There is one serious and obvious to all problem that currently prevents the use of neural networks in mobile devices: power consumption. Many of their artificial intelligence systems require powerful, multi-core processors. Their use is extremely impractical when it comes to a device that is supposed to be placed on a human hand.
However, Massachusetts Institute of Technology has found a solution. It recently demonstrated Eyeriss, a new chip that enables the installation of neural networks in extremely low-power devices. Although this processor has 168 cores, it uses ten times less electricity than the GPUs in modern phones. Thus, you can safely use it without fear that the processor will instantly "eat up" the battery.
The secret of Eyeriss is that the chip does not use information exchange whenever possible. Each of the nuclei (which essentially act like neurons) has its own memory and compresses information every time it is sent. This keeps the amount of work to a minimum. Neighboring cores can "communicate" directly with each other, so they do not need to access a central source (that is, main memory), since everything they need is "at hand". On top of that, a special delegation scheme gives the cores as much work as they can get done without having to ask for additional data.
While there is no information on how soon Eyeriss technology can be expected to appear in any commercially available devices, it can be assumed that its impact on self-learning computer systems will be huge. Users will be able to get smartphones (or any other low-power devices) that get by with local AI processing instead of having to communicate with a remote Internet server, which is associated with latency and security concerns. Many of the devices available to consumers will be better able to adapt to new situations or navigate in space. It is also important to mention that one of the leading researchers from NVIDIA's company participated in the development of the new chip, from which one can conclude thatthat the new technology will have practical applications in the foreseeable future.