Xayn provides web search as it should be. It offers you not only a privacy-protected personalised search experience but also gives you a search-less access to the internet in form of a personalised curation of content. Based on your past interactions with Xayn, the AI anticipates what you are looking for, offering you curated suggestions. The AI, all data, and your interaction history stay under your full control and reside completely on your device.
We developed our privacy-protected search AI because privacy is a fundamental human right. With Xayn we protect this fundamental right, while delivering a great experience through AI-based personalisation. However, we even go one step further. With the help of complete decentralisation, user control and open-source transparency, we're also fighting AI misuse.
In general, we see many applications of AI personalise the experience of users, e.g., in search, advertising, and video streaming. Many companies like Netflix, Google, Facebook or TikTok rely on personalisation or even hyper-personalisation for delivering better content faster. This way, they are convincing, winning and keeping more users because of the superior user-experience they’re offering. Essentially, personalisation is based on understanding user behaviours and interests to adapt and match content.
However, the data that is collected for training models for personalisation can also be used to understand the following: What does the user need to see or read, in order to have or change certain believes or interests? In this case, AI is used to "re-programme" humans to influence their way of thinking. Instead of tailoring content selection to users’ believes, those beliefs are influenced if not manipulated with selected content. This is certainly a very dangerous development – not only on the individual but also and especially on the social level. This ability to shape beliefs through hyper-personalisation is possibly even endangering our democracies and makes the concern about misuse of AI to influence human decisions more relevant than ever before.
We already saw the implications of lacking trust in AI in the debate about TikTok in the US. Such trust issues have certainly come to full force in the US election of 2016. Technically, it is certainly possible to influence elections through AI by adapting (visual) content to try to influence the way of thinking of millions of voters. For example, these trust issues hold true for Google, Facebook and Co. And as we’ve seen in past elections with scandals like those around Cambridge Analytica, the power of tech was used to influence people’s (political) decision making. The question hereby is, how much we can trust these companies to act ethically and to not sell access to either the data or – even worse – to the AI algorithms themselves. In short, whoever builds and controls the AI, is the one who can effectively influence human thinking. As such, we asked ourselves: What is our responsibility as a tech and AI company? Can we do something to change this frightening path that the industry seems to take? How can we make sure that our AI is truly controlled by the people?
The answer lies in radical transparency and user control, paired with decentralisation. All three aspects define our entire company DNA. Xayn’s AI gets feedback from you whenever you left swipe content as relevant or right swipe it as irrelevant.
The Xayn AI therefore gives you a better and better experience over time and you, as a user, have the full control over it. You can also turn off the AI for a generic search experience whenever you so desire. Furthermore, for the first step of decentralisation, Xayn applies edge AI in combination with federated learning to further improve the quality of our AI models. This creates a virtuous and privacy-protecting cycle of effective, personalised search experience for all users.
Xayn's AI based on federated learning is built on XayNet, which we open-sourced to create transparency and to encourage collaboration with others. The XayNet protocol has been designed so that it meets the legal demands of data-privacy regulations such as the EU GDPR. It also offers sufficient robustness to system failures and resiliency to attacks. XayNet uses a publicly verifiable random function to select – in each round – who can submit an encrypted AI model for aggregation, and who can participate in the decryption of the aggregation of encrypted AI models. This is akin to – but different from – homomorphic encryption, is very efficient, and makes the aggregation process both fair and democratic.
Now, Xayn is going one step further in its fight against AI misuse by incorporating IOTA's Tangle technology as a trust anchor into XayNet. We firmly believe that the trustworthiness of the XayNet protocol and Xayn itself is further enhanced by its integration with DLT technology. Therefore, we are integrating The Tangle into XayNet to record the hash of the decrypted aggregated AI model that was computed in a round of federated learning. Xayn will decide whether to trust this aggregated model based on the familiar trust metrics that The Tangle offers. This creates a further necessary step to validate that only the user shall have the full control over their AI model and it moves away the trust that you need to have in us or any other contributor into the AI. Thus, we wrap decentralisation, user control and AI transparency in The Tangle as an additional trust layer.
This is merely the beginning of further IOTA integration opportunities for Xayn and our vision. At present, the orchestrations within that protocol, including the aggregation itself, will be done by a central federated learning coordinator. The coordinator's behaviour is by and large verifiable by round participants. But there are compelling reasons for making this coordinator decentralised. In fact, the reasons for such a major change are like those that led to the Coordicide of The Tangle for IOTA. We are in the process of extending our XayNet framework so that XayNet’s coordinator becomes decentralised while the spirit of the original protocol – which sees to its legal compliance, robustness, and resiliency – is being maintained.
Stay tuned for future news on how IOTA'S DLT technology can help realise this process.