Background and context – Sorty (GAR–LCIA Hackathon)

This post relates to our participation in the GAR–LCIA Hackathon, an international competition organised by Global Arbitration Review and the London Court of International Arbitration in collaboration with Stanford, Google and Three Crowns. Within the competition, participants are tasked with identifying concrete problems in international arbitration and developing practically useful, AI-based solutions that can be demonstrated in working form. The task was thus not to analyse legal rules in the abstract, but to understand legal practice and translate that understanding into a technical tool of real use.

‍

Against this background, we chose to approach the task by first observing and mapping the professional reality the solution was intended for. Rather than starting from the technology, we contacted more than fifty senior lawyers working in international arbitration and commercial disputes, whom we met with a structured list of questions. The conversations focused on their actual working day: how workflows are structured, where complexity and lost time arise, and how AI is currently used – or deliberately not used – in practice.

‍

The observations gave a relatively consistent picture. There is a clear interest in AI, but its use is often cautious, fragmented and marked by low trust. Recurring themes were the difficulty of integrating new tools into established work processes and the concern that the technology risks increasing complexity rather than reducing it.

‍

To complement the user perspective, we also held a dialogue with a lawyer with particular responsibility for AI matters (AI counsel) at a law firm, who both works on integrating AI into the firm's practice and has developed tools herself. This gave further insight into the structural and organisational conditions for AI development in legal practice. A central observation was that the greatest challenges are seldom technical in nature, but rather concern resistance to change and the difficulty of getting users to depart from established ways of working.

‍

The overall conclusion of these observations was that the problem seldom lies in the technology as such, but in how it is used. AI can in itself achieve a great deal, but as a rule requires substantial input from the user – and thus a familiarity with the tool that far from everyone possesses. Since lawyers also work in different ways, tools that require extensive hands-on effort must be adapted to each individual user's way of working, which in practice constitutes a considerable obstacle to broad and lasting use.

‍

Against this background, we chose the opposite starting point: a solution that requires almost no input from the user. The tool, which we have named Sorty, was developed and programmed by Oliwer Martens and me ourselves. Sorty is an AI agent that works in the background from the very start of a matter. It reads incoming material, links it to the correct case file and gradually builds up a structured model of the dispute, in which the parties, roles, procedural posture, background facts, issues in dispute and procedural history are gathered in one place. Submissions, exhibits and the exchange of written pleadings back and forth are thus captured automatically and structured into a single matter per dispute, without the user needing to steer the process. The name Sorty reflects this core function: automatically sorting and structuring a matter's material, so that the lawyer can devote their time to the legal substance rather than to the administration surrounding it.

‍

The result is what we describe as a digital twin of the dispute, and it is here that the practical value of the solution emerges. First, the twin gives arbitrators and parties alike an immediate and comprehensive overview of the dispute, without the manual work otherwise needed to hold together a growing body of material. Second – and this is one of the central points of the tool – the model makes it possible to use AI as a tool for analysis and experimentation in relation to the dispute. Disputing parties can test different questions, perspectives and hypotheses against the digital twin and let the AI help explore the various dimensions of the dispute, without having to experiment live. It thereby becomes possible to experiment in a controlled environment, decoupled from the risks of trial and error directly in the ongoing proceedings.

‍

Sorty was developed within the framework of the first GAR–LCIA Hackathon, which brought together 44 entries on how AI can be used to develop international arbitration. Our entry was highlighted in Global Arbitration Review's subsequent review of the competition, in which Sorty was singled out together with a few other entries.

‍

The tool is thus a direct result of our observations and has been designed with a focus on a low threshold, intuitive use and minimal impact on the lawyer's existing workflow. The accompanying video shows how the solution works in practice and how the user interacts with the tool. This post should be read as a contextualisation of the task and as documentation of the observations that formed the basis for the development work.

‍

Link to Global Arbitration Review's review: https://globalarbitrationreview.com/article/what-ideas-emerged-the-gar-lcia-hackathon