Do the math: ACM publishes strategy on monitoring use algorithms
The ACM worries that the use of algorithms may lead to the creation of cartels, or nudge consumers towards a purchasing decision that is not in their best interest. Therefore, on 10 December 2020, it published a new policy document (in Dutch) setting out what businesses can expect when the ACM checks their algorithms. On the same day, the ACM also launched a trial with online music library Muziekweb to improve the ACM’s knowledge about the categories of data that are likely to be relevant in such investigations. All signs indicate the ACM’s intention to become more active in this area.
Although the ACM acknowledges that the use of algorithms has great benefits for businesses and consumers, it also notes that they can have harmful effects when used to violate consumer protection or competition rules. For example, companies cannot use algorithms to implement personalised pricing based on characteristics such as disability, religion, ethnicity or race.
If the ACM has a reasonable suspicion that an undertaking’s use of an algorithm is illegal, it can oblige the undertaking to provide the source code of the algorithm and all other relevant documents (such as internal communications, input data, log files, test and debugging reports, scrum boards and GitHub repositories). On the basis of this information, the ACM will look at aspects including (i) the algorithm’s role in the alleged violation; (ii) the purpose for which the algorithm is used; (iii) the basic principles of the algorithm’s design; and (iv) the input for and the results of the algorithm (“input-output analysis”).
The position paper also sets out certain challenges the ACM is likely to face when it investigates an algorithm, such as:
- The possibility for self-learning algorithms to change from one moment to the next, for example by using artificial neural networks. The ACM notes that it can use advanced technical tools to understand the functioning of such complex algorithms.
- External parties and integrated use of software. Algorithmic applications can be part of an ICT environment with links to (sub)systems and datasets that may belong to external parties. The ACM warns that companies that purchase third-party software may still be liable if they use that software to violate consumer protection or competition rules.
- Cross-border elements. The ACM’s view is that it can copy data, even in a cross-border situation, if one (or a combination) of the following three elements is satisfied: (i) the company under investigation owns the data, (ii) manages the data; or (iii) uses the data.
- Privacy concerns. If an algorithmic application processes personal data, the ACM needs to structure its investigation in such a way that it complies with all applicable privacy rules.
- Deviation from standard investigatory practices. If algorithms are simple, the ACM can copy the code and store it in its own environment. However, if a company uses a third-party cloud service, the ACM may only be able to investigate if it also uses the same third-party cloud service. This could force the ACM to deviate from its standard investigatory practices.
The ACM’s new focus on algorithms is likely to lead to formal investigations and potentially to complex litigation relating to the boundaries of its investigatory powers. For example, it is not certain that companies under investigation will readily accept the ACM’s view on the scope of its powers as set out under points (3) and (5) above.
This article was published in the Competition Newsletter of January 2021. Other articles in this newsletter:
