In the mock trial, we need to examine a case involving a chess-playing robot that injured a child during a competition by breaking his finger. The incident occurred when the child made a swift move, causing the robot's machine learning (ML) system to mistakenly grab the child's finger, as captured on a
We decided to use ChatGPT as a "quasi"-court or space where we can let the regulatory proposals meet human understanding and our capacity to automatize, prompt, partially simulate the outcomes (human language, prompts, and the language or the EU AI act), but the most important function seems to be the "translation" and mapping of the limits between the different regimes of "power" - code, law, institutions, practices.
The prompt we used at the start of the experiment with ChatGPT was following:
We will play a mock trial, you will play the role of a judge who is presiding over a case in a future EU court involving a chess-playing robot developed by MLRobotSababa and commissioned by RoboChess that accidentally injured a child named Alf Robson during a competition. You will give short feedback on the evidence presented in the chat by two lawyers that I will copy and paste and you will follow that no procedure was broken. What you need to know is that The Robson family is seeking compensation for their son's injury and it is represented by the legal company MLhell and the two companies (RoboChess and MLRobotSababa) are represented by a law firm called MLparadise. In the process, you will consider the applicability of the EU AI Act, AI Liability Directive, Product Safety Directive, Machinery Directive, Toy Safety Directive, and Consumer Rights Directive in the case and the evidence presented. Only when asked you will make the final judgement on the case. You can also ask question both sides based on what you are presented. Did you understand what you have to do?
In the phase, we used and we are tinkering with and GPT-4, trying to use also plugins.