Over the past year, the industry has made great strides in enabling the intelligence in the form of trained AI models that can be ported to anything from an applications processor used in a car or smartphone to a microcontroller used in a security camera or appliance. This is referred to as inference processing rather than the deep learning training that is performed on massive server systems in the cloud. And, there is significant value in making devices around us, usually referred to as edge devices, intelligent in terms of privacy, security, reliability, and efficient use of network bandwidth. However, OEMs appear hesitant to make the shift as seen at the Consumer Electronics Show (CES).
The industry was quick to adopt the “smart” moniker for everything from watches to homes to buildings and even cities, but the vast majority of all these “smart things” are really just connected things. The intelligence often lies in a remote cloud server. Smart speakers, smart doorbells, smart appliances, smart lights…are all just connected devices that leverage intelligent models in the cloud. So, you may ask, what’s wrong with this model?
Image a smart house that has smart doors, windows, AC, lights, appliances, etc. This can eventually add up to hundreds of smart solutions all connecting to a wide variety of different cloud services, depending on the OEM and/or service provider. In most cases, the consumer has no control over what data is collected, how often it is collected, how it is used, or by whom. The first issue this brings up is privacy.
Also, consider that each one of these internet connections is a potential hole in your network that can be exploited by hackers. That brings up the second issue – security. And, if your internet connection goes down, you may have no control over all those “smart things.” Even worse, imagine your lights coming on at full intensity after a power outage waking and blinding you in the middle of the night. This was the case with Philips Hue lighting, the issue has since been corrected. This brings up issue number three – reliability.
And finally, you have all these devices transmitting data continuously over your internet connection. If you have unlimited data, that can be just fine, but if you are on a limited data plan like those of us using satellite, then you have another problem, which is network bandwidth costs and data caps. Even consumers on an unlimited data plan may experience slower response times when data traffic is high or if your service provider chooses to regulate your data usage. How long would you be willing to wait for your lights to turn on?
Now, I can understand why most of the device vendors chose not to deploy more localized AI models. For one, the technology was not available when many of the IoT solutions were initially developed. A second reason is a combination of cost and complexity. But there is also a third and more concerning reason: they don’t want to give up access to the raw data, which can be more valuable than the device. Consider all the potential buyers for this data. Wouldn’t your electric company want to know more about your usage patterns, your health insurance company about how much TV you watch or games you play, or any consumer company about your interests and buying patterns? To be fair, much of this data is already collected every time you hop on a browser, purchase something over the internet, or stream content, but those are all conscious decisions by the consumer, not information generated from ambient devices around you.
I am very disappointed at the lack of device OEMs making solutions that are more intelligent at the edge. I believe those companies that do improve edge intelligence will ultimately be the industry leaders. Even Apple, who does not exhibit at CES, took a swipe at this issue with a billboard that says “what happens on iPhone, stays on your iPhone”. I’m not sure I want Apple to know everything about me either, but it helps make the point for more intelligent things, not just connected things.
This topic even led to a debate between my colleague Kevin Krewell and me during a meeting with a major technology provider that enables many of these devices. There is the case for sharing data that can be used for the greater good, like auto and traffic information that could potentially save lives in real-time or improve AI driving models over time. But, even if you could make that case for every application, the networks and cloud services could not handle the exponentially increasing amounts of data that these devices are creating. We also did a podcast to recreate this debate on the Smart Home that you can listen to on Sound Cloud.
So, my message to the industry is that there is a difference between a connected device and an intelligent device, and the latter is important. And to the staff at Ring that argued with me that the cloud is more secure than local intelligence, think again! I would also venture a guess that most of the current solutions and services are in violation of the EU’s General Data Protection Regulation (GDPR). The way forward is smart things that are intelligent, not just connected.
originally posted on Forbes.com by Jim McGregor