Decentralised, transparent AI solution from Xayn solves privacy-convenience dilemma.

Instead of having to rummage through countless documents in search of the term “Berliner”, Google provides the information we require: The algorithm knows exactly whether we mean a resident of the German capital or whether we are interested in a doughnut recipe. Netflix's AI knows our taste in films better than our partner and Facebook reminds us when our favourite band has a concert without being asked. But such convenience usually comes at a high price: "Most Silicon Valley companies such as Google, Netflix and Facebook do not create convenience with merely a nice design, but with the help of their users' data," says Leif-Nissen Lundbæk, CEO and co-founder of the Berlin tech company Xayn.

Location, payment data, contacts and usage statistics - data like these are collected by the technology providers and linked to the users. This leaves many feeling somewhat uneasy: According to the 2019 Cyber Security Insights Report, 92 per cent of users worry about their digital privacy. And the fear that their data will be hacked, sold to third parties and used for purposes other than originally intended is justified. An example of what happens when things go wrong is shown in the Cambridge Analytica scandal. Nevertheless, 64 per cent of users accept these risks in return for a better user experience.

^{Xayn CEO Leif-Nissen Lundbæk © Xayn}

Data protection or user experience?

For Leif-Nissen Lundbæk this comes as no surprise: “Privacy should not be a distinguishing feature, but it is. I often compare it to sustainability.” The customer would prefer a product that protects their data - provided they do not have to pay extra for it or compromise on quality. But that is exactly the case at the moment. “You often have to choose between user experience and data protection,” says the Berlin resident, describing the dilemma. The American search engine DuckDuckGo does not collect any data from its users, but it delivers poorer search results. The user pays for privacy protection with his time, because it takes longer to find the right results. Lundbæk is far from content with the status quo: “We can solve this trade-off between data usage and user experience with a technology that reconciles the two,” he says, “[a technology] that is just as convenient as Google, but also gives the user maximum privacy protection."

Xayn search engine: "Web search the way it should be".

In December 2020, the start-up Xayn launched a new search engine that allows internet users to have the best of both worlds: data protection and user experience. Founded in 2017, by Lundbæk along with co-founders Felix Hahmann and Professor Michael Huth, the company's eponymous search engine offers "web search the way it should be". The free search engine, which is available for Android and iOS, gives users the option to enable the AI for personalised search results. It also allows you to curate your content with one click: a swipe right confirms the search result is useful. If you swipe left, you decline. This allows users to influence what they will be shown in future. They can also save and sort their preferred content in collections. Users not only have full control over the AI ​​algorithms and content displayed, all data and interaction history remain in their hands - or rather on their own device.

The Xayn app features an AI infrastructure that combines federated or decentralised machine learning with homomorphic encryption. It was developed in 2014 as part of a research project at Oxford University and Imperial College London. The resulting framework that emerged four years ago has now been published as an open source project under the name Xaynet. "If you want to learn from data, you need large volumes of it," said the mathematician and software engineer Lundbæk in an interview with Datenbusiness Podcast. "Normally you bring the data to the algorithm for this. We turned it around: We bring the algorithms to the data and then collect all of the knowledge, not the data itself." The behaviour of the users is already evaluated on their smartphone. The generated AI models are then encrypted and transmitted to a server and incorporated into a global AI model, which is then transferred back to the end devices for improvement and correction. "This means we can ultimately decrypt the global model without ever being able to decrypt the local analysis," says Lundbæk, who now leads a 28-person team. "We don't see anything from the individual, only the total result of the all local users." To ensure that only the user has control over his AI model, Xayn also uses other decentralised technologies. The IOTA Tangle serves as an optional "trust anchor" to validate the integrity and transparency of the search AI.

Data protection: EU at the forefront of technological innovation

The benefits of the federated approach - built-in data protection and privacy protection as well as lower costs for local AI models - sound convincing. Central models still prevail, but the tide will turn: the EU Data Strategy forecasts that by 2025 80 per cent of worldwide data will be processed on end devices. European companies in particular "have no other choice but to switch," says the CEO of Xayn. The legal framework in the EU is already moving against the business models of Facebook and other tech giants designed to resell data. That is why data protection is “simply essential for survival.” The strict General Data Protection Regulation is not a disadvantage for companies: “Universal values ​​such as data protection are our top export to the world,” advocated Lundbæk at the DigitalShift conference in Munich in 2019. "Innovative developments in the field of AI that endorse these values ​​could put the EU at the forefront of technological innovation.

Berlin success story

From its headquarters in Berlin, Xayn's decentralised AI infrastructure gives it the potential to compete globally. "We think that Germany plays a major role as a privacy driver", which is why Lundbæk is convinced that choosing Berlin as a location to launch the start-up in 2017 was the right choice. Since Brexit appeared an inevitability in Great Britain at the time, he decided to search with "great foresight for a Central European alternative". As luck would have it, Felix Hahmann, one of the founders of company, was then-employed at the Daimler automotive group in Berlin, so the choice was not difficult. And Lundbæk has certainly not regretted the decision: "All the developers are leaving Oxford, where we were initially, to move London or the home countries," he explains. "We always wanted to have all our people in one location, and Berlin has the advantage that everyone wants to move here." In addition to access to the city's vast talent pool, the open community and good cooperation between companies attract many tech start-ups to the capital. "On top of this, many investors are looking to Berlin," says Lundbæk. Xayn, which was initially financed by the founders themselves, also benefited from this. After two financing rounds, the venture capitalist Earlybird, Business Angel from Asia and Dominik Schiener, inventor of the crypto currency IOTA, have become investors in the company.

Great little search engine

Xayn's strong investor base enabled it to develop solutions for industrial partners such as Daimler and Porsche, and at the same time devote a year to refining the search engine. The strategy worked: "After just one month we already have around 15,000 active users and a relatively strong daily growth," says the proud CEO. "So among the small search engines, we have become a relatively big player right from the start." For the first quarter, the company aims to gain 100,000 users, primarily from the European market. A browser version is expected to be added to the app version in the first half of 2021. In addition, the search feature will be boosted with other sources such as a product search function. “In addition, we want to provide a search engine for companies because they have problems obtaining relevant results,” says Lundbæk. Civil servants spend no less than 37 per cent of their time searching the Internet, according to an internal study conducted in a major German city. This is an area where the AI ​​can help authorities become more efficient. Testing of premium features for end customers is also planned: "The fact that the AI ​​operates on the device makes it easy to add sources," says the CEO. "So you can also search in the cloud, for example."

From federated learning to federated analytics

According to the Berlin entrepreneur, 80 per cent of Xayn's activities are focused on the search engine. In addition, the infrastructure is continuously updated: "The question of how users use our product is very important," he explains, "Every other company uses Google Analytics or other trackers. The problem is that you can see everything that the user does - except for the search queries. The disadvantage of this is that the product cannot be optimised for our scenarios. We are unable to see whether there are problems or how it is being used." As a result, the team is developing data protection-compliant local analysis approaches. "We are expanding federated learning to include federated analytics," adds Lundbæk. "That may also prove be interesting for others."

The Xayn team plans not only to expand its search engine with technological innovations. “We want to show that something like this can be implemented,” says Lundbæk. The company's decision to launch a search engine was based to the expertise of the founders. They also took into account the fact that internet search is one of the largest markets: after all, "30 per cent of internet traffic goes through search engines," reasons the expert. "But I see other areas that can be improved - not just by us." The open source platform XayNet gives other start-ups and companies the opportunity to build similar products. "The video search on YouTube could offer a privacy-protection alternative," he adds by way of example. "It can also be used for streaming services. They don't need the data for advertising and are app-based so that the AI ​​can easily be trained on the device or in the browser. The existing recommendation AI could be gradually switched to this technology. ”It would be a new kind of issue-free viewing pleasure - without having to worry about our data.