Wie beeinflusst ChatGPT die Jurawelt? Werden Urteile bald nur noch von Robotern gesprochen? Und was sagt eigentlich die Rechtssprechung zu den neuesten Entwicklungen der generativen KI? In der neuen Kolumne Richter und Roboter von ARIC und Fieldfisher beleuchten Dennis Hillemann und Stephan Zimprich im Wechsel einmal monatlich Themen aus KI und Recht. Im Februar geht es um KI-generierte Linkedin-Posts.
Lawyers have little time. ChatGPT and comparable language models are therefore being enthusiastically received – above all to drastically increase the output of marketing measures. This is a real risk for social networks.
Rarely has a new tool been received as enthusiastically as ChatGPT. The voice AI reached one million users just five days after its launch – faster than Instagram and Spotify. This makes ChatGPT by far the fastest growing consumer application in the world to date. In the slipstream of ChatGPT, specialized applications are also flourishing, which offer greater factual accuracy or annotation, for example, or which can be used to quickly reword existing texts in such a way that copyright risks are minimized as well as the recognizability of the text as “AI-generated”.
Law firms have also immediately recognized the potential. The major law firm Allen & Overy, for example, recently announced that it had acquired licenses for the “Harvey” chatbot for all of its 3,500 lawyers. The tool is designed to help lawyers create texts such as memos and merger documentation.
However, most lawyers are currently focusing on a completely different use case: chat AIs can drastically accelerate marketing measures. The output of posts on professional platforms such as Xing or LinkedIn has multiplied for some lawyers. Even before ChatGPT, many profiles on these platforms were AI-generated; now the proportion of AI-based posts is increasing rapidly.
It is often difficult for human users to recognize differences between AI-generated and “real” texts. Even with appropriate training, studies have shown that the recognition rate is only just over 55%. If the texts are trimmed to “human” before publication using tools such as Sudowrite, the recognition rate is likely to be significantly lower still. Even an AI finds it difficult to distinguish AI texts from real texts.
This does not bode well for the future of professional networks. The networks are flooded with more and more (AI-generated) content, if you want to be noticed, you have to produce more and more and faster. AI texts dominate through sheer volume. Man-made content is barely visible – too slow, too time-consuming, too little. Users are less motivated to spend time on the platforms. A profile on Xing & Co would then only be important to stay connected and to publish content yourself (with the help of an AI, of course). If interaction with others in the network is then automated, there is no longer any reason to spend time on a platform yourself. And you can use like bots that react to keywords and distribute likes for the corresponding posts.
It won’t take much to create a largely closed AI interaction system on social networks that does not require human intervention: AI-generated profiles create AI-generated texts that are liked by AI-supported bots, whereupon new texts are generated from the texts with the most likes. Real people, that much is certain, are unlikely to voluntarily enter into this kind of content roller coaster.
For social networks, the reliable technical recognition of AI content (and its blocking) is therefore likely to be essential for survival. For lawyers and everyone else, the question is how long the networks will continue to function as communication channels in the market – and what will replace them.
Stephan Zimprich specializes in cases with a technology background and advises clients mainly from the digital sector in the areas of data protection, competition law, media law and IT law.
