Conclusion

Investigating the role large online platforms’ recommender systems play in shaping individual and collective experiences — both online and offline — comes with great challenges. In the absence of better insights into how these systems function and better access to data from or about platforms, their impact remains poorly understood. At the same time, many harms caused or added to by platform recommender systems are already well established. It is therefore critical to both rein in these harms while creating the foundation for a better understanding and higher scrutiny of these systems and the companies that deploy them.

This paper seeks to provide a snapshot in time in this fast-moving field and debate. It does not and cannot offer an exhaustive list and discussion of potential options aiming to tackle the challenges we face relating to platform recommender systems. It does, however, lay out the steps Mozilla thinks are suitable and necessary at this point in time to build a better recommendation ecosystem. Our proposal is systemic in nature, prioritizing a concert of actions over quick individual fixes. Our goal is to hold platforms accountable while providing the public, experts, and individuals with bespoke tools to scrutinize recommendations and the engines driving them.

At the same time, more transparency, accountability, and control don’t obviate the need for thinking more deeply about how recommender systems can be designed in a way that is more aligned with the public interest. Exciting work on this question and others is emerging and close attention should be paid to it going forward as our understanding of the issue space will evolve.^[1]

Ultimately, our contribution to this discussion should be seen as an invitation to engage and provide feedback in order to advance the debate and move towards a consensus of what is necessary.