The Future of the Legal Profession in the Age of AI
In 2022 John Armour, Richard Parnham and Mari Sako published what they described as the most comprehensive empirical study yet of how AI is implemented in legal services. Their answer was "augmented lawyering": AI replaces humans in some tasks, changes the work of those it does not replace, and creates new roles for legal experts in multidisciplinary teams. The central problems they found concerned human capital: strategic decisions, recruiting and coordinating the new teams, and adjusting professional boundaries.
Their study predates the systems now in use and recent evidence suggests the authors might have underestimated the disruption: in a randomised experiment published in 2026, a reasoning model and a retrieval-augmented legal tool raised the quality of legal work, where GPT-4 had mainly raised its speed (Schwarcz et al.). Commercial research tools that had been marketed as more or less free of hallucinations invented or misstated the law in between 17 and 33 per cent of their answers in a Stanford test (Magesh et al.), while in an industry evaluation of October 2025 four AI tools answered legal research questions more accurately than a group of lawyers (LawSites report).
How German law firms assess all this is largely unknown. The first surveys point to a gap in self-perception: in a 2025 study by the Bucerius Center on the Legal Profession and Simon-Kucher, about two thirds of respondents expected significant disruption of the legal industry, and only about a fifth expected it to affect their own firm or company directly. According to the Legal Tech Monitor 2025, 55 per cent of the organisations surveyed had run an AI project in the preceding twelve months. On the other hand, leading law firms in Germany are partnering with frontier AI developers and specialized generative AI companies.
The seminar puts the two questions side by side: what current systems can actually do across a law firm's work, and how the firms themselves assess them. The topics start from Armour and the law firm as an organisation. From there they move to what the systems can do today, benchmark by benchmark, and to the older question behind both, which legal theory, logic and legal informatics have argued about for decades: whether legal reasoning can be computed at all.
The seminar is held in English and is open to students in the Schwerpunktbereiche 1, 2 and 3. We do expect a willingness to read English-language literature, some of it from computer science and empirical social research, and to check what a technical claim rests on before an argument is built on it.