Subject: Analysis of a Professional Journal
Date: Mar, 31, 2016

Introduction
I am evaluating a journal article titled: Algorithmic accountability. The article was published in Digital Journalism, in November 7, 2014. The author of the article is Nicholas Diakopoulos from College of Journalism, University of M aryland .

Evaluation
This article focuses on the concept of “Algorithmic Accountability Reporting” as a way of investing biases and influences employed by algorithms in todays society and how new age computational journalists should approach it. This article is directed at journalists who are scrutinizing algorithms to understand biases and false analysis portrayed by algorithms.

The article is well structured. Text is organized in coherent sections which logically connects the entire article. The article starts with the brief introduction which outlines the points which will be covered in the article. Introduction also answers the questions of – what is this article about, who is the target audience, what are the current issues faced in journalism and how methods described in this article will help address those issue.

Author, then mentions few real world examples of software companies which collect user data and then build ingenious algorithms to classify, group and eventually target people for their benefits – and how in doing so – they often open risks and flaws. The author exposes potential flaws by raising very valid questions about the decisions made during the development of algorithms. These questions hook the reader and draws readers’ attention to the purpose of the article.

On the issue of transparency of algorithms, the author points out that while some organizations are inherently transparent, some need to be pushed by the government. He cites “voluntarily examples” like Google’s biannual transparency report and “Government mandated” reports like restaurant inspection scores, automobile safety tests score. He notes that transparency is often “limited” – due to competitive dynamics. Government incentives are often not enough. Sometimes policies exist, yet their enforcement is rickety. The author has well researched this part and provided ample examples. Companies do not expose their trade secrets and won’t leave their services open to gaming and manipulation. Consequentially, the author emphasizes on the role of 3rd parties in Watchdogging. “Watchdogging” or “Investigative Journalism” – on “Algorithm Accountability” is the key focus area – of this article. The author wants more and more people (journalists) to take interest and groom their skills to hold algorithms “accountable”.

In the next section, the author proposes using methods like “Reverse Engineering” for “Algorithm Accountability Reporting”. It is the process of extracting the knowledge or design blueprints of a system – by rigorous examination, drawing on domain knowledge and deduction. The author describes algorithms as “black-boxes” – which take an input and produce some output. He suggests that by varying the inputs in enough ways and paying close attention to the outputs, one can start piecing together a theory of how the algorithm works.

However, there are often instances where inputs to algorithms are not known, just a final set of results are known (e.g. University rankings list). For such scenarios, the author encourages journalists to interview or initiate dialogs with designers of systems, occasionally do some document reviews and actively track policy changes of company to get insights into the motives / intentions behind algorithms.

I find his approach quite logical. A mix of reverse engineering and moral policing through interviews can act as a nice check and balance for maintaining algorithmic accountability of widely used systems.

To support the theoretical component above - the author himself investigates 4 topics and shares his findings - based on own black-box testing he performed and dialogues he initiate with eminent people in the domain. This shows the conviction with which the author applies his theories. The topics for his case-studies are: “censorship in search-engine autocomplete algorithms”, “Targeting Political Emails”, “Price Discrimination in Online Commerce” and “Executive Stock Trading Plans”. He faces several challenges in investigating these topics and finds innovative ways to overcome them and pursues his goal of “understanding” the algorithms. In some cases, he is successful and in some he isn’t. Any case, his results are interesting and gives reader a lot of points to ponder.

To conclude, the author proposes a methodology – to investigate algorithms (for accountability). It involves identifying a meaningful target, sampling the algorithm, and finding the story. He provides an ample list of questions that an investigative journalist should purse – and has also given advice to journalists - for when they are in a predicament about which leads to follow. The author has linked contexts back into his case-studies –which makes it a good read and bolsters his points.

He finally, concludes the article by summarizing the main topics addressed in the article and opens up a discussion about how journalists should train themselves to carry out algorithmic accountability investigations and navigate legal, ethical, and technical challenges in doing so.

Conclusion
Overall, I feel the author conveyed his points clearly and demonstrated expertise by himself applying investigative techniques to study a few algorithms. He immaculately used real world examples, to elaborate his views and raised intriguing questions at crucial points. Throughout the article he maintained a relationship with the reader by summarizing and hinting at things to come. His open ended discussion about the future - towards the end is thought-provoking and I enjoyed reading it.

Reference
Nicholas, D. (2015). Algorithmic Accountability. Digital Journalism, 3(3), 398-415. doi: 10.1080/21670811.2014.976411