My YouTube algorithmic play artefact is presented here in the form of a preamble and a three-part timeline. Screenshots have been presented below. Real-time videos and high-res screenshots can be found in my lifestream blog (LB).
Preamble
Prior to setting up a new YouTube account, I wanted to try out an experiment I first learned about last month. My mobile phone (in standby mode) in hand, I engaged two willing friends in a lengthy conversation about our experiences with Netflix. My aim was to ascertain whether my mobile was eavesdropping on us.
Part 1 – Tabula rasa
11.36 a.m.
Conversation over, I set up a YouTube channel choosing the US as my location.
11.41 a.m.
Here we can see how rapidly YouTube’s algorithm has kicked in. Never having inputted any data, I was eager to see what I was being presented with. The first thing that caught my eye was an advert for Netflix; my mobile had indeed been eavesdropping and, only a few minutes after my friends and I had ended our conversation, it was clear that it had surreptitiously passed its intelligence to YouTube.
Below the Netflix advert appeared a trailer for ‘Wonder Woman 2’, not a film I would choose to watch, but clearly the algorithm (through my earlier mobile activity) had identified that I was interested in superheroes; the focus of my recent MOOC.
11.46 a.m.
With no personal input of keywords to inform it as yet, YouTube’s other suggestions (a Sean Connery interview, an essay memorisation framework, the Kabs family and a 911 call for a pizza delivery) were all random. A potential exception was a video of tourist mistakes in Vienna prompted by my LB blog posts on the city a few weeks ago. (This is not something I would have watched.) The video featuring the second movement from Berlioz’s ‘Symphonie Fantastique’ had also been informed by my earlier posting of a YouTube video of the same piece of music in my LB.
Part 2 – Feeding Data to YouTube
11.50 a.m.
I now start to input keywords related to spilling red wine on a carpet and receive instant recommendations.
I watched several of the suggested videos; some in full some in part to intentionally influence the algorithm.
Part 3 – Results
11.57 a.m. – 12.09 p.m.
Only 20 minutes after launching my channel, the results (accompanied by adverts for Dyson products and for carpet cleaning services in London – my location at the time) come thick and fast accompanied by a slew of uninvited (and what I initially felt were untargeted) ads. On closer investigation the ads were all thematically home and lifestyle related. The ad for the Sony Xperia mobile may have been prompted by my use of an iPhone for my search.
Carrying on the theme, I searched for carpet cleaning, cleaners and for the latest, cheapest, best vacuum cleaners.
12.14 p.m.
From showing me adverts on vacuum cleaners and associated services, the algorithm now moves one step further and helpfully tells me where I can buy what I’m looking for; John Lewis.
Two-days later, Netflix, Dyson and videos on carpet cleaning now feature prominently on my channel.
1 How has the algorithm affected the options you are given or what you can see?
By setting up a new YouTube channel, I was curious to see what videos and adverts would initially be selected. I was not surprised that my mobile had listened in to my conversation on Netflix and had instantaneously informed YouTube. The only videos in the initial selection that drew upon what the algorithm felt to be my ‘interests’ (one music and one touristic) were drawn from earlier LB posts.
From helpfully showing me videos of how to clean a red wine stain on a carpet, the algorithm then attempted to influence the type(s) and range of materials and vacuum cleaners that I could buy to solve my problem through sponsored masthead adverts and a limited range of products and related services from YouTube advertisers. These may not necessarily have solved it, but their inclusion and prominent ‘calls to action’ were clearly intended to influence what and even where to buy.
2 How have your actions changed what the algorithm has done?
By consciously choosing the length of time I would spend watching a video I knew that my decisions would change any future suggestions.
3 Have other people been involved in shaping results?
Yes, people such as YouTube viewers, advertisers, sponsors, Millennial YouTube stars, and YouTube algorithm engineers have helped to shape results.
4 Do the results feel personal or limiting?
I found the results limiting here, but I also expected them to be so. I appreciate that YouTube’s raison d’être is to maximize revenue via advertising services.
5 What are the ethical issues at stake with your chosen algorithm?
A major ethical issue is that YouTube’s algorithm was not designed to help visitors find videos of what they’re looking for but, according to Chaslot (a former YouTube AI engineer), to get them addicted to YouTube. He also claims that YouTube has promoted, among other things, ‘terrorist content, foreign state-sponsored propaganda, extreme hatred, […] inappropriate kids’ content and innumerable consipracy theories’ (2019).
Is there data here that should be private?
Yes. Data harvested from users of platforms such as YouTube is often sold on to third parties without the users’ consent. This begs the question as to who then bears responsibility for possible data misuse or inadequate data protection. The development of privacy-preserving practices should therefore be prioritised.
In September 2019, Google and YouTube were fined $170 million to settle allegations by the US Federal Trade Commission and the New York Attorney General that YouTube’s video sharing service illegally collected personal data from children without parental consent.
6 What are the implications for digital education implied by your chosen algorithm?
Used judiciously, the YouTube algorithm can positively support digital learning. According to Deutsche Welle, a recent study by Krüger (2019) suggests that parents and teachers underestimate the time spent by German 12-18 year-olds on YouTube to help with homework, exam preparation and to support classroom learning and, importantly, its untapped potential as a learning tool.
In addition to its use in flipped classroom scenarios, I can see the YouTube algorithm helping course designers to craft personalised learning programmes, to foster independent and self-directed learning, and to encourage student and teacher creativity.
It is also helpful to teachers and lecturers unable to attend professional development events such as the annual ALT (Association for Learning Technology) conference to be able to access its proceedings which are freely available on its YouTube channel.
In HE, universities could use platforms such as Panopto to video lecture series with the aim of curating a repository (accessed through a VLE such as Canvas) of core recordings for any one programme. This could also be supplemented by suggested related videos in the same subject and by relevant interdisciplinary videos. Panopto helpfully allows for the embedding of interactive activities such as quizzes and polls. PowerPoint slides and screen content can also be shown alongside videos. Panopto also offers an analytics feature which enables tutors to monitor video selection and watch-time. This would help tutors in updating, reviewing and removing content.
To encourage a multimodal approach to learning, links to related reading materials, podcasts and visuals could be added by tutors. Suggestions for additional multimodal resources should be invited from students and tutors, reviewed for currency and, if relevant, added to the repository.
For assessment purposes, students could set up individual video streaming channels, share content with peers and engage in tasks such as individual/paired and small group video compositions.
Readings
Brody, Ben and Bergen, Mark (2019) Google to Pay $170 Million for YouTube Child Privacy Breaches. www.bloomberg.com. Available at: https://www.bloomberg.com/news/articles/2019-09-04/google-to-pay-170-million-for-youtube-child-privacy-breaches. Accessed 13 March 2020
Chaslot, Guillaume (2019) The Toxic Potential of YouTube’s Feedback Loop, Wired. Available at: https://www.wired.com/story/the-toxic-potential-of-youtubes-feedback-loop/ Accessed 12 March 2020.
DW (2019) YouTube in Schools: A Digital Revolution in the Classroom. www.dw.com. Available at: https://www.dw.com/en/youtube-in-schools-a-digital-revolution-in-the-classroom/a-49049423. Accessed 12 March 2020.