Menikö juna jo? Tekoälyn sääntelemisen mahdollisuuksista
Nyckelord:
tekoäly, sääntely, sääntelymahdollisuudetAbstract
Has the train already left the station? Options for regulating artificial intelligence
The article examines the regulatory options for AI technologies. It is argued that AI regulation proposals and literature frame AI in terms that cause misunderstanding in the discussion/debate and suggest it is about unpredictable and autonomous learning systems that are difficult if not impossible to control. However, this is not the case. The article proposes six new articulations for regulation-relevant AI properties. The properties are technical agency, complexity, interpretability, non-linearity, and the extent of input and output spaces, (in) determinacy and dynamism of the systems. When these articulations guide analysis, new regulatory problems and opportunities arise. The article then moves on to explore the regulatory tools that could be used to address the immediate adverse effects of each of these properties. The outlook is a cause for optimism but also concern. The immediate adverse effects of AI technologies can be addressed, but the solutions are not easy. Meta or process regulation, simulation-based performance regulation and appropriate explanation and transparency regulation may help in controlling the potential harm of AI. Simultaneously, care must be taken to ensure that indeterministic and dynamic systems are not allowed to enter safety-critical environments. Regarding the control of indirect damage, there are few options. The means for controlling the potential harm of AI are few in number and there is little political will for regulatory projects in sight.
