Constructed Message: ‘This is not my idea of being able to make progress.’
Device Used: Blackberry Priv
My first attempts to construct a meaningful sentence were with the more conceptual statements. Neither ‘Education is not about’ nor ‘As a society we are’ led down any path that made a meaningful sentence. Only with what I consider to be a more narrow beginning could I get a sentence that even made sense.
Initially I used the built in texting app on my phone to try. I then wondered where the predictive algorithm lay – was it the domain of the app or of my phone? So I tried both on WhatsApp (which I have only recently added to my phone) and on my e-mail client. In both cases the predictions were identical. First mystery solved – the predictions were the responsibility of my phone – not the individual apps.
Given the limited nature of the predictions I was seeing, it is clear to me that the predictive text is primarily reliant on the things I tend to say in my own texts, messages, and e-mails. Especially in the choice of nouns, I recognized words that I use on a fairly regular basis. The thing is, I don’t really do a great deal of text writing on my phone. I tend to answer only the e-mails that require a simple response, and I infrequently text and message only a small group of family and personal friends. I seldom use my web browser and do not have apps that rely on Web content. This doesn’t leave the predictive algorithm a lot to work with. I imagine many of my classmates found much more robust responses on their phones.
I am the person who can expose the Achilles heel in the predictive text algorithm. As Cathy O’Neill has identified, this kind of algorithm works because its abilities are based the premise that it can ‘collect historical data about people, profiling their behaviour online, location, or answers to questionnaires, and use that massive dataset to predict their future purchases, voting behaviour, or work ethic” (2017). Since I provide my phone with extremely limited raw data to work with, its predictions are narrow and quickly fall apart on a conceptually large idea.
Predictive algorithms ultimately rely on a constant stream of data. They begin with a training data set, but the presumption is that the learning is continuous and that more and more data generates more and more accuracy in the result. If an algorithm is starved of data it will not learn; it will not increase its accuracy. Data starvation can come in two forms. It may be quantity, or it may be a lack of variation in the data. My phone is over 5 years old, and there have been thousands of e-mails and texts in that period of time. The thing is, there has been very little variety. A huge percentage of those are ‘Thank you’ e-mails, and ‘Hey, let’s have coffee’ texts. I have cruelly starved the predictive text feature on my phone. I admit a sin for which I offer no apologies.
References
O’Neil, C. (2017, July 16). How can we stop algorithms telling lies? The Observer. Retrieved from https://www.theguardian.com/technology/2017/jul/16/how-can-we-stop-algorithms-telling-lies