Mobile OS maker Jolla is back and building an AI device

On the fringes of MWC, the world’s biggest mobile phone show, TechCrunch is getting a demo of a new kind of AI device. The small black box sitting in front of us on a table in a hotel lobby in Barcelona is designed to tap into the generative AI boom — making it another example of how the wheel of innovation is spinning up again in mobile.

Its maker — Jolla, a born-again startup with the combined forces of around 40 seasoned staff behind this new push — is building what’s being tentatively described at this early product design stage as a private and secure “blackbox for your life”.

The device is intended as a private-cloud-cum-AI-router that makes the user’s data accessible for running AI queries. So this device-in-the-making is akin to the personal server concept that’s been kicking around in privacy circles for years — but reconfigured for the generative AI age.

Here at MWC 2024 we’re hearing a lot of chatter about “next-generation” smart assistants, with telco talk rife with predictions that the end of apps as we’ve known and tapped on them is nigh as generative AI models rewrite the user interface rules. Jolla is jockeying to insert itself into this buzzy conversation by putting the focus squarely on privacy and security — since the assistive data processing will take place on the AI device itself.

“We don’t believe it’s a good model to put AI run locally here in the phone. It’s not secure,” argues co-founder Antti Saarnio. “You can never make smartphones secure enough for that. We believe that in this AI era privacy becomes much more important.”

The machine learning databases Jolla is architecting to host the user’s personal information inside the AI device are being designed to intelligently respond to natural language querying and/or proactively surface insights — drawing on trusted access to the user’s mobile data and digital services.

Jolla is building on open source AI large language models, including Meta’s LLaMA 2, for the foundational layer of generative AI smarts. Saarnio says the role for the team is basically architecting the user interface and function design — rather than needing lots of AI engineers itself, as they’re able to draw on powerful open source general purpose AIs.

A private data hub for AI

Discussing the vision for the product, Saarnio argues that amped up AI models, such as large language and multimodal models, and the powerful data mining and inferences associated with generative AI, change the game when it comes to processing personal information — whether it’s emails, calendar, contact lists or obviously sensitive stuff like health data — bringing both the potential for AI-powered insights to create huge value and utility for individuals; but also deep risk if the user has to offer up intimate access to their information in order to tap into these services.

This trade-off is the problem Jolla wants to fix by designing a way for users to gain the superpower of advanced AI without having to let data-hungry third parties gobble their privacy.

“This device will hold all your private information,” he says, laying a hand on the small rectangular box we’re discussing. “It runs its AI here. All your data will be here and nobody else will have access to it — only you.”

AI is increasing data concentration and leading to the development of rich databases of personal information that can be “dangerous” if exported, he suggests. So what if people could just own their own safe and secure hub where all that processing happens?

Assuming Jolla’s product is as safe and private as they claim it will be, the level of data protection a user might gain would depend on how much AI processing they’re willing (or able) to route via the device. But if they practice good privacy hygiene generally such a product could help people limit their privacy risk when using generative AI.

“If people use different email clients, and so on, we need to connect to those applications via our API,” Saarnio notes. “But the key here… is we’re saying that your concentrated data is in safe place.”

“Surprisingly, we haven’t seen anyone who is really focusing on data privacy and keeping the data secure,” he adds. “I think it’s a good place for us to play.”

The Finnish mobile phone startup tried very hard to disrupt the traditional mobile smartphone business a decade+ ago, with their alternative, Linux-based Sailfish OS platform. It’s that same familiar brand name stamped in tiny letters on the 3D-printed prototype. But the product represents a leap in a new direction for the mobile also-ran.

Saarnio, an original Jolla co-founder, has returned to what his LinkedIn account simply refers to as “New Jolla” — reenergised and raring to go after winning a long legal battle to reclaim the business and its assets from majority Russian owner, Rostelecom. The mission: Rebuild Jolla anew by building new AI products. “I think we can completely rebuild this. But we need to do new things,” he tells TechCrunch.

The Jolla brand name still has value, as Saarnio tells it. He talks in terms of co-designing the product with the community of Sailfish users and Jolla enthusiasts still excited about what alternative tech can do. But — to be crystal clear — the AI device won’t need special mobile hardware to function. The goal is for this to piggyback on top of mainstream mobile platforms.

Adding a layer of security and privacy to platforms and apps is exactly the point. The broad ambition, if Jolla can pull in the users and build momentum, is to develop the product into a platform of its own — further extending the embedded AI assistant’s utility by attracting other developers to build apps that operate through its protective wrapper of privacy and security.

“That’s our only playfield,” admits Saarnio of the focus on privacy and security. “Obviously, big companies will do similar things… [But] we can become an open application platform as well for others who are focusing on [AI innovation] so we don’t need to do the lifting only ourselves.”

He also predicts big changes incoming for mobile apps over the next few years, driven by AI. “It will become almost like a protocol level,” he suggests, describing a scenario where mobile apps are pushed into the background as generative AI interfaces take over. “You [won’t] really go and open apps anymore. So this whole visible UI level is layer is disappearing.” 

“I don’t know what it means. Basically, nobody probably yet knows, but it’s a major change,” he adds, arguing Jolla is well positioned to take advantage of the shift give its expertise in operating system development. “An operating system player like us, we have a good role there. Because our job is to integrate with different applications. Even the app runtime that we have is quite good for that. It gives us access to integrate into different applications, their APIs and so on.”

Jolla Mind²

The prototype sitting on the table currently bears the name Jolla Mind² — a play on the idea of AI providing humans with a ‘second brain’. Jolla also described the system as an “adaptive digital assistant”.

The main target customer looks likely to be Android smartphone users as it could offer them a way to tap into the utility of advanced AI without having to expose all their personal data to Google (or other non-trusted third parties) — with all the associated risks. (Google remains an attention-mining advertising company, after all. While Microsoft, a major investor in OpenAI, has its own digital advertising business. And Facebook-owner and LLaMA developer Meta’s business model is still pure surveillance advertising.)

On the flip side, Apple already offers its own privacy promises and on-device AI processing. It also has its own generative AI products in the works. So value certainly looks harder to build on iOS (where this kind of functionality is likely to be natively embedded). But what Apple will do is not giving Jolla pause. Android remains the dominant mobile platform, with billions of users worldwide, so there’s more than enough of a potential addressable market to target — especially if advances in AI trigger a rise in privacy concern.

“Very large companies nowadays are very tempted to use our information for AI training — because that’s the name of the game at the moment. So I wouldn’t trust any company [not to use my data for AI] training,” says Saarnio. “If the future would be similar to the past — only a couple of massively big companies were able to take the whole market with AI — then I think it’s not very good news for us human beings.

“In this past mobile era, we were the data product, basically. But if we let AI steer us we become digital robots, basically — the manipulative capability of AI is so big that we don’t have a chance. So that’s why I think we need digital clarity. And that’s our role.”

“Basically, you need AI. But you need to be mastering the AI so that the AI is not mastering us,” he adds.

Saarnio himself built the first product mock ups, after spending the last 1.5 years learning to program — upskilling from his former finance-focused background to close in on the AI opportunity.

He describes Meta’s decision to open source LLaMA as the “break point” for this new wave of AI innovation. And is — necessarily — hopeful there won’t just be a repeat of the last mobile cycle where Big Tech took most of the spoils. “I think it will be difficult for big players to own this agenda,” he predicts of generative AI. 

During a brief demo of the JollaMind² prototype — the first time it was shown to press, we’re told — we saw Jolla’s AI assistant software running on a mobile phone which was interfacing with AI systems and databases running on the prototype hardware. The developer running the demo posed a couple of natural language questions and received contextually relevant responses back, voiced by a rather deep-throated robot voice, which added to the spooky thrills (so Jolla’s app is doing speech generation too).

In one of the demo examples, the developer asked the software “who is Natasha?”. Jolla’s AI assistant responded by parroting my TechCrunch biog — without needing to be provided with additional context since the AI has access to calendar info, emails etc where it could infer the subject of the enquiry from just a first name.

In another demonstration, the developer asked the AI “what tasks do I have for Tuesday?” — and got a read out of their calendar items with observations about the priority of different items.

Both examples are fairly utilitarian, in terms of what AI assistance features can do, but the point is the user of JollaMind² can rest assured their information remains safe from prying third parties.

Another envisaged use-case Saarnio sketches is email triage: Where the AI assistant could keep tabs on incoming emails and surface priority messages, including knowing whether to override a ‘do not disturb’ setting to make sure something important isn’t missed.

While such AI-powered email parsing functions do already exist (and can even be baked in by email providers themselves, as Google does with Gmail ‘smart’ features), you have to be comfortable providing a cloud-based entity with access to your inbox to tap into such smarts — which is a privacy and security trade-off.

Jolla also sees potential for business users — such as a smart minutes type feature. Saarnio gave the example of lawyers meeting with clients who could ask them if they could record the meeting to make it “memorable” — meaning it would be captured by the device and a searchable record of when (and what) was said retained to aid productivity but with all the sensitive data remaining on the private cloud device.

In instances where a user might actively want to go outside Jolla’s system in order to make use of a specific third party AI, the product could pop up a permissions screen for them to approve sharing their data — so the user could manually accept extra risk on a case-by-case basis. The device could also let users configure which kind of questions they only want processing locally (health queries, say) and which ones they’re happy for the device to cast a wider net out to third party AIs.

The sneak peak we’re getting is just that: A glimpse of what’s coming down the pipe. The final form-factor, features and specs — even, for example, such as whether it might have a screen, built-in speaker, its own battery or keyboard, or stay as a minimal black rectangle — remain tbc. So much of the product design and detail — and even core utility features — is yet to be nailed down. But Saarnio does say the device will be priced below €300.

He also points out the hardware can act as an upgrade for a user’s existing mobile phone, saying it can bring advanced AI processing capabilities to even mid-tier mobiles. So an old smartphone you have lying around in a drawer could get a second lease of life with Jolla’s AI layer bolted on top.

Whatever form the device finally takes, Jolla is envisaging it as something generally static — rather than a second mobile device you have to carry with you. So the likely scenario is it gets stored safety by the user at their home or office. But it’s available to be queried remotely through the user’s own mobile device (with, likely, a layer of biometric security looped in to ensure the dynamic database of their life it contains remains safe).

Back to the future

So what happened to old Jolla? Wind back the clock to August 2021 and the company as it then was had just turned profitable after a decade of hard slog selling its alternative OS and app support platform. Then, in spring 2022, disaster struck as Russia invaded Ukraine, putting Jolla’s non-Russian shareholders in a bind and desperate to cut ties with Russia.

Saarnio says company shareholders tried for a long time to negotiate with Rostelecom directly to buy back its majority share. However the Russian telco would not budge. So, in the end, they changed tack to litigation — and a court adjudicated corporate restructuring process in Finland enabled them to buy back the old assets. The court decision on that restructuring came around October last year — handing the team a clean slate to take a fresh swing at mobile innovation, now with AI front and center.

The new incarnation of Jolla will be fully focused on AI devices. But the legacy story doesn’t end entirely. Separate “independently fundable” business entities are being established — one to look after Jolla’s tech focused on in-car systems and app support for car makers, and another for Sailfish mobile devices for businesses, per Saarnio. But he and his team will be fully focused on rebuilding Jolla as a new, and they hope smarter and more agile, AI startup.

Mark your calendar: May 20 will be the official announcement of Jolla’s first AI product. The date has been picked as it’s a decade on from the “Jolla Love Day” announcement of May 20, 2013, when the Sailfish maker revealed its first ever mobile phone. Which is a neat way to underscore the cyclical nature of technology.

In just under three months’ time the exact shape, specs and feature set of the AI device will be clearly defined and a pre-ordering campaign will likely kick off at that point too. As for shipping, Saarnio is confident they can get the JollaMind² to market later this year. Expect Jolla’s new baby to arrive in Q4.