Overview
One of the most memorable points I retained from this talk is the concept of “datafication” how all of our services and our public service platforms are all now digitalized, by extension making all our inputs available as data points. The speaker argues that since there are more data, the algorithmic decision-making process becomes much more inaccurate due to the reductive nature of creating a decision from this significantly larger data set. For my reflection I wanted to look further into what is algorithmic decision-making, and how “datafication” feeds into this.
Datafication
Datafication by definition of Mayer Schoenberger and Cukier is the process of transforming our social interactions into quantifiable data, which then becomes trackable to consumer platform for optimization of prediction algorithms (Cukier & Mayer-Schonberger, 2013). I can see examples of this in my everyday life, an easy one is Facebook Marketplace, this app would always show me secondhand items being sold within the vicinity of my area, even show suggestions of items I usually browse for. I can only assume my views are being tracked to determine which items I am looking for.
As shown above, I’ve recently searched for a car and some TV sets, and these are the results available first to me, with my exact area listed. This brings me to the next section, if all these data points are available online, they must go through a system to provide these recommendations, that would be algorithmic decision-making.
Algorithmic Decision-making
According to Breidbach, algorthmic decision-making is the process of machine automatically completes task or make decisions for us (Breidbach, 2024). The decision making process usually comes as a result of advance machine learning or prediction models. In order for the accuracy of these model to increase, more data point is needed, as shown in this study where adding 10000 samples increased accuracy from 85% to 92% in a classification model (Academy, 2023). In order for these decision-making processes to improve in accuracy and predict our behaviors more, more data needs to be collected, leading to a vicious cycle of data collection. At this point I inferred that privacy becomes less of a concern, and only data volume becomes of importance to service providers, not to mention data quality can varies from collection point.
Another major point that the speaker mentioned is accountability, since algorithmic decision-making is also present at the public service level, who would be help accountable if a valid claim is rejected due to system error?
Reflection on Privacy
As I looked further into the topic, the more I believe there must be an ethical framework created to bound data collection from encroaching too far into privacy. Despite all the benefits that datafication and algorithmic decision-making can provide to consumers, this is done at the cost of their own privacy and corporate accountability.
References
Academy, E. (2023, August 8). How does adding more data to a deep learning model impact its accuracy? – EITCA Academy. EITCA Academy. https://eitca.org/artificial-intelligence/eitc-ai-dltf-deep-learning-with-tensorflow/tensorflow/using-more-data/examination-review-using-more-data/how-does-adding-more-data-to-a-deep-learning-model-impact-its-accuracy/#:~:text=The%20additional%20data%20helps%20the,positive%20impact%20on%20its%20accuracy.
Breidbach, C. F. (2024). Responsible algorithmic decision-making. Organizational Dynamics, 53(2), 101031. https://doi.org/10.1016/j.orgdyn.2024.101031
Cukier, K., & Mayer-Schonberger, V. (2013). Big Data: the essential guide to work, life and learning in the age of insight. https://digilib.umsu.ac.id/index.php?p=show_detail&id=19457
Promoting greater fairness, accountability, and transparency around algorithmic Decision-Making in content moderation. (n.d.). New America. https://www.newamerica.org/oti/blog/promoting-greater-fairness-accountability-and-transparency-around-algorithmic-decision-making-content-moderation/
Leave a Reply