Dalia Hatalova Blog Post # 2/15

I was interested in this article that dates back to 2015 and concerns itself with Google's algorithmic mislabeling of Black people as gorillas at the time. It is an interesting illustration of the possible biases of algorithms, so I'm retracing steps to last week's texts. For Taina Bucher, "algorithms, understood as forms of organizational logic, some with specific affordances that both enable and constrain" (Bucher 23). These constraints are apparent in Google's mislabeling, which was a result of the algorithm's inability to adhere to an organizational structure. The author of the Wall Street Journal article shared here (though there were many other publications covering it at the time), Alistair Barr, writes, "machine-learning systems don't understand the difference between mistaking a chimp for a gorilla, which may be OK, and mislabeling a human as a gorilla, which is offensive" (Barr). As such, in addition to failing to categorize correctly, the algorithm may also fail on account of not realizing the emotional impact of certain mislabeling. Barr's conclusion points toward the fact that "machine-learning systems often reflect biases in the real world. Some systems struggle to recognize non-white people because they were trained on Internet images which are overwhelmingly white" (Barr). Similarly, Ed Finn warns that "algorithms and their human collaborators enact new roles as culture machines that untie ideology and practice, pure mathematics and impure humanity, logic and desire" (Finn 47). However, the additional component is that while more substantial data could have prevented this algorithmic failing, it was not necessarily reproducing an existing mislabeling due to cultural prejudice but creating a new one through a failed algorithmic organization. Tarleton Gillespie writes, "when we turn over the provision of knowledge to others, we are left vulnerable to their choices, methods, and subjectivities" (Gillespie). Google's mislabeling highlights the potential perils that lie in trusting our imperfect machines to create meaning. José van Dijck writes, "non-human elements, like algorithms, affect how people act and how they are controlled" (van Dijck 151). While their organizational abilities will grow, errors will nevertheless make mistakes and errors that will be harder to spot but nonetheless continually extant, making it essential to view machine algorithms as other than perfect and objective organizational structures.