AI vs homework

January 19, 2024

Another installment in the series of AI doomer content is this video “AI can do your homework. Now what?” by Vox. It’s best to refer to these types of videos as mostly opinion pieces because they are edited and curated to spin a narrative. You are meant to walk away from the video with a specific take and they aren’t always so clearcut.

Whether or not the use of chatGPT enables students to cheat is not the biggest issue. Rather the video leave out the context of what subjects and classes are more prone to ChatGPT use by students. Since the video focuses on the general use of ChatGPT, the viewer doesn’t know.

But more than anything the video ignores the hypothetical question, is it ever in the students best interest to cheat? Absolutely. Not all classes taken for a degree requirement have the same level of relevance in the long run. Is the short term benefit of passing the class worth the long term consequence of not having learned anything? That’s for the student to decide.

The positive argument for cheating and plagiarism is meant in a satirical sense. The reality is no one thinks ChatGPT replaces the act of serious research. Instead of banning the use of AI chatbots, schools need to encourage the use of developing AI chatbots to understand their shortcomings. A common example of this is how much LLMs rely on Wikipedia and web crawling websites (which may pull information from Wikipedia. Even though Wikipedia has improved, it still has limitations. A video by Captain KRB, “The Cryptids of Wikipedia” discusses some example of malicious Wikipedia editors who operate by making minor errors as a form of harassment. It’s also possible for minor errors to be introduced unintentionally. These errors can propagate by the number of websites that copy Wikipedia data, which in turn make their way into LLM datasets.

Students need to be judged on the veracity of their ideas. If the assignment can easily be gamed by using ChatGPT, then it probably isn’t challenging enough. Students need to see what ChatGPT creates to understand its inherent shortcomings. It’s no AGI (yet).