Thanks to the committee for the invitation today. It's a pleasure and a privilege to appear before you.
I am a full-time faculty member at Carleton University in communication and media studies. I teach in the areas of media and gender, law communication and culture, and algorithmic culture and data analytics on the more technical side. I'd like to share some concerns and considerations about the role of algorithms in the context of networked communications, such as those for social media, search, and, in particular, what is broadly conceived as automatic content curation by algorithms.
There's been some discussion of this already, obviously, so I'll focus on three things: defining algorithms and their operations; the trade-off between user interfaces and the increasing complexity of software; and, the impact of algorithmic content curation.
I want to be clear at the start about what I mean when I refer to an “algorithm”. In very simple terms and in the context of information systems and networked communication, it can be thought of as a series of computational steps or procedures that are carried out on information as an input to produce a particular output. For example, a search term typed in as input to “Google Search” produces an output in terms of search results.
Also, they don't operate in isolation. Algorithms are part of a complex network of digital devices, people, and processes constantly at work in our contemporary communication environment.
Embedded in any algorithmic system is a capacity for control over the information it analyzes, in that it curates or shapes the output, based on multiple factors or capacities the algorithm uses to generate the outputs. Again, in the case of Google Search, their suite of algorithms takes in the search term, personal search history, similar aggregated history, location, popularity, and many other factors to generate a particular set of filtered results for us.
The rather amazing thing about any of the algorithms incorporated into our contemporary communication is that these computational systems know much more about us than we know about them. They're often mysterious and non-transparent, as has been mentioned: a black box that governs our information landscape, persistently at work to shape information flows, determining what information we see and in what order we see it, and then nudging us towards certain actions by organizing our choices.
Algorithms do govern content automatically, but they do so because they have been designed that way. The capacity of algorithms to curate or sort information has been designed to sit behind the user interface of our popular search and social media applications, so we don't directly interact with the algorithm. Curation and filtering of information is sometimes something that we can see happening, but it's not entirely clear how it is happening. For example, the simplification includes things like swiping and tapping, and clicking icons in our mobile apps—highly simplified behaviour.
The extraordinary complexity of algorithms in automated curation is thus deeply hidden in the software and digital infrastructure necessary for networked communication, and this leads to a sort of distancing effect between us as human users and the complexity in the systems we are interacting with, such as Google Search, for example. It becomes difficult for us to connect our simple button choices or search queries to any wider effect. We don't necessarily think that our own individual actions are contributing to the ranking and sorting of other information searches or the popularity of a particular newsfeed post.
Social media companies tell us that reaction buttons like “Like” and “Don't Like”, or love or angry icons, are a way to give feedback to other users, stories, and posts, and to connect with the issues, ideas, and people we care about, but this effectively trains us to input information that feeds the algorithm so that it can generate its output, including ranking posts and shares based on these measures.
I was recently reminded of the powerful ways algorithmic curation happens. In the context of a group of Facebook users making a few original and offensive posts, the situation quickly escalated over a week, and hundreds of reactions or clicks on all those “like”, “angry”, or “haha” buttons continually moved up that cyber-bullying incident in people's newsfeeds. As Facebook itself notes on the relevancy score of a newsfeed algorithm, “we will use any Reaction similar to a Like to infer that you want to see more of that type of content”. These simple actions literally feed the algorithm and drive up the issue.
I also find Google's auto-complete algorithm even more troubling. While Google likes to make grand public assurances that their auto-complete algorithm—the drop-down of suggestions you see when you're searching—is completely objective and won't link personal names with offensive auto-completes, it still drives users to problematic content via its complex and comprehensive knowledge graph.
Google's knowledge graph combines search results in one page with images, site links, stories, and so on, but it still combines information that is problematic. For example, the Google auto-complete algorithm still points us to details of the late Ms. Rehtaeh Parsons' horrific case that were propagated by Internet trolls and continue to feature in Google's “searches related to” suggestions that appear at the bottom of the search page, pointing to images and other problematic content.
Recent changes to automated curation techniques point to our need for sustained efforts to build digital literacy skills, as discussed earlier, that steer young people into thinking more critically and being ethically minded in terms of what's going on. I would argue that we also need, then, a specific effort to educate young people about what algorithms are, not in their mathematical complexity, but generally how it is that they're operating with these simplified user actions that young people are so eager to participate in.
Visibility and publicity, and shares and various Snapchat scores are part of the new social accounting that young people value, and it's driven by an increasingly subtle yet complex infrastructure: an algorithmic milieu of communication and control that leaves very little in the hands of users.
Algorithmic sorting, ranking, and archiving is persistent and ceaseless. It churns away continuously as social media and search users navigate, click, view, search, post, share, retweet, @mention, hashtag, and react. As users, these actions and immediate results feel dynamic and vital. At its best, it affords us efficiencies in information retrieval and communication, and at its worst, it amplifies some of our most problematic and prejudicial expression, action, and representation online.
Thank you. I look forward to your questions.