Chat GPT, a Large Language Model (LLM) developed by Open AI, has gained substantial attention for its potential impact on various aspects of human life after its official launch in November 2022. From aiding scientific research to helping college students generate essays for homework, the power of generative artificial-intelligence (AI) is gradually affecting the lives of humans.
Introduction:
Launched a year ago, ChatGPT has raised concern among educators due to its ability to generate seemingly well-researched essays in response to assignment questions. This development has prompted a global reassessment of traditional evaluation methods, leading to measures such as the prohibition of ChatGPT access on campuses, as witnessed in several universities, including The University of Auckland [1] and Imperial College London at earlier stages. Institutions are implementing changes such as a reduction in take-home assessments and the incorporation of AI detection tools, such as GPTZero, to scrutinise assignments.
Despite its generative powers, professors argue that the utilisation of ChatGPT constrains individual creativity and critical thinking by producing content derived from pre-existing online documents, composed of other’s beliefs. This raises concerns about academic integrity and prompts the question of the extent to which students should, if at all, rely on ChatGPT in their academic pursuits. Or alternatively, whether students could utilise this tool to their advantage while avoiding plagiarism.
Questions remain about whether ChatGPT can be made accurate and reliable enough to be deemed trustworthy as educational aids. In a mathematical research conducted by Cornell University[2], ChatGPT exhibited a relatively low accuracy rate, correctly answering only approximately half of the questions in a benchmark set comprising relatively simple mathematical problems typically addressed by students aged 12-17. This limited success rate raises concerns about its reliability, particularly in educational contexts. Additionally, the Large Language Model (LLM) demonstrated a propensity for inventing references or citations, generating potentially unreliable information that may be misleading or harmful.
Despite these risks, an outright avoidance of ChatGPT may not be the best approach. Instead, a more judicious strategy involves categorising its applications based on its strengths and weaknesses, thereby facilitating targeted improvements for educational purposes. An illustrative instance of this approach is evident at Arizona State University, where generative AI models like ChatGPT’s GPT-4 and Google’s Bard are deployed. Employing a technique known as retrieval-augmented generation, these models are directed to generate responses sourced exclusively from specific scientific papers or lecture notes. This method serves to reduce the likelihood of errors and allows for the incorporation of opinions that the AI may assimilate. Furthermore, exploring alternative variations of ChatGPT that integrate additional programs for calculations could contribute to enhancing the reliability of computational outputs. This nuanced approach acknowledges the potential value of ChatGPT in education while actively addressing and refining its limitations.
Hence, employing ChatGPT and other Large Language Models (LLMs) without a thorough consideration of their drawbacks is counterproductive and potentially unreliable. In the worst-case scenarios, such usage may impede students’ ability to learn and independently formulate their thoughts on a given topic. It is crucial for individuals to adopt a balanced perspective, acknowledging both the potential benefits and inherent risks associated with these technologies. Caution is necessary to seize the considerable opportunities presented by AI, and to ensure that the deployment of ChatGPT aligns with a commitment to improving, rather than exacerbating, educational outcomes and societal well-being. Achieving this balance is central to harnessing the transformative power of AI responsibly and effectively.