According to the latest research from Strategy Analytics, one out of every three smartphones sold worldwide this year will use artificial intelligence (AI) to power virtual assistants. But cloud-based AI which powers virtual assistants results in slower response times – On-device machine-learning will significantly speed smartphone AI.
Ville Ukonaho, senior analyst at Strategy Analytics, said “Google has a narrow lead in total smartphones sold with onboard virtual assistants in 2017. That lead will only grow as Android smartphone sales, with Google Assistant onboard, continue to expand into lower price tiers.”
Virtual assistants have already become common in premium tier smartphones. In 2017 already over 93% of premium tier smartphones with a wholesale price above US$300 (€257.88) sold worldwide have a virtual assistant integrated out-of-the-box. The penetration will expand fast to lower price tiers, mainly with the help of Google Assistant. In 2020 we estimate that over 80% of the smartphones sold with a wholesale price of over US$100 (€85.96) will have a virtual assistant integrated natively.
Exhibit 1: Global Virtual Assistant Smartphone Sales % Share by Type: 2017 & 2022
As AI-powered virtual assistants become common across smartphones, the speed with which they are able to accomplish tasks and return results will become key differentiators. AI is at a nascent stage with lots of improvements and lessons to be learned. One of the major issues for AI powered virtual assistants right now is that very little of the actual computation that powers the assistants is done on the phone itself. According to Ville Ukonaho, “AI is computational intensive and most of the heavy lifting is done in the cloud. This requires a solid data connection, something that isn’t always available.”
Over the last few years, since the first virtual assistants on smartphones emerged, beginning with Siri on Apple’s iPhone, most of the data processing has been done in the cloud, due to the limited processing power of smartphone CPUs. AI applications require huge amounts of data processing, even for small tasks. Until recently smartphones did not have the computational power to handle this. However, recent advances in smartphone operating systems and related software and components have brought increased processing power to the newest flagship smartphones.
Exhibit 2: Intelligent Virtual Assistant Landscape
Ken Hyers, director at Strategy Analytics says that “A number of vendors have created more advanced processing engines or are combining the power from the CPU, GPU and DSP to form a subsystem capable of handling complex machine learning and other computational AI tasks.” However, only high-end flagship smartphones will have these advanced AI processing engines and subsystems, meaning that on-board AI-powered virtual assistants will become key differentiators for premium smartphones.
“Neil Mawston, executive director at Strategy Analytics stated that “By combining software enhancements through machine learning and hardware enhancements in the form of AI engines, we can expect the abilities of virtual assistants to improve significantly over the next several years. This will result in increasingly responsive virtual assistants and more interactive experiences from the devices.
Key questions which still need to be considered include:
- Which vendors will drive AI to the all important mid tier?
- Which segments or clusters will pay for or value AI features?
- How strongly will AI contribute to brand purchase consideration? For how long?
- Will AI become table stakes?
- How will AI on smartphones interact with the smart car and smart home use cases?
The full report, Global Artificial Intelligence Technologies Forecast for Smartphones: 2010 to 2022, is published by the Strategy Analytics Emerging Devices Strategies service, details of which can be found here.