Legal content losses in algorithmic translations of law

When it comes to integrating and hybridizing autonomous systems in our societies, an algorithmic translation of regulations is necessary at one point or another. Whether it is to make sure the autonomous car stops at the stop sign, or to embed the obligation to distinguish civilians from combatants, developers of AI based systems evolving in our societies need to make sure that their products comply with regulations that would otherwise normally affect human beings having direct control over machines.
What is the legal value of an algorithmic translation of a regulation? Understanding if an algorithmic translation of a regulation is the same thing as a human-language version of a regulation is important in order to better understand which weights are on whose shoulders when it comes to making legal decisions. It also serves to find the best way to hybridize artificial intelligence, human beings and law.
Algorithmic translations can be made both by robots and by humans. While the first ones translate the world around them into an equation where a specific element obtains a normative weight impacting its behaviour, the robot doesn’t actually understand the concepts of rule and regulation, it only gives a certain factor a prohibitive weight based on the data it is given to build upon. Human beings creating robots can make algorithmic translations of rules that the robot must follow. In both methods, content in the law is lost in translation. It seems impossible to make a comprehensive translation of a regulation into an algorithm for two main reasons. Firstly, any regulation supposes a certain number of implicit elements that are not defined in the regulation or in connected pieces of regulation, but that are foundational to its normative effect (eg. biological concepts of mortality and time related to the sanctions in criminal regulations are an integrant part of its effectiveness). The second reason is that it is impossible – with the current level of development in cognitive neurosciences – to make a comprehensive charting of all the implicit elements of the regulation, part of which are in our brains. If we can’t export that implicit knowledge into an algorithmic translation, then the algorithmic version of the regulation has lost part of its legal content in the translation process.