Articles 101 and 102 of the Treaty of the Functioning of the European Union (hereafter ‘TFEU’) regulate anti-competitive practices within the EU. In order to minimise abuse of dominance which leads to anti-competitive behaviour the Court of Justice of the European Union (hereafter ‘CJEU’) has ruled that data holders could be obliged to allow data access under the requirements set out by the law.[21]
Data or “information”, allows users or consumers on the Digital market to make informed choices or transactions, but de facto data does not have any value in itself. Digital data affects transaction patterns and thereby prompts shifts within the economic value chain. So for example, in relation to digital car data, aftersales of a car are always affected by receiving access to the car data itself. Hence, there is a vertical relationship between the upstream data markets and downstream markets for transactions in goods and services: changes in access and conditions in the upstream market will affect the downstream market.[22]
From a competition law or market perspective, for businesses operating online this has led to competitive strategies being put in place between sellers which in turn lowers prices of goods and services which triggers positive and negative effects throughout society.[23] Economies of scope occurs where for example business which are free services in return for collecting consumer behaviour patterns online which can later be monetized by individuals deciding use a connected device for example. The economic incentive of doing so is that the market value of individual personal data is very low compared to the consumer surplus value of the free service they receive in return from the data aggregator also known as economics of scope, mentioned below. Economics of scale in ML occurs where marginal cost of additional use of the algorithms have are lowered where movements are made into making the connected device function more efficiently which leads to economies of scale in ML.[24]
Economies of Scale
Artificial intelligence and Machine learning mechanisms thrive on access to (big) datasets as necessary inputs for training algorithms. ML has changed the production process for innovative data driven ideas and therefore ensuring access, and transfer to (personal) data becomes an important issue for promoting innovation. ML is a statistical technique and their closest and best estimates are dependent on the large size of the datasets generated by connected devices. For example, in IoT and/or connected devices several variables may be present in the large datasets and therefore it is crucial that robust predictions are generated to diminish the existence of any errors. ML requires larger datasets than humans need in order to learn as it requires thousands or millions of observations to learn some basic responses in contrast to humans who may only need a few observations. For example, “a self-driving car algorithm can handle most traffic situations after having “learned” to drive from millions of kilometres of data inputs; a human driver only needs a few thousand kms of experience to become a proficient driver.”[25] The algorithm is able to comprehend millions of kms of data input in a shorter timespan in comparison to a human driver needs to learn to drive several kilometres. Likewise the algorithm can drive many cars at the same time; a human driver can only drive one car at the time. Hence, investing in high quality datasets transmitted by connected devices for the purposes of training ML algorithms is expensive yet once trained the marginal cost of additional use of the algorithms have shown to be low which leads to economies of scale in ML.[26]
Economies of Scope (Data Aggregation)
When merging two or more datasets economies of scope transpires where the benefits of using a combined dataset is higher than using each dataset separately.[27] These datasets need not be completely separable, instead they should complement each other. For instance, web surfing data may produce insights on consumer behaviour and therefore combining these with mobile phone data may produce more insights, compared to studying both datasets separately. Some business models of aggregators merge data from various sources into a single consistent dataset which may for example allow for targeted advertisement such as in the example of combining consumer behaviour data with the content they view online. For example, individual car data on driving performance is valuable alone for insurance and maintenance purposes and there is no need to aggregate that data with other cars data. On the other hand, car navigation data needs to be aggregated by a navigation service provider in order to identify traffic jams and send this information back to drivers. All the data produced in abundance here is thanks to connected devices. There are considerable economies of scope in the aggregation compared to the marginal value of each individual car navigation dataset.[28] It is the ML that aggregates such data through trained knowledge of algorithms. [29] The same conclusion can be withdrawn in the case of connected devices where the connected devices producing large datasets make use of ML that aggregates such data to improve the accuracy of the connected devices functioning efficient and to their optimal capacity.There are considerable economies of scope in the aggregation compared to the marginal value of each individual dataset.[30] These datasets may contain within software of a connected device which may be protected by the SGDR under the DbD.
Data portability i.e. “the ability to move, copy or transfer data”[31] is one of the instruments dictating control over data by an individual. As mentioned above, data can be seen as the ‘new money’ in today’s society so therefore it is important to weigh in the benefits of regulatory intervention regimes and how this affects the sharing, transfer, and trading in data. Now that it is established that the abundance of data generation and collection is shaping the livelihoods of our present and future, it is crucial to understand the legal basis of access, store and transmit it to third parties.
References [20] Josef Drexl et al, ‘Position Statement of the Max Planck Institute for Innovation on the European Commission’s “Public Consultation on Building the European Data Economy”’ [2017] Max Planck Institute for Innovation & Competition Research Paper No. 17-08. [21] Case T-17/21 Miquel y Costas & Miquel v EUIPO (Pure Hemp) OJ C72, 1.3.2021. [22] Bertin Martens ‘An economic policy perspective on online platforms’ [2016] JRC Digital Economy working paper 2016-05. [23] Bertin Martens, ’What does economic research tell us about cross-border e-commerce in the EU Digital Single Market?’ [2013] No 2013-05, JRC Working Papers on Digital Economy, Joint Research Centre. [24] Bertin Martens,’The impact of data access regimes on artificial intelligence and machine learning’[2018] JRC Digital Economy Working Paper, European Commission, Joint Research Centre (JRC). [25] Bertin Martens,’The impact of data access regimes on artificial intelligence and machine learning’[2018] JRC Digital Economy Working Paper, European Commission, Joint Research Centre (JRC). [26] Bertin Martens,’The impact of data access regimes on artificial intelligence and machine learning’[2018] JRC Digital Economy Working Paper, European Commission, Joint Research Centre (JRC). [27] Bergemann et al, ‘Markets for data’ [2012] >https://economicdynamics.org/meetpapers/2012/paper_538.pdf> Accessed 12 April. [28] Bergemann et al, ‘Markets for data’ [2012] >https://economicdynamics.org/meetpapers/2012/paper_538.pdf> Accessed 15 April. [29] Bertin Martens,”The impact of data access regimes on artificial intelligence and machine learning“[2018] JRC Digital Economy Working Paper, European Commission, Joint Research Centre (JRC). [30] Bergemann et al, ‘Markets for data’ [2012] >https://economicdynamics.org/meetpapers/2012/paper_538.pdf> Accessed 12 April. [31] Commission, ‘Commission Staff Working Document on the Free Flow of Data and Emerging Issues of the European Data Economy Accompanying the Document Communication Building a European Data Economy’ SWD(2017) 2 final, 47.
Comments