Showing posts with label Broadcasting. Show all posts
Showing posts with label Broadcasting. Show all posts

Wednesday, February 7, 2024

Algorithmic Gender and Race Bias: a Canadian Broadcast Policy Concern?

My new article "Discoverability and Algorithmic Recommendations in Video Streaming Platforms:Algorithmic Gender and Race Bias as a Canadian Broadcast Policy Concern" with Fizza Kulvi, Faiza Hirji, Manveetha Muddaluru, Emmanuel Appiah, Leandra Greenfield, Erica Rzepeci, and Christine Quail recently came out.  

Our article examines discoverability and algorithmic recommendation systems in video streaming systems like Netflix or YouTube. We were particularly interested in looking at gender and race bias in algorithmic recommenders, and how concerns about those biases were seen:

1) by creators, policymakers, and industry members, and 

2) by parliamentarians in debates about Canada's Online Streaming Act.

We found that while algorithmic video streaming recommenders are often portrayed as neutral and as responding to users' interests by parliamentarians and video streaming platforms, our interview participants - and research on recommendation systems - are far more skeptical.


Tuesday, June 7, 2022

The Online Streaming Act (Bill C-11) claims to level the playing field … but for whom?

The Online Streaming Act (Bill C-11) claims to level the playing field … but for whom? Nawshaba Ahmed and I ask, in our piece published in The Conversation.

 See also my comments to the Heritage committee on algorithmic recommendations and C-11.

Wednesday, June 1, 2022

Comments to the House of Commons Standing Committee on Canadian Heritage on Bill C-11

44th Parliament, 1st Session
Tuesday, May 31, 2022, 6:30 p.m. to 8:30 p.m.
Watch on ParlVu

My name is Sara Bannerman (she/her). I’m a Canada Research Chair in Communication Policy and Governance at McMaster University.

Today I will focus on discoverability and algorithmic bias. 

Governments around the are working on measures to ensure that algorithms are accountable.

There is a common misconception that streaming platforms’ recommend what the user wants to see.

Platforms show us what they want to show us. They show us what will keep us watching ads, purchasing advertised products, contributing data, and subscribing.

Platforms are not neutral.  They serve their business interests. As well, there are three types of bias I am concerned about. These biases can affect both users and content providers. 

First, there could be bias if algorithms are used to select content for carriage on a streaming service--

  • predicting how many viewers the content will attract. 
  • A poor algorithmic showing could sink a content provider’s chances of being shown.
Second, there can be bias in the recommendation algorithms that users use to discover content.
  • Recommenders often display popularity bias, recommending what’s popular and concentrating users’ viewing on a smaller catalog of content.
    • This can be unfair to artists in the “long tail” and to users who like something other than popular content.
    • It could be unfair to Canadian content—including user-generated content,

Third, users’ own biases can be amplified
  • Beyond users’ biases towards or against Canadian content, if some users have a gender bias, for example, this could be amplified in the recommendations that respond to past viewing habits.
  • Such biases can form a feedback loop that can spread throughout the system.

The research in this area is only recently developing.

CRTC intervention in the algorithms raises many difficult problems.  It may not be the first, most likely, or best answer to them.  The CRTC has said, today, it does not want to play that role. But the Commission could play a role in bringing such problems to light.

There are concerns that requiring discoverability would infringe on freedom of expression.
Streaming services’ user interfaces and recommendations may be forms of expression, and, if so, regulatory interventions could constitute a limit on that expression.
There are legitimate concerns that promoting some content could mean demoting other content, among other concerns.
Sometimes limits on expression are justified, but they must be justified.
To understand whether any justification exists, or even just to understand, we need data.

It may be that the best role for the CRTC will be to monitor and call attention to problems, not just with the discoverability of Canadian content, but also with recommender biases relating to other Canadian values, so that civil society and others can intervene.
The CRTC can only do that if it has data.

The provisions on information-provision and information disclosure in the bill are important to the study and examination of discoverability algorithms and data, and to the CRTC’s potential work with outside organizations on this matter. 

It may be necessary to require platforms to collect certain data to permit these examinations to happen.
  • The general powers section of the Bill could include (under n and o) the phrase “collection and provision”
  • The general powers section of the Bill could name information on discoverability as information the Commission can demand.
I strongly disagree with proposals that would allow a company to prevent the disclosure of information in section 25.3.[1]

The Canadian broadcasting system has often served dominant groups.
It has also been open to change and improvement based the work of civil society and others.

We need to ensure that the discoverability mechanisms of online streaming platforms are also open to critique and change through public transparency, debate, and data. 

Thank you.



[1] See Appendix for details


Appendix

Amendments to C-11, for consideration

 9 (1) (n) the collection and provision to the Commission […]

9 (1) (o) the collection and provision to the Commission […]

9 (1) (o) (v) information related to discoverability; and [NEW]


Comments on Netflix’s Proposed amendments to Bill C-10, for consideration

I object to Netflix’s proposed amendments to Bill C-10 section 25.3, which could equally be proposed to be applied to C-11.

If their proposed 25.3 (4 and 5) (a) (a.1) is adopted (I do not support this), it should read “must not knowingly be publicly disclosed or made available […].” This would permit the CRTC working privately with groups to examine discoverability mechanisms and other matters.