Researchers from Delft University of Technology are on a mission to enhance their BitTorrent client, “Tribler,” with decentralized AI-powered search capabilities. A recent demonstration highlights how generative AI models enable novel content searches without constraints. The overarching objective of the research project is to recalibrate the power dynamics of the internet, transitioning authority from governments and large corporations back to consumers.
Twenty-five years ago, peer-to-peer file-sharing ignited a revolution on the internet, enabling users to search for and share content with strangers on a global scale.
In the years that followed, online media consumption surged, often driven by shared content without proper authorization. However, these early pioneers of piracy laid the groundwork for innovative business models that emerged later.
While the original ‘pirate’ ethos has evolved, with centralized streaming platforms dominating the landscape, there are still pockets of innovation driven by technological progress, particularly in artificial intelligence (AI).
One such example is the Tribler research group at Delft University of Technology, which has been developing its Tribler torrent client for nearly two decades. Their aim is to advance decentralized technology, shifting power away from corporations and governments and towards individual users.
Their latest endeavor involves combining decentralized large language models with decentralized search, creating a framework called “De-DSI” (Decentralised Differentiable Search Index). This innovative approach allows users to search for content stored across peers using decentralized AI-powered search. The process is entirely peer-driven, without the need for central servers, ensuring greater user autonomy and control.
While the concept holds immense promise, the current demo version is still in its infancy, with limited datasets and AI capabilities. Nonetheless, it represents a significant step towards realizing the potential of decentralized AI implementations within the BitTorrent ecosystem.
In essence, De-DSI operates by distributing the task of training large language models across peers in the network. Each peer specializes in a subset of data, which can be retrieved by others to generate optimal search results.
The proof of concept demonstrates the feasibility of this technology. However, integration into the Tribler torrent client is still in progress, with the aim of releasing an experimental decentralized-AI version by the end of the year.
While the researchers view this as a significant breakthrough, immediate improvements for users may be limited. Initially, AI-powered search may be slower, and for users with specific search queries, the benefits may be minimal.
However, the researchers are committed to refining the technology, envisioning a “global brain” for humanity as the ultimate goal. Beyond torrent searches, decentralized machine learning could be employed to combat spam, provide personalized recommendations, and optimize torrent metadata.
The researchers’ primary objective is to empower the public, reducing reliance on large corporations and centralized control. As Johan Pouwelse aptly puts it, “The battle royale for Internet control is heating up.” Driven by their idealism, they are dedicated to gradually shifting power back to citizens, a mission they’ve been pursuing for nearly two decades.
The limited De-DSI proof of concept and associated code are available on Huggingface, while detailed technological insights are provided in the accompanying paper. Meanwhile, the latest fully decentralized version of Tribler can be accessed on the official project page.
