For those looking to the future of artificial intelligence on the smartphone, it’s clear that this combination holds limitless potential. True, on one level, a decade into the smartphone revolution, innovation has slowed down significantly. Each new phone brings a faster processor, prettier screen, higher resolution camera, the usual new features. But quantum leaps in functionality are less common than the smartphone’s earliest years.
If there is to be a revolution in smartphones, it is likely going to come with the advent of artificial intelligence in the device. AI has countless uses and applications, but in the case of the smartphone, the omnipresent nature of the device enables these uses still more.
Because you have your smartphone on you at all times, it can learn your behavior. Because people do so much business and ecommerce through their phones, it knows your interests. And thanks to sensors, it will know what’s happening with you physically and even mentally.
AI will allow your phone to track, interpret, and respond to patterns and behaviors that it recognizes as important to you. It will learn your route to and from work and warn you of an accident or construction. It will learn what restaurants you like and let you know of a special at that restaurant. Or it can even monitor your health and detect a heart attack.
“AI [on smartphones] is focused on assistants. [Phone makers] really want to do this digital assistant stuff,” said Will Stofega, program manager for mobile research at IDC.
There are already digital assistants, of course. Apple has Siri, Android has Google Assistant, Microsoft has Cortana and Samsung has Bixby. All do roughly the same thing: voice-driven searches. It’s just a matter of how well they do it, which Stofega said still isn’t all that great.
“Software assistants work out well except that what they are doing is searching for something and matching. The knowledge isn’t always accurate or can’t always get to the answer you need. What we will see is a big advance in the ability of devices to learn for themselves,” he said.
Jack Gold, president of mobile consultancy J. Gold Associates, says he’d like to take that notion one step further. “You are talking about a concierge sitting on your phone. It’s intelligence vs search. Today Siri and Cortana are search. It’s going through an encyclopedia and finding something as opposed to going through that info and making something out of it,” he said.
AI, Your Smartphone and…Data
For your smartphone to use AI to understand your interests and tastes and emotions – and prioritize the information presented to you – means the phone will need a lot more data than what it collects now.
The more data your smartphone collects, the more data it can make use of and the more intelligence can be extracted, but smartphones are finite, closed spaces. The latest generation of phones has only just now reached a minimum storage capacity of 64GB. For the longest time, 16GB was standard issue, and that was filled fast.
Not just that, but how much of your personal information do you want your phone to know? Some people might balk at having no secrets with their iPhone, because that means Apple might know, too, and who else?
“It’s all about the tradeoff. For AI to work properly needs to collect the data, build the algorithms, so that implies you may need to collect info you may not want collected. But you can argue that battle around privacy the barn door is already open. So the question is will it help me. I think it will,” said Stofega.
Chances are massive data won’t be processed on the phone any time soon. Mobile processors have advanced considerably since the iPhone’s introduction in 2007 but they still can’t handle the processing needed for real AI.
“AI on phones today is about back-end services,” said Gold. “It’s mostly about learning what I’m doing but it’s not taking place on the phone it’s taking place in the cloud. That said, AI is a hard to implement technology. if you are going to do anything significant you need a lot of data to model and you do machine learning. All of that is a very compute intensive process. There are no smartphones available today that can do real AI.”
Your Smartphone Knows Your Mood
In the 1970s, mood rings hit the market and caught the public’s attention in a big way. The rings had a stone that changed color depending on your mood. Their accuracy was debatable, and like any fad it ran its course, although they are still available online.
Argus Labs is a company working on technology that will turn your smartphone into a mood ring. It will know your mood and how likely it is to change. It’s also working on non-phone devices, including a radio player for cars that will pick a song to play next based on your mood. It could also be used by advertisers to provide products based on the mood of the listener.
It’s an advance over prior attempts at mood study. In 2015, a study by Northwestern University Medicine found that the more time you spend using your phone and the fewer places you visit, the more likely you are to be depressed. It tracked text messaging, app usage and location throughout the day. Participants in the study were also asked to rate their feelings of sadness via popup messages on their phone.
That reflects the difference between the current system and what AI can offer. The Northwestern study was based on manual pattern matching by humans. They looked at the smartphone data and came to a conclusion. The Argus Labs project will use the sensors in your phone to detect your mood without human intervention.
AI Evolution: Smartphone Hardware and Software in Tandem
In addition to the challenge of data processing, for AI to really work on your smartphone will require tight integration of hardware and software, an advantage Apple has over Android. Both Stofega and Gold say the two have to go together, a challenge since the Android OS is made by Google and licensed to OEMs.
“One reason behind the development of the [Google Pixel smartphone] is to really zero in and make a great AI personal assistant experience. They said they need to develop software and hardware in tandem to do this,” said Stogega.
The end game of the Pixel is not the Pixel phone but the ability to build out a smarter system. So it helps to do both the hardware and software. Google’s last smartphone effort, the Nexus, was well-received but it was Android on an OEM phone, and Stofega said Google were only involved after 90% of the hardware design was done.
“Can you do AI without integration? Yes, but to get the best optimal experience so everything melds together, you need that integration,” he said.
Gold thinks the situation is even tougher than just hardware and software. “In the early days [of AI] all the algorithms that are developed are going to be proprietary. Apple will do something, Google will do something, Lenovo will do something and they may not be compatible. They will be unique. It’s the database and OS wars all over again as it is in any new marketplace,” he said.
Chip Rush for AI on the Smartphone
This has led to a gold rush in the AI chip market as VCs line up to throw tens of millions at startups and the big companies acquire the ones with the most promise. Intel, desperate for new business with the PC market saturated and no longer growing, has been the most aggressive. It paid $16.7 billion for programmable chipmaker Altera, $15 billion for driver assistance company Mobileye, and $400 million for AI software and hardware startup Nervana.
Others are growing organically. Nvidia, the leader in GPUs, has focused on applying its GPU technology to AI processing. It has already made significant inroads with GPUs in high performance computing so it knows this field well.
Google made headlines with the announcement of the Tensor Processing Unit, or TPU, which it claims offers a 15-30X improvement in processing power over traditional CPUs. And Qualcomm has its own projects, along with the acquisition of a machine learning startup called Scyfer.
This activity leads Gold to believe more AI systems will move into the phones. “With the progression of chips, eventually some of that AI processing gets down into the processors. But you are still a few years away and you are still going to need machine learning on the back end. We’re just going to transfer some of the algorithms down to the phone,” he said.
But he adds AI – on the smartphone and elsewhere – is in its very early stages. “Anyone who thinks AI is not a work in progress doesn’t know anything about AI,” said Gold.