User:Anuradha reddy
Algo News - The Space Between Contemplation and Execution
Algo News is a by-product of explorations into news, language and algorithms. We constantly rely on machines to make accurate decisions for us, whether we are purchasing airline tickets, trading stocks or searching for Chinese takeaway. Our reliance on machines has left us being unmindful of its potential, where even the slightest inaccuracy is readily dismissed as a malfunction or as ‘bad data.’ Algo News subverts this notion by putting the machine through various stages of malfunction. The machine is programmed to source news content through different media channels with an algorithm that repeatedly auto-generates new stories of each news item. The new content is published back into mass media channels via Twitter and the newspaper. Through these misrepresentations, Algo News opens up spaces for contemplation on machine-thought and its execution against a backdrop of our own intentions for the machine.
Luciana Parisi (2013) claims that randomness has become the condition of programming culture. In her view, exposing randomness in a program does not necessarily justify cultural and political biases. However, I argue that it plays a crucial role in sensitising us to different modes of computational thought and in generating new possibilities. In this sense of the word, it even breaks open existing ideas behind algorithmic expectation. Similar investigations of machine-learning systems are common among artists experimenting within the boundaries of design, politics, literature and culture. For instance, James Bridle coined the termed ‘The New Aesthetic’ to highlight the increasing appearance of the digital in our physical environment (“The New Aesthetic,” n.d.). According to Bridle, machine glitches, misrepresentations and unintended occurrences (e.g. rainbow plane, “On the Rainbow Plane | booktwo.org,” n.d.) act as agents that sensitize us to the visions of the machine and its interpretation of our world.
Drawing from literature and contemporary discussions on machine learning, it is becoming increasingly crucial to highlight the algorithmic processes behind such systems. Several examples illustrate how this could be done. Khovanskhaya et al (2013) designed a web interface that reveals what it learns from a user’s browsing habits. In doing so, it encourages the user to playfully interrogate the system by manipulating data, and to interpret what the machine learns from those changes. Another approach is to automate an algorithm or a ‘bot’ that interacts with other entities in a network. For example, Rob Dubbin created a Twitter bot named ‘Olivia Taters’ under the disguise of a young teenage girl (Madrigal, A. C., 2014). Taters’ often conversed with real people and other bots like her. By following her conversation, one might gather a glimpse of the nature of the algorithm i.e. the manner in which it picks up on random ‘keywords’ from other Tweets. In this way, designers might attempt to open up new possibilities for understanding and interacting with machine-learning systems. The experiments could be designed in such way that it exposes inner workings with unexpected behaviors or outcomes.
Early into the prototyping process, experiments were conducted using existing tools such as Google Translate, X-ray Goggles, IFTTT and so forth. I have chosen to elaborate on one of the experiments that became crucial for the remainder of the project. Google Translate is a popular multilingual translation service that uses machine intelligence to learn from human-translated text content on the web. It attempts to make intelligent approximations of the original sentence through context-recognition and pattern matching. However, it is rather difficult to explain why the machine chooses certain words or sentences against many possibilities. The experiment entailed using the Google Translate algorithm to translate a news article in English to another language and back into English. The outcome showed slight variations to the original meaning of the news content. When this experiment was repeated several times across several languages, the new content lost its original meaning on the whole and a new story emerged. It was reminiscent of the popular kids game ‘Chinese Whispers’, which involves re-telling a story from one source to another, gradually turning nonsensical or amusing to hear. The results from the Google Translate experiment were ambiguous, comical, and yet revealing of its learning mechanism i.e. providing subtle hints in the way it made intelligent guesses through the pattern recognition algorithm. These experiments were conducted locally up to this point and no programming was involved. As the results from the experiment seemed promising, a prototype application was created with the aim to auto-generate news stories using the Google Translate machine-learning algorithm.
The prototype for this piece originated via Twitter. Twitter is a micro-blogging service that relies heavily on machine learning based recommendation algorithms in order to recommend followers by learning from its users. Even though Twitter is meant for real people, there are a fair number of automated twitter bots that actively post Tweets and respond to previous Tweets. Moreover, keywords are crucial for triggering interactions between users. This insight led to choosing Twitter as a plausible platform for publishing mistranslated news items which could proliferate and trigger interesting reactions. As a result, a Twitter profile titled ‘Algo News’ was created which would post the result of 10 mistranslations (via Google Translate) as a Tweet. The original news was sourced from ‘The Guardian’, which takes a rather neutral tone compared to other news sources. The idea was to assume a neutral position in the initial stage that could assist interpreting the machine’s doing after several mistranslations.
An automated algorithm that Tweets mistranslated news items was created using the GoSlate and Twython API. As a result, ‘Algo News’ rapidly started receiving attention through the keywords it generated. It produced about 400 tweets, several of which were retweeted and ‘favorited’ by unfamiliar users (possibly bots as well). However, Algo News' followers continue to fluctuate until this day. The outcome of the algorithm was also materialised as a newspaper where content was not just text-based but it also included imagery from a Flickr API fused into the algorithm. In doing so, it presented the opportunity to reflect on the state of news consumption and the implication of involving machines as actors in the process of disseminating mass media.
One of the key takeaways from undertaking this work was the crucial role of experimentation and prototyping in interrogating black-boxed systems. It was during the process of prototyping that several biases and challenges were brought to the forefront. To begin with, it is problematic to determine if sentences are modified due to change in grammar or if the algorithm recognizes a shift in the context. For instance, in the table below, the word ‘Alibaba’ is mistranslated to ‘Pope’. Alibaba refers to the Chinese e-commerce company which the algorithm failed to recognise. Anyone might choose to overwrite the algorithm to make the right translation but it also implies that the algorithm can be intentionally trained to understand a false statement.
Language Translation
English Alibaba: China’s answer to Amazon makes £4.4bn thanks to Singles' Day (ORIGINAL)
Chinese Alibaba: China's answer to the Amazon £ 4.4 billion, due to the Singles' Day
Japanese Alibaba: China's answer to the Amazon £ 44 億 is, for the singles of the day
Hindi Alibaba: 44 億 £ China's answer to the Amazon for a single day, is
Arabic Alibaba: response 億 £ 44 Amazon China in one day, it
Russian Alibaba: response 億 Amazon China 44 pounds in one day
German Alibaba: response 億 Amazon China 44 pounds in one day
Spanish Alibaba: Answer 億 Amazon China 44 pounds in one day
Italian Alibaba: Response 億 Amazon China 44 pounds in one day
French Pope: response 億 Amazon China £ 44 in one day
Portuguese Pope: 億 response Amazon China 44 pounds in one day (TWEET)
Similar results exemplify the possibilities of interpreting how machines execute programs through such prototypes. This prototype, in particular, resulted in ambiguous, humorous and witty outcomes that became sites for contemplation about how we might leverage from differing perspectives brought forth by machines. In this way, the notion of ‘bad’ data has been subverted and made useful in engaging people with machine-learning data. As such, the project highlights the role of the experimentation in harbouring the distinction between our expectations for the machine and the machine's executions of it . The resulting randomness, as put forth by Luciana Parisi (2013), extends beyond programming culture and into our physical lives, into who we are, and what we know. It is indeed crucial to accept randomness as a part of our mundane lives and design ways to shape the scope and our expectations for machines through such explorations.
References
Khovanskaya, V., Baumer, E. P., Cosley, D., Voida, S., & Gay, G. (2013, April). Everybody knows what you're doing: a critical design approach to personal informatics. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 3403-3412). ACM.
Madrigal, A. C. (2014, July 7). That time 2 Bots Were Talking, and Bank of America Butted In. The Atlantic. Retrieved from http://theatln.tc/1mBxOB4
On the Rainbow Plane | booktwo.org. (n.d.). Retrieved from http://booktwo.org/notebook/rainbow-plane/
Parisi, L. (2013). Contagious architecture: computation, aesthetics, and space. MIT Press.
The New Aesthetic. (n.d.). Retrieved January 30, 2016, from http://new-aesthetic.tumblr.com/?og=1