Is the government picking the wrong place to start regulating algorithms?

In preparation for our October 26th evening we’ve here republished discussion leader Sara Bannerman’s article Is the government picking the wrong place to start regulating algorithms?  Originally published on Policy Options May 31, 2021.

Bill C-10 – the current proposed revision to Canada’s Broadcasting Actproposes to regulate the discoverability of Canadian content (CanCon) online. Discoverability is a key feature of online platforms like Netflix, YouTube or Spotify. These services make it easy to discover new content through algorithmic recommendations that show up in banners, recommendation rows, playlists, “watch next” recommendations and search features. There is a place for regulating algorithms, but regulating algorithmic recommendations to promote CanCon is the wrong place to start.

Recommendation algorithms have been charged with creating filter bubbles, echo chambers and political polarization. While these concerns must be taken with a grain of salt, regulating the algorithms that generate these recommendations is a big deal. Algorithms affect what we see online, viewing trends, ad dollars, subscriptions, and decisions on current and future content creation. Regulating them could have significant cultural and financial implications.

Bill C-10 is interesting because it opens the door to regulating recommendation systems for the purposes of promoting Canadian content. This could involve requiring a “Canadian” row on the Netflix recommendation page, prominent placement of Canadian songs on key Spotify playlists, or a certain number or percentage of recommendations of Canadian content to viewers in Canada.

The power the CRTC already has

The Canadian Radio-television and Telecommunications Commission (CRTC) has, since 1999, interpreted the Broadcasting Act as saying that much online audio-video material is “broadcasting,” and therefore subject to its regulatory authority. This includes a vast array of online audio-video material, from podcasts to cat videos, webcasts and livestreams. While the CRTC has, in its own view, had the power to regulate online broadcasters for more than 20 years, it has largely declined to use that power. Perhaps it anticipated a backlash, or that attempts to regulate online “broadcasters” would attract a Charter challenge of its claimed jurisdiction over the internet. The passage of Bill C-10 into law would clarify the CRTC’s contested regulatory authority over online broadcasting. This could lead to CRTC regulation of discoverability.

Which of these things is the CRTC likely to do?

While the CRTC has largely declined to regulate online entities, an emboldened CRTC might pursue various strategies, depending on the law’s final form. A preliminary draft policy direction associated with the bill would call on the CRTC to ensure that “Canadian programming in English, French and Indigenous languages is available, prominent, and easy to discover.” Requiring Netflix or Spotify to hold a certain amount of Canadian content available in their library would be one possible strategy, following an approach adopted in Europe.

Another strategy would be to require recommendation systems to promote a certain amount of Canadian content in banners, rows, “watch next” features and playlists. It is not clear what strategies the CRTC would adopt. All are potentially on the table with the details left up to the CRTC.  This uncertainty is concerning.

Do algorithms speak?

Algorithmic search results and recommendations are forms of speech and expression. There are important debates about when algorithmically generated content is protected as speech, but it seems clear that Netflix or YouTube recommendations are, generally speaking, forms of expression.

In Canadian law, “expression” is defined broadly. Parking a car in protest can be a form of expression.  Netflix’s decision to algorithmically promote and recommend certain content is also a form of expression. After all, algorithmic recommendations convey meaning and the commercial preferences and business decisions of the platform. Requiring recommendation systems to promote a certain amount of Canadian content would limit platforms’ freedom of expression. Sometimes limits are justified. Is that the case here?

Is mandating discoverability justified?

In its recent plans and priorities documents and its policy submissions, the CRTC has not, to my knowledge, articulated a pressing problem requiring regulation to promote the discovery of Canadian content online. Indeed, some Canadians have found massive audiences on YouTube, Spotify and Netflix. It is easier than ever to find Anne Murray’s “Snowbird” – with lyrics – or her duet with k.d. lang online.  Somehow, the CRTC is concerned that the abundance of content on the internet may make it difficult to find or recognize Canadian content. It’s as if they’ve never heard of a search engine.

Somewhat more reasonably, the CRTC raises concerns that platforms have an edge over traditional broadcasters, given the range of discoverability features platforms offer. For instance, traditional cable does not offer recommendation and search features. There is also a concern about the position of dependency artists and content owners are in with regard to recommendation algorithms as creators’ fortunes can turn on algorithmic recommendations. Finally, there is the possibility that recommendations based on past pattens of consumption could trend toward recommending more and more foreign content if it dominates current viewing patterns.

Currently, Spotify may highlight “Canada’s most streamed,” but that might not always be the case – particularly if government pressure to highlight Canadian content ends. Is requiring every fifth “watch next” recommendation on YouTube to be Canadian a reasonable or minimally invasive way of confronting a potential and possibly unlikely future problem of Canadians disappearing from recommendations? I’m not convinced.

Discoverability regulation could become a vehicle for official CanCon to maintain its dominance – to try to control the loss of traditional radio and television viewership to a multiplicity of other sources. Traditional media could use discoverability as a tool to maintain a central position in the Canadian media landscape.

Mandating CanCon discoverability could give traditional Canadian media a boost over other online outlets as a reliable source of Canadian news, entertainment and information. However, it is important to recognize the ways that traditional media – and the Canadian content funding that supports it – have failed to represent diversity. While Canadian media industries are now making greater efforts to ensure equity, diversity and inclusion, Canadian content has a long way to go. Promoting official CanCon could risk promoting content that is not representative or fully diverse, perpetuating existing bias through algorithms.

If Canada were to step into the world of algorithmic regulation for recommender systems, there are many possible – even urgent – policy goals that could be adopted. These include algorithmic recommendations that could be required to “turn down” or prevent recommendations of illegal content such as hate speech. Regulation could require platforms to make their recommendation algorithms more transparent. Platforms could be required to publish statements on how algorithms work and what they recommend – even to report on issues of gender or racial equity in what gets recommended.  Regulation could require platforms to give users the power to prioritize the values and priorities that recommendation algorithms use.

Priorities 

If the government decides to regulate algorithms, there are lots of places to start. Algorithms can discriminate, recommend hateful content and polarize. They can sully brands and they can be used to profile, manipulate and track. Regulation could make algorithms transparent. They could intervene to address the worst form of recommendations, give people ownership, control and insight, and stop targeted ads.

There are many ways to promote discoverability, such as investing in, or providing tax incentives for, the promotion of Canadian content. In relation to recommendations and algorithms, “discoverability” should mean opening the black box, allowing us to discover how recommendation algorithms work and creating mechanisms of accountability. As Canada moves into the world of regulating algorithms and discovery, “discoverability” should be rethought.

This article first appeared on Policy Options and is republished here under a Creative Commons license.