In April 2022 the DRCF published two papers related to algorithmic processing. One looked at the specific harms and benefits posed by algorithmic systems, including those powered by machine learning methods, and the other looked at the merits of algorithmic auditing, as a means of documenting the risks associated with these systems and assuring stakeholders that they are safe to use.

The DRCF says that the purpose of the research was to reveal areas of common interest among the DRCF members, ensuring that any future interventions made as individual regulators are aligned and complementary. The intention was also to highlight emerging ideas for how to mitigate the risks posed by algorithms, and to signal to industry those solutions that appear the most promising, and which need further exploration. This includes interventions that: (i) promote transparency for users and developers (e.g. via model and system cards); (ii) give people more control over the systems that shape their daily lives; and (iii) enable people to seek redress where they have been harmed.

The DRCF also published a call for input, inviting views from stakeholders on its findings and conclusions. In the case of the paper on algorithmic auditing, the DRCF specifically asked for views on several hypotheses, which set out a vision for the role that could be played by regulators in using algorithmic audits and supporting the development of a third-party auditing market. It also asked stakeholders for their opinions on where the DRCF should focus its energy in the future.

In total, the DRCF received 31 responses. Most were supportive of the two papers and welcomed the four digital regulators collaborating on this topic. There was widespread agreement that algorithmic processing creates opportunities and risks across the entire economy and that there is value in regulators pooling knowledge and working together to promote best practice.

In respect of the first paper, the DRCF says that most respondents felt that this had captured the majority of pressing issues, among them risks to fairness, transparency and competition. However, some felt that the analysis could have paid closer attention to the impact of algorithms on the environment, as well as to workers and their experiences of the labour market. While some respondents disagreed with several claims made in the paper, e.g. that the use of algorithms on social media may be driving a polarisation of views in society, in most cases the feedback received sought to build on the DRCF’s analysis, as well as to highlight some of the challenges in addressing the risks identified. For example, several stakeholders cautioned that efforts to improve algorithmic transparency could prove counterproductive if they allow systems to be more easily gamed.

The DRCF received a similar set of responses to the second paper. Stakeholders were generally supportive of the hypotheses outlined by the paper, including: (i) that there would be value in regulators clarifying how third-party audits could support the regulatory process; and (ii) that there may be a role for some regulators in producing guidance on how third parties should conduct audits. However, not all the ideas drew support. For instance, some stakeholders cautioned against the idea of creating mechanisms for information sharing between third party auditors and regulators due primarily to security and privacy concerns. Several also underlined the challenges involved in accrediting auditors, which could be significantly costly for regulators and create disproportionate burdens for auditing firms.

The DRCF says that it will use the feedback gathered to inform the next stage of its project on algorithmic processing and relevant work across the individual regulators. Over the coming work programme year (2022/23), the DRCF plans to undertake several new activities, including looking at how to improve the procurement of AI systems (such as content moderation or age verification technology), and a stream of work where the four regulators will share lessons learnt and best practice in how to conduct algorithmic audits. To access the DRCF response document, click here.