software has seen millions of messages and developed a model. When it sees a new kind of message, it classifies the same and then extracts the relevant entities. It includes technologies like machine learning (ML) and natural language processing (NLP).
The differentiation between AI and rules-based models is important here, because other companies have tried rules-based models like this before. “It is very easy to get 10-20% of the messages users receive because you can just create one template," Sheth said.
Rules are accurate 100% of the time as long as their parameters are met, but AI is correct 98% of the time regardless of the parameters. Sheth claims the software achieved 97-98% accuracy in internal testing.
He also explained that while the app does include AI capabilities, it doesn’t learn as it goes right now. That’s because it’s on the device and doesn’t share any information to the server side. There are ways to change this in future, but at the moment, the company just plans to send updated AI models are improvements are made.
The software can only deal with the English language right now but Gupshup also wants to read vernacular messages in future. While OnePlus is the first company to implement its software, the company is also working with other phone makers to get the same on their devices.
The feature is also available only on OnePlus phones in India right now, though Gupshup is expanding the platform to cover text messages in other countries.