"I Tested a Popular A.I. Slop Detector. It Felt Empowering." Article Reflection No. 195 (11/7/2026)
Reflection:
In The New York Times article “I Tested a Popular A.I. Slop Detector. It Felt Empowering.,” journalist Brain X. Chen discusses his experience with an AI-usage resource that can potentially discern individuals’ writing compared to that by AI technology. This resource is called Pangram and, although it is claimed to be relatively effective in (e.g.) identifying when AI has been used, it still falls short of being able to discern whether an image has been created by AI technology or by humans themselves. Chatbots using AI involve the use of a “‘decision tree’” system in which one choice made by the chatbot further informs a subsequent choice, and so forth, according to the article.
About the part from the article about the accuracy of the tool in identifying—as writing created by an individual—Charles Dickens’ works, I wonder about whether it’s possible for the AI to be taking into account how Dickens was an “actual” real-life author, the context of how he did not live during an age of AI, and how his works are widely recognized (recognized in a way that is not associated with AI). I think this idea is contrasted by another development discussed in the article, specifically regarding how Pangram was able to accurately distinguish AI-generated writing and the writing of the journalist himself. However, I still wonder about the potential influence of these aforementioned points.
Another question that came up for me as I read this article is this inquiry after how the content of a photo may influence the “quality” and accuracy of AI detectors. For example, a part of the article discussed how Pangram could not check whether certain images—specifically, ones pertaining to (e.g.) violence—were generated by AI. I found this interesting because images conveying strong emotions, such as the types of images discussed in the previous sentence, are ones that I believe are more important in terms of the effect of determining whether they are generated by AI or not. These types of images are the ones that most provoke emotion and thus, have the higher potential of being spread widely. Given this, I wonder about the reason behind how and why the tool is unable to apply the tool to these types of photos. Is it in part due to the risks that are involved, such as the risk of incorrect judgement in this context?
Additionally, through a quote from Hany Farid, a professor from Dartmouth University, the article seems to emphasize on the importance of recognizing and keeping in mind how information consumption should be primarily sourced from platforms that are backed with credibility. I find it comforting to know that there are sources of information that are considered reliable. Perhaps my perception of this is biased as well (because, frankly, who is not biased in this world?) but I do believe that news platforms that are credible help bring hope.