Latest Posts

Algorithmic Play Artefact

AI, artefact, Block 3: Algorithmic Cultures, Education

For my little piece on Algorithmic Play, I decided to look at the Instagram AI and how it assesses my kids on a computerlikes and dislikes. And from there, how it suggests to me, people and businesses to like in the future. I focused mainly on the “people feed” of my Instagram account.

The link to the short (7min) presentation is here.

I covered how Instagram is owned by Facebook and that they share algorithmic code between both companies, as well as user contacts and profiles. I discuss how Instagram makes suggestions for me based on my search history, my demographic, and my previous likes. I note how paid subscriptions are also entering my feed and being cross referenced with demographic and search history, and how all of of this demonstrates an“increased entanglement of agencies in the production of knowledge and culture” (Knox 2015)

Also attached is the transcript of the audio presentation below.

transcript Instagram Algorithmic Play

Instagram popularity model algorithm

Block 3: Algorithmic Cultures, Lifestream, machine learning

The most interesting thing about the Instagram popularity algorithm that pulls images from a dataset of 10,000 images of faces, is that NONE of the faces are real; they are created by computers. The algorithm scores to these faces according to how beautiful it considers them and the predictability of those images being ‘liked’ by other users. None of the faces are real, they were created by an algorithm that looks for predefined signs of beauty;  big eyes, rosy lips and mostly female. It assembled these faces according to what it predicts will be seen by humans as attractive.

random face

The second most interesting thing about this algorithm is that the above image is the top ranked face from the set of 10,000 machine generated random faces.  I was a bit surprised that this is deemed to be the ‘most influential’ photo, because it has noticeable imperfections, her teeth are not great, the left side of her jaw is swollen. The popularity score is designed to predict which photos will be the most liked on Instagram, and these might not necessarily be the most realistic photos or attractive faces. In addition, as the video below explains, the 32 highest scoring photos alongside this image are mostly female and mostly Asian. The theory is that the researchers from the university of Hong Kong who trained the Instagram popularity model, used a data scraping method that collected too many Asian photos, too many females, and was therefore biased.

 

How does Instagram use AI?

AI, Block 3: Algorithmic Cultures, instagram, Lifestream, machine learning, Youtube

AI inside Instagram is used in 3 ways-

1) newsfeed (sorting your posts),

2) Targeted advertising based on your demographic and what you ‘liked’ in the past and

3) Deep learning-to manage community moderation.

Deep learning looks at words in the comments in posts, it groups the words together and considers what is good and bad text. Bad text might be what it considers trolling, hate speech or words associated with cyberbullying. The algorithm is a closely guarded secret, we do not know how the model works or what type of comments they are targeting or how many comments are being removed. Therefore we don’t know how biased it is, and how much this algorithm is “used to coerce, discipline, regulate and control: to guide and reshape how people, … interact with and pass through .. systems”. Kitchin 2017

Kitchin, R. (2017) Thinking critically about and researching algorithms, Information, Communication & Society, 20:1, 14-29, DOI: 10.1080/1369118X.2016.1154087

 

Doubts About Data-Driven Schools

AI, Articles, Block 3: Algorithmic Cultures, Education

As we all know, the collection of data on students has become commonplace, from grades, state test resumescores, attendance, behavior, lateness, to graduation rates. Each year however, this information becomes ever more detailed with more and more data points being collected per student.

Educational transcripts, unlike credit reports or juvenile court records, are currently considered fair game for gatekeepers like colleges and employers. These records, though, are getting much more detailed, and how will this influence a student’s ability to gain access to these institutions in later years?

Article Complet : https://www.npr.org/sections/ed/2016/06/03/480029234/5-doubts-about-data-driven-schools
via Pocket

New York City Moves to Create Accountability for Algorithms — ProPublica

AI, Articles, Block 3: Algorithmic Cultures, machine learning, Pocket

This article discusses the development of an ‘accountability bill’ in the US, which aims to punish different races of peoplecompany algorithms that appear to discriminate against people based on age, race, religion, gender, sexual orientation or citizenship status.

Whilst it would be a good law in theory, in reality, it would be very hard to implement. Algorithmic systems are so difficult to break down, designed to be increasingly complex, gathering millions of data points and  “woven together with hundreds of other algorithms to create algorithmic systems” (Williamson 2017). Added to that,companies are very secretive about how their models work and what type of parameters make up the design of the algorithms, so that “the rules generated by (algorithms) are compressed and hidden”  (Williamson 2017). Discrimination would be very hard to prove as being deliberate in these cases, but hopefully it will encourage scientifically sound data builds, validated in appropriate ways, and to eventually make them more transparent to the public.

Article Complet : https://www.propublica.org/article/new-york-city-moves-to-create-accountability-for-algorithms
via Pocket

Williamson, B. 2017. Introduction: Learning machines, digital data and the future of education (chapter 1). In Big Data and Education: the digital future of learning, policy, and practice