Artificial Intelligence and Compliance
On December 13, 2019, an interdepartmental work group presented its report Challenges of Artificial Intelligence commissioned by the Federal Council [1]. Section 1.6.16 addresses the use of artificial-intelligence in the judicial system. The matter of compliance is also relevant here. The European Ethical Charter on the Use of Artificial Intelligence in Judicial Systems [2] and their environment distinguishes between four types of use: (1) “uses to be encouraged”, (2) “possible uses, requiring considerable methodological precautions”, (3) “uses to be considered following additional scientific studies”, (4) and “uses to be considered with the most extreme reservations”. What does this mean specifically in terms of compliance?
Rule-based AI
Simply put, there are two different methods used for artificial intelligence: rule-based methods that are based on expert systems, such as those developed by Ed Feigenbaum at Stanford University in the 1960s. The advantage of these systems is that for each result, it is clear and verifiable which rules were applied. Jurists are generally familiar with the following syllogism: “All men are mortal.” “Socrates is a man.” “Therefore, Socrates is mortal.” Rule-based systems automatically evaluate these kinds of inference rules.
The second type of AI method employs statistical learning. In order to minimize errors, a machine learns a behavior based on examples. Depending on the task, the error rate of the machine may be lower than the human error rate. The functions used in these cases are of an abstract nature and do not provide any set of arguments. Testing for statistical significance has to suffice. This statistical logic is supposed to work in any scenario calling for a risk-based approach. In our view, the category of “uses to be encouraged” set out in the European Ethical Charter is reasonable. Assistance systems designed to prevent frequent errors are also to be encouraged. These systems include, for example, ones that verify the completeness and consistency of compliance files.
Links:
[2] https://rm.coe.int/charte-ethique-fr-pour-publication-4-decembre-2018/16808f699b
