Rose Bialk/The Mossy Log
In the present day, discussion around how artificial intelligence is integrated into student life at Lewis & Clark is unavoidable. Some faculty are firmly against the use of AI tools in their classrooms, while others encourage students to use them as a part of the curriculum. It is my belief that, before policies restricting and punishing using this technology are made, policies that dispel the misinformation surrounding its abilities must come first.
This misinformation surrounding the abilities of AI is spread by AI companies themselves. Beyond their products, they want to sell the idea that AI tools are hyper intelligent and border on conscious. This is why the user interface for ChatGPT is akin to a texting app — to make the user feel that they are conversing with a conscious machine. It is also why OpenAI, the company that owns ChatGPT, uses verbs associated with critical thinking in humans when describing ChatGPT. It claims the software can “answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.”
However, this idea that they try to sell does not exist in reality. Programs such as ChatGPT are large language models (LLMs), systems which predict what word is most likely to come after another to generate answers to prompts. These LLMs are trained on terabytes of text, scraped from the internet and databases like Project Gutenberg, a library of over 75,000 e-books. The model analyzes all of this data, searching for patterns in its language, so that it can generate new sentences based on those patterns. If an LLM is prompted to generate a sentence telling a friend you will be late to an event, it will search for patterns in its training data that correspond to the prompt. It will note that the word “late” is often near words such as “sorry,” or “apologies.” With this pattern established in the model, it generates the phrase “I am sorry, but I will be late to this event.”
Therefore, it is essentially a scaled-up version of the “suggested words” above your keyboard you see while you text. This is not to say that the technology is not impressive or complex — its ability to find all of these patterns in language is incredible — but LLMs cannot actually “think” in the same way a human can. These models are just outputting one likely word after another and are thus incapable of outputting coherent, quality work for an academic setting.
It is worth acknowledging that some students are more aware of LLM’s limitations and use them regardless, because it’s an easy way around doing academic work. However, I do not believe that takes away from my belief that policies should focus on dispelling the misinformation spread by AI companies first. I think jumping to enforcing punitive or restrictive policies does not give students a fair chance to educate themselves and make an informed decision on the tools they use. Some students are not aware that the programs they rely on are not capable of the thought necessary to help them in their work. I believe it to be unfair to punish them for this ignorance, because it is fueled by massive marketing campaigns funded by multi billion dollar corporations that want AI to be in every single part of our lives. They must be informed and given a choice.
If access to these tools is removed with little explanation, the result will be an endless cat and mouse chase, where unaware students find ways to mask their LLM usage and faculty spend their time trying to catch them in the act. This exact situation played out in my high school, where LLMs were prohibited for certain assignments or classes. Students were constantly discussing the best way to hide their use of AI. It was so incorporated into their academic life that they could not conceive of how they could complete work without it. Instead of pushing students to avoid LLMs, these policies pushed them to waste their time figuring out ways to hide their use of the tech.
The failure of these policies was that they did not explain to students the flaws of LLMs, and to the students who did not know better, the policies took away their super-intelligent machines that helped them. A lot of them were not aware that these machines did not “understand” a single word that they were spitting out. When students are aware of the misleading ideas they are sold, they are given the choice to no longer buy into them.
In my Intro to Rhetoric and Media Studies class, we were given the task of writing our class’ policy on AI, specifically LLMs. Prior to the task, we had dedicated time to learning about the inner workings of the technology, and the limitations of it, especially in an academic setting. I do not believe it to be a coincidence that after being educated on the problems with LLMs, the class collectively wrote an AI policy that heavily restricted the technology’s use. Students will only stop their AI use if they are properly informed of what they are using and are not operating on the false ideas that AI companies are selling.
When I see how students view LLMs, I am reminded of the Mechanical Turk, an 18th-century machine which could seemingly play chess, beating master players in the game. It appeared to be a marvel of engineering, a machine with a conscious. In reality, the automaton was controlled by a talented human chess player under the table. The Turk was still an engineering feat, but at the end of the day, it was a parlor trick that made money off of audiences who did not know much better.
This is exactly what is happening with LLMs in the 21st century, and I believe that it is the duty of LC to ensure that its students are aware of the smoke and mirrors behind the parlor trick. If they do not, any policy which goes against it will surely result in undeterred use.
About the Author
