Like any other new technology before, artificial intelligence (AI) challenges any legislator to strike the perfect balance between enabling the benefits and minimising the harms. While the Global North appears to favour a harms based approach to AI regulation, the Global South tends to approach AI regulation predominantly as empowerment. After a brief introduction (I.), the article first contrasts these two approaches (II. and III.) and then demonstrates the difference in regulatory focus regarding three key areas: liability, access to data and economic dependence (IV.), before drawing conclusions on the major factors influencing the direction in which the regulatory pendulum will swing (V.).

The article compares the privacy legislation and associated PIA guidance applicable to private-sector organizations in Canada (excluding Quebec), in Quebec, in the European Union and in California, USA. For that purpose, the article briefly describes the role of a PIA (I.); provides a brief overview of the relevant domestic legislation and associated guidance (II.); sets out the triggering events that require the undertaking of a PIA under each regime (III.); outlines the required conduct of the PIAs under each regime (IV.); sets out the required “passing grade”, if any – i.e. the threshold which the PIA results must surpass in order to proceed with the product/service/initiative (V.); sets out the penalties for non-compliance under each regime (VI.); and concludes with a comparative analysis of the details of each regime’s PIA requirements (VII.).