INTRODUCTION

• 1. Imagine that for your next holiday you are interested in flying from Brussels to Windhoek, Namibia, because you want to go on a safari in the much praised Etosha Park. On social media you read about a small start-up in London which developed a new AI tool that can help you book air travel. Instead of having to manually compare countless possible flights on a myriad of booking websites, this new tool autonomously makes a comprehensive analysis of all possibilities. Based on some pre-specified variables, the AI system books the best flight on offer. You are, for instance, happy to pay any price below € 1.000 for a return ticket to Windhoek, but of course you are aiming for ‘as cheap as possible’. You only want a direct flight, if possible, in the morning. Based on these preferences, the AI tool will book a seat and, in doing so, conclude a contract that binds both you and the airline.

While this example sounds straightforward, and is technically not farfetched, it stands in sharp contrast with current Belgian contract law, which is by no means ready to deal with an AI tool, even such a simple one. What happens, for instance, if the AI tool makes a mistake and books you a flight to Cape Town instead? Or what happens if you refuse to pay the ticket price because you just found a cheaper ticket anyway? These are typical contractual issues, but here the problem pertains to an intermediary and autonomous system which is not behaving as expected, something any AI tool is statistically bound to do in a certain percentage of cases.

• 2. We believe it is important to have a contractual framework that is adapted to AI because the use of this technology brings about many advantages. As a digital tool, it allows for faster decision-making, which in turn may help in cost-cutting, e.g. via supply chains based on just-in-time principles. Furthermore, decisions made by AI systems are generally more accurate and less biased than those made by humans. On top of that, the use of automated systems helps to eliminate clerical errors.

INTRODUCTION

• 1. Currently, economic actors do not only track online behaviour of consumers but also gather information about the offline world through domestic appliances that are connected to the internet – also known as the ‘internet of things’ or IoT. The information collected can range from objective facts, such as the ambient temperature, to very personal data concerning its users. This contribution focuses on one specific application of AI in a domestic IoT setting: smart home assistants (SHAs).

• 2. Smart home assistants can be approached from two angles. First, there is the futurist point of view. According to this viewpoint, a technology can only be called a ‘smart home assistant’ if it pulls together data from different devices in the home, builds a real-time profile of the conditions in the home, and takes actions according to a combination of such analyses and commands issued by the home owner. However, when observing the real-life landscape it is clear that we are still far-removed from a widespread implementation of such SHAs. Although consumers increasingly own individual smart devices and household appliances, the only IoT devices that manage to achieve a considerable uptake percentage are home assistants, smart speakers and smart watches. Furthermore, there is scepticism among consumers regarding the need for a central control system vis-à-vis individual controls by different devices and apps, which is possibly explained by the lack of trust consumers have in the security of these devices.

• 3. This contribution therefore takes a realistic angle and chooses to investigate the currently most widely used SHAs: the combination of a smart speaker with a virtual personal assistant (VPA). Around 24% of the population in the United States now have at least one such a device and adopters on average have 2.6 such devices in their home. A recent UK survey revealed that 22% of those interviewed owned such a device. In the Netherlands around 19% of the population is estimated to have a smart speaker in their household. For Belgium, no numbers were found.