AI on-device

Late last year, the personal information of 31 million Android users was exposed thanks to a popular keyboard app. The app, AI.type, stored this data on a server owned by the company’s co-founder, rather than on a secure server. The information contained everything from names and emails, to exact locations, web searches and IP addresses. This incident was one of the many 2017 breaches that shared a common thread – the exposed data was stored on a single, centralized server.

Unsecure centralized servers act as easy targets for hackers looking to access sensitive information in bulk, and AI platforms and other services hoarding millions of users’ private data that rely on these systems will continue to experience breaches. This reliance is driving the need to shift to a decentralized approach where AI can digest large amounts of information and then distribute that data to multiple devices as opposed to a single database or entity.

The bitcoin frenzy has shifted attention to blockchain technology and its decentralized nature. Blockchain is increasingly becoming an accepted method for enabling users to share information and conduct transactions freely and securely. Currently, Walmart’s food safety solution has been working with IBM’s Blockchain Platform to bring transparency and efficiency to food supply chains, starting in China. Google’s Deep Mind is also utilizing blockchain to encourage transparency and protect personal health data used in their AI engines. The shared, decentralized data layer that blockchain provides is accessible to all stakeholders. It promotes the transparency and safety that organizations and consumers alike seek – something that is lacking in the age of the data breach. What’s more, blockchain technology, when coupled with AI, can open the door for decentralized authentication. AI has the capability to digest large amounts of information and share it through blockchain – rather than a centralized entity controlling where data is stored, often in poorly protected systems.

The many breaches that have occurred in the last year have shown us that passwords, pins and fingerprint authentication are no longer making the cut when it comes to our protection, which makes blockchain’s architecture so appealing. According to Gartner, fingerprints or “Touch ID,” the most popular mainstream method of authentication, is only around 75 percent successful due to contaminants such as dirt and sweat. But what if, in addition to blockchain, decentralized authentication could be made possible across mobile devices through AI? Living in a sensor-based world revolving around smartphones and other IoT devices, this will become a necessity. Several entities are currently working on bringing AI chips on-device to life.

Google is deploying AI to improve the quality of the Pixel 2’s camera, to compete with Apple’s iPhone X, which has two lenses while the Android phone only has one. Using a custom chip, Google is implementing a technology called RAISR to make zoomed-in shots sharper. RAISR uses machine learning that is trained on real photos to make intelligent decisions when it comes to filling in details. The chip accelerates processing tasks like running AI software and merging several photos into one “HDR+” image in order to prevent dark shadows and over-exposure. Meanwhile, the iPhone X uses a Neural Engine as part of its A11 Bionic chip, which incorporates machine learning and deep learning to recognize a user’s face, whether for animojis or for FaceID’s authentication.

Gartner forecasts that security technology combining machine learning, biometrics and user behavior are poised to reduce passwords to account for less than 10 percent of all digital authentications through 2022. Implementing AI on-device can help accelerate this process. AI and machine learning algorithms will go beyond recognizing physical characteristics such as the features on one’s face. This technology can be trained to quickly, and continuously, adapt to user behavior and build personalized models, learning everything about the way they interact with their devices from the way they type and swipe, to the hand they prefer to hold their device in. These traits will be analyzed to differentiate between true users and fraudsters. Hackers may be able to crack passwords and duplicate fingerprints, but the behavioral signature of a human consumer cannot be decoded.

AI on-device not only eliminates the use of passwords; it promotes decentralization. Rather than storing sensitive user data (such as names, credit cards, or how they use their apps) on a centralized server, having an AI chip on one’s smartphone or tablet enables data to be kept on the user’s device, encouraging user privacy. Companies like Google are currently working on a new method of AI training that allows algorithms to be trained on-device so that data used to improve their apps never leaves the device. Another benefit to users is that they will have everything they need on their smartphones to conduct transactions, meaning there will be no government or banking intermediary in the middle. Users will avoid potential skimming and identity theft, while financial institutions, retailers and online merchants do not need to store customer credentials, reducing the risk of a data breach.    

AI on-device will only continue to grow with the demand for better privacy and security. This decentralized approach will allow users to take back ownership of their personal information, while protecting them from major breaches – making it much more difficult for ill-intended actors to steal valuable details.