“As more devices become intelligent and connected the model of sending everything to the cloud will not scale. To address that, and also to improve privacy, security, and latency, AI will need to be distributed from the cloud to the wireless edge, including devices —with on-device processing, sensing, security and intelligence. That brings AI to the core of Qualcomm’s focus: the emerging wireless edge comprised of a massive number of connected devices, like phones, cars, robots, drones, sensors, meters, etc.”
What kind of research within AI is Qualcomm working on and why?
“AI touches numerous fields in Qualcomm. Qualcomm AI Research is focusing on model and source compression, computer vision with deep learning, and automatic speech recognition among other things. These research topics have numerous applications in autonomous driving, the Internet of Things, and further development of Snapdragon, Qualcomm’s signature system on a chip.”
What will surprise us the most in AI over the next five years?
“Oh sorry, my crystal ball is in the repair shop this week! Meanwhile, I dare making a few vague and wild guesses… We will have more powerful AI algorithms running on your smart devices, making things like voice recognition, language processing, and personal assistants faster and more power efficient. While the use of AI outside the cloud is mostly focused on inference today, we will see advances in training networks directly on edge devices, allowing for continuous learning. We expect breakthroughs in the generalization power of AI by turning parts of the networks on or off depending on the data or situation. Much like we don’t use all parts of our brain all the time, algorithms can turn on or off the parts that are needed. This enables networks to be more energy efficient, learning multiple tasks at once and with far less data. A step further will be AI algorithms that are better at learning from fewer examples and we expect to see AI being better at explaining its decisions.”
What is the biggest misconception about AI?
“Like with any new technology, there are a number of misconceptions. One of them is that we are approaching general artificial intelligence, the so-called singularity that is portrayed ominously in the media. We’re nowhere near AI developing human-like conscience. However, we think AI will make breakthroughs in many specialized applications in fields like diagnosis in healthcare, self-driving cars and IoT.”
Peter Rauber is senior Director Engineering at Qualcomm Research Netherlands.