The latest Gboard features rely on machine learning to give you automatic suggestions for emojis, stickers, and text relevant to what you write, but you need to know as much as possible about your digital life in order to function accurately.
But where is this data processed to provide recommendations for responses, and how does Google maintain your privacy?
Google has insisted for years that the Gboard does not keep any data about keystrokes, nor does it send it to its servers, and that the only time Google knows what you type in the Gboard is when you use the app to search for something in a search engine. Google, or enter other data into Google services, which is data that you must see even if you use another keyboard application in your phone.
But the Smart Responses feature gives you automatic suggestions about what you write, and this has potential privacy implications more broadly, since the feature relies on real-time analysis of everything you type in the smartphone to provide useful suggestions.
Says (Xu Liu) Xu Liu, Director of Engineering Department in the (Gboard) App Development Team: “We are always working on developing the (Gboard) to become more intelligent, and we want to give you the correct prediction of emojis and texts but we do not want to record anything you type, and there is no Text or content is transmitted to any of our servers at all, so this is a big challenge, but privacy is our first focus.
To achieve this privacy; Google runs all the necessary forecasting algorithms locally in your device, so you don’t see your data or send it anywhere. And one more thing; Google doesn’t rely on the Gboard app itself to do the processing, but it does on your device itself.
But in light of the development of cyber attacks targeting smartphones, are the precautions taken by Google sufficient?
Some experts suggest that granting the Gboard app access to content that feeds these algorithms will have significant risks, for example: Malicious apps may try to attack the keyboard app to access data that it should not be able to see.
So the Gboard app development team had an idea: Why don’t the Gboard app itself completely come out of the equation and rely on Android to run machine learning analytics to identify response recommendations?
Indeed; The Android system runs all your applications and services, which means that you have already entrusted it with all your data, and to access this data you must hack your phone to control the operating system itself.
So when the Gboard in Android 11 gives you three suggestions related to what you write; These suggestions are not from the app itself, but rather depend on the app and the operating system at the same time.
“It’s a smooth experience, but it depends on two layers,” says Xu Liu. “The first is the keyboard layer itself, and the other is: the operating system layer, but the second is not visible to you.”
It is worth noting that the entire Android system is supported by (unified education) technologies. Federated Learning From Google, it is a way to build machine learning models from data sets that come from all different sources and are never combined, such as: using data from all users’ phones to improve prediction algorithms without transferring data from their devices.
This does not mean that adding the Smart Responses feature in the Gboard does not have any security risks, because the more software you design, the greater the opportunities for the emergence of errors and defects that can be exploited by hackers, but the development of the recommendations engine to be run via the Android system itself instead of an application Gboard is an important security step by Google.
Kenn White, security engineer and project founder, says:Open Crypto Audit Project): “For many of us, almost all of our personal data is in our phones, and it depends on our confidence in this technology, but there is a huge burden on Google to make sure that these types of features are based on strong security factors.”