Edge computing: should a car's AI think on board or in the cloud?

A car at ninety kilometres an hour. A pedestrian suddenly steps out. In a tenth of a second, the car covers two and a half metres. Should its AI think in the car, or in a distant data centre? In 1998, MIT researchers founded Akamai, to serve web pages from distributed servers closer to internet users. In 2009, researchers proposed placing a small data center near the phone, one network hop away, for faster responses. In 2016, an IEEE Internet of Things Journal paper laid out edge computing: processing data close to its source, along with its challenges. That's edge computing, in three ideas. One: the trip costs time. In the cloud, data leaves the camera, crosses the network, gets analysed in a data centre, and the answer makes the whole round trip. For a web search, nobody notices. For braking, it's critical. Edge computing does the opposite: it processes data where it's created. Two: fit the AI into the device. A model is millions, even billions, of numbers: its weights. Quantisation rounds them from thirty-two bits to eight: a quarter of the memory. Another technique, pruning, cuts the least important links. The model then fits on a chip, like NVIDIA's Jetson or Qualcomm's Snapdragon. Three: share the work. Urgent and sensitive tasks stay on site: answers in milliseconds, and the data never leaves. The heavy lifting goes to the cloud. The right place depends on the job, not on fashion. At KOAEE, the same question: where does the data go? Our assistant runs in the cloud, a cloud of our choice: our servers are in Frankfurt. From the farm shop to the estate agency, answers appear in under three seconds. For a conversation, it's perfect. For emergency braking, you need the edge. To understand AI in 150 seconds, subscribe to the channel. The AI that stays between you. Try it at koaee.ai.