Will Edge AI save drivers from hitting pedestrians?
Artificial intelligence (AI) is often criticized for its enormous carbon footprint and privacy issues, as the applications are run on the cloud. Edge AI developers and supporters claim it solves these sore problems.
According to Facts & Factors research, the Edge AI hardware market is anticipated to reach around 2160M units by 2026. Intel, NVidia, Samsung Electronics, and Huawei Technologies are the top market players.
Recently, San Diego-based Edge AI company Kneron raised additional "eight figures" funding from Foxconn and Winbond. The neural processing unit (NPU) that the company developed runs sophisticated AI applications without relying on the cloud. Kneron's founder and CEO Albert Liu claims that Edge AI solves data privacy issues and diminishes environmental concerns that surround AI.
With Edge AI, algorithms are processed on hardware devices locally instead of sending data to the cloud. It saves energy and time, crucial, for example, when a car tries to define an obstacle in front of it.
“Cars need to send the image or the video to the cloud, do the recognition, and send the response back to the vehicle. So it will cause some problems, for example, if there is an internet delay, the risk for you to hit the pedestrian or the car will be higher. In some applications, especially for application speed, Edge AI has great advantages,” Liu explained.
Kneron designs chips for a wide range of uses, from autonomous driving to smart door locks, payments, and surveillance. Kneron’s work with Foxconn will focus on the automotive industry through Foxconn’s MIH open platform for electric vehicles.
When we talk about AI, a huge carbon footprint comes to mind. One recent example of colossal emissions is the IBM Debater – the first AI system that can debate with humans on complex topics. Project Debater digests massive texts, constructs a well-structured speech on a given topic, delivers it with clarity and purpose, and rebuts its opponent. Debater was able to perform very well against Harish Natarajan, who holds the world record in many debate competition victories. The problem is that it consumes a lot of energy. The machine that was used to run the project debater required almost a hundred processor cores and one terabyte of memory. For this one debate, it would consume more than a hundred times the power of the human brain.
With Edge AI hardware, when computations are done locally, power consumption doesn’t seem to be a problem. For example, a smart door lock with an Edge AI chip can run on a simple battery and last, according to Liu, about a year.
Manuel Le Gallo, a researcher at IBM, explained that most inefficiencies arrive from the architecture of our computers. Standard computers are based on von Neumann’s architecture, which means that computing and memory units are physically separated. Moving data from the memory to the processor consumes 100-1000 times more energy than a processor operation. To solve the problem, IBM has developed mixed-precision in-memory computing. The bulk of computation is done in computational memory units, and that results in lower power consumption.
It illustrates that local computation, when data doesn’t have to be sent anywhere, reduces power consumption.
The thing is that Edge AI chips can run only AI tasks. Just as the CPU (central processing unit) is designed for performing logic operations, GPU (graphic processing unit) is meant for graphics, such as games, and NPU (neural processing unit) is designed for AI tasks.
Edge AI chips are praised for their contribution to privacy. As US citizens are heavily surveilled through CCTV cameras, there are fears that smart home devices, such as smart door locks that we willingly buy, can be turned into surveillance cameras, for example, by the police. Surveillance can make any of us less confident to access healthcare, to join a protest, to go into certain areas, and can even make our normal behaviors like waiting for a friend or putting our hood up appear suspicious
A smart door lock with an Edge AI chip, Liu claims, solves that problem.
“Cloud AI means that everything you detect is sent to the cloud. Privacy is a big issue. Let's say someone is hacking into the cloud system. They can watch and monitor everything that happens in your home. Edge AI means that there's a small brain in your device. All the recognition is done by the Edge. It won't connect to the cloud all the time. Whenever some accident happens, they will only connect to the cloud sending the warning message instead of sending the video or image to the cloud. It is a major difference,” Liu explained.
Moreover, with Edge AI chips, any device can be turned into an AI device. You can plug in the chip like a USB key, download specialized AI software, and, for instance, enable your computer with facial recognition capability.
More from CyberNews:
Subscribe to our newsletter