Data from IMARC Group puts the value of the smart speaker market at $5.08bn last year, and it is expected to hit $21.94bn by 2027. Such widespread adoption is laying the foundations for the smart home of the future.
People already demonstrate a high degree of trust in voice control/interaction. However, as more and more people buy into the smart home and devices proliferate, there is an ever-bigger price to pay when it comes to data security and privacy.
The balance between convenience and privacy
To put it another way, Alexa might be helpful when she turns off the lights, but the reliance on a permanent link to the cloud means that she is also leaving the back door open from a privacy point of view.
That is not just a technical fact, it is also an ethical question. It is certainly convenient for your devices to be listening for their particular wake word. But it is also a major consumer concern that every audible event in their house is being captured, digitized and streamed to the cloud.
Of course, this mechanism isn’t just borne out of technical necessity – it is also fuelling the ad-based business models of Amazon, Google, and others. However, this might prove ultimately self-defeating. There is a much lower ceiling to the smart home market if consumer concerns around data privacy cannot be solved – not just for today’s smart speakers, but also for whatever devices come next.
What this means is that there is a technical balance to be found if the smart home is to truly thrive. Without the ability to recognise people and respond to commands it is difficult to see what the smart home is “for”. At the same time, we must avoid a situation where people feel like they are constantly under surveillance in their own home.
How we find that balance is very much the million-dollar question for the smart home industry right now.
Enter edge AI
The key to delivering on a more private smart home is to make devices more intelligent in and of themselves.
TVs, soundbars, smart speakers, and even remote healthcare monitoring devices all have one thing in common — they all want to become “smarter”. But currently the only way to do that is through the cloud. Commands and signals given to a smart speaker are not processed by the device. Instead, the data is transmitted to the cloud for interpretation, contextualization and then instructions and actions are sent back to the speaker.
There is an alternative to the cloud-based IoT – the Artificial Intelligence of Things (AIoT). The AIoT model involves putting intelligence and processing power directly in the end point device – enabling the device itself to interpret and action commands locally – cutting the cord with the cloud.
The problem is that delivering this edge intelligence has been easier said than done. To date, the chips that can deliver this intelligence are expensive, difficult to work with and time consuming to design into products.
Delivering the AIoT
There is no doubt that the chip design challenges of the AIoT are significant. To end the reliance on the cloud requires an entirely new type of processor that brings together AI, DSP, control, and I/O in a single device, versatile enough to let the designer determine the balance between the four. All of that must also be delivered in a small package and with a low overall BOM cost.
Creating such an infinitely programmable device with fast processing and neural network capabilities is no small task. But it is essential to smart home privacy. It enables collected information to be processed locally while keeping personal data secure and executing actions almost immediately.
It is also worth noting that having this sort of intelligence freely available at the edge of networks will make it viable to add more sophisticated capabilities to smart home devices. With a new class of processor, presence detection, face and image identification, and even life sign monitoring can be added to devices – capturing rich, contextual data to build an intelligent understanding of the operating environment within a closed loop system, without the need for the cloud, or indeed, without exposing data to the cloud.
Of course, severing the link with the cloud isn’t the only requirement for secure smart home devices. Advanced security features including secure boot, one-time-programmable key storage, true random number generation and custom security instructions are also crucial to protecting consumers’ data.
Security and privacy are non-negotiable for the smart home
Making devices capable of processing data locally, and reacting to the results based on local AI, would represent a huge stride forward in data privacy.
There is an argument that says this will be too much of a trade-off compared to the personalisation and user experience that cloud-based systems currently enable. But it is not the case that personal information is required to deliver smart home services. Visual and audio sensors, enabled by edge AI, are more than capable of determining the difference between a child’s voice and a parent’s – enabling the device to ignore commands from children to turn the oven on, or to order age restricted items.
Over time these capabilities can scale up to enable very sophisticated functions without transmitting any personally identifiable information to the cloud. For example, visual and sound sensors could work in concert to observe a room in which someone injures themselves in a fall, alerting the emergency services as a result.
Delivering on this more sophisticated vision of the smart home requires devices that have the intelligence and the collective sensor array that is capable of painting that picture. The shift towards privacy, and prioritising AI-enabled sensors over data collection, is the absolutely crucial pre-requisite to making this version of the smart home a reality.