(*Title of this article is generated by ChatGPT)
Artificial intelligence, or AI, in short, has made the headlines multiple times. From the chess engine Deep Blue to AlphaGo, the self-taught AI for the game of Go, AI has shown its potential to excel in certain fields and to outperform its human counterparts. However, until recently, no AI had been able to write essays and answer questions based on research, as well as to compose works of fiction and non-fiction based on the user’s command. This spiked excitement, as well as a fear that this new breakthrough may result in threats to academic integrity and existing rules around intellectual property.
Chat GPT (Chat Generative Pre-Trained Transformer) entered the stage of AI about six months ago now, on November 30th of 2022, with the stable release on January 30th of 2023. It is a product of the company Open AI, aimed to answer people’s questions and perform tasks. Officially considered a language model, Chat GPT uses existing information on the internet to “learn” how humans speak, as well as research factual information for answering questions. Not only can it model standard modern English, but it can also mimic certain writers’ styles such as Shakespeare’s. Within the first five days of its initial release, Chat GPT gained around a million users, a popularity level that took Facebook ten months to achieve. The algorithm “learns” by initially having human AI trainers teach the essentials of composing responses, followed by having a large amount of data being fed to the AI for it to learn the formats of English. The AI trainers would then ask questions and rank the answers given by Chat GPT, helping the AI understand the preferred way of answering questions.
The result of this repeated training is an AI that is capable of writing entire essays without copying directly from the internet, while also using adequate research to back up its writing. Chat GPT’s ability to write soon captured the attention of teachers at a multitude of institutions, from high schools to universities. The biggest concern around Chat GPT’s implication in academia is the concern around plagiarism. Due to Chat GPT being an AI that generates instead of copies, the works generated by the language model would pass regular plagiarism tests like any other original work. The works created by Chat GPT have reached a level of authenticity that a new way of checking for academic integrity is needed. An additional problem lies in that despite Chat GPT’s answers still having certain flaws, there is no definitive evidence to show if a student used the AI. In other words, it is extremely hard to prove that an essay is made by Chat GPT, all that the teachers can do is suspect. Furthermore, AI can also answer many questions with very high accuracy. Famous Youtuber, AP History teacher, and grader Steve Heimler used Chat GPT to answer three SAQs from AP World History, AP US History, and AP European History, with the AI earning full marks on all three questions. In response to such concerns, Edward Tian, an undergraduate at Princeton developed a program for checking the likelihood of an essay being the AI’s work. His program analyzes the sentence length and word choices, as well as many other variables to make its judgments. Despite these concerns about academic integrity, Chat GPT can be beneficial for research efficiency. For example, Chat GPT can facilitate academic research by summarizing and analyzing large amounts of given academic essays without needing the researcher to read each individual essay. It can also help proofread written work for not only grammatical inaccuracies but also potentially better word choices. Applications on these aspects.
In addition to its impact on academia, Chat GPT also creates ripples in the discussion of moral standards around. Despite Open AI setting limitations on sensitive information, Chat GPT can still produce controversial pieces that may offend or be untrue. The information feed used to train the AI is also not politically neutral, with many considering it left-leaning and even double standard. Open AI never programmed the AI for fixed responses, therefore it is impossible for the company to know exactly how the AI would respond to a given prompt. However, the AI cannot be considered responsible for its own responses either, since it does not “own” the things it writes.
There are still limitations to Chat GPT: the AI may not always be specific in its answers, it can “hallucinate” and start speaking absolute nonsense, and it can give wrong answers to questions. However, the development of this AI is another step toward an unknown future. Maybe one day AI will write better essays than us, just like how Stockfish can crush even Magnus Carlson with ease.
