Devices with combined cloud technology and neural networks integrated into their hardware – to reduce potential stability, privacy, and latency issues – are now a reality.
One of the reasons for the incorporation of artificial intelligence (AI) into smartphones is to achieve a more natural interaction between the device and its user. Giving devices the ability to understand and decode meaning contained within audio, video, and image files both enriches the user experience and contributes to the emergence of new value-added services. For example, if a phone detects a mood by analyzing the tone of voice, information about user sentiment can be used to personalize how the phone works and what content it provides at any given moment.
Smartphones empowered with artificial intelligence rely on one of two approaches:
Method 1. The device acts like a passive or “dumb” terminal and is limited to sending data to a cloud-based service, which in turn responds appropriately. In this case, all the intelligence and required processing takes place in the artificial intelligence provider's DPC (Data Processing Center).
Method 2. The device itself is able to intelligently process data; therefore, there is no need to send data over the Internet.
Each of these two options has advantages and disadvantages. On the one hand, the first approach does not require significant processing capacity in the terminal. Having a good Internet connection suffices. However, there will be a delay in the response because time is needed to upload the data, complete the processing in the cloud, and receive the results over the web.
The second option requires having a device with sufficient processing power, so the results can be calculated locally. If the device is powerful enough to perform the processing, this option avoids Internet traffic and associated latency, and thus achieves a smoother response and user experience.
To date, the first method has been more common, since devices haven't been equipped to efficiently execute these complex algorithms and the only suitable hardware were servers found in big data processing centers. "But with the current boom of applied robotics and cognitive services, there has been an emergence of smartphone chips specifically designed to perform artificial intelligence tasks. So, it will be increasingly common to have a mobile device equipped with specialized hardware for this kind of activity,” says the cognitive neuroscientist Raúl Arrabales, director of Artificial Intelligence in Psicobotica.
Accordingly, there are already smartphones on the market with integrated neural processing units (NPU) embedded in their chips. This transforms these devices into hybrid terminals enabling artificial intelligence by combining the power of the cloud with the speed and response immediacy of native processing.
Advantages of AI hybrid
Factors like latency, stability, and privacy are improved by the combination of artificial intelligence in the cloud and AI that comes with the hardware of the device itself. As Arrabales points out, if the terminal device only uses cloud-based services for this functionality, the response will not only be slower, but there will also be the added risk that data is transmitted over third-party infrastructure for analysis. "From the standpoint of privacy and response speed, as users it is in our interest that our data is analyzed locally in the device.” the head of Psicobótica stated.
Nonetheless, we have to bear in mind that many technology service providers like Google, Microsoft, and Apple are interested in having access to data because the quality of their artificial intelligence systems largely depends on being “trained” with user data.
"The combination of both scenarios will result in more possibilities," according to Arrables. Looking forward, the ideal scenario will see developers deliver services that keep the advantages of the two approaches without the drawbacks.