generative ai: generational implications

Special thanks to E.J.M.M for contributing to this article

Generative AI is a branch of artificial intelligence that can produce content across various mediums, such as text, images, video, and audio. The most popular forms of generative AI include “large language models” or LLMs such as ChatGPT and Gemini AI, as well as “text-to-image” models such as DALL-E 3 and Midjourney. These tools produce text and visuals that are nearly indistinguishable from those created by humans. Advances in multi-modality—a quality of AI models to interpret and generate content in multiple mediums rather than just one—have greatly expanded the capabilities and improved the ease-of-use of generative AI products. For example, ChatGPT is capable of processing and generating documents and images, alongside its typical text-based prompts. As a result, the popularity of AI tools has skyrocketed in recent years. 70% of Gen Z report using the technology, among which 52% cite using Generative AI to make informed decisions^[s]. With wide-scale adoption among the youth, it is becoming increasingly urgent for the education sector to develop strategies and policies for accommodating AI within the curriculum and to encourage its ethical and responsible use.

Before generative AI, plagiarism has been a prevalent problem in academic writing since time immemorial. With the advent of the World Wide Web, academic text from all disciplines and geographies have never been so accessible; what used to require going to a library and sifting through dusty collections of books and papers is now available anywhere to anyone with an internet connection. As such, the creation and implementation of protocols for the prevention, detection, and punishment of plagiarism became essential to every academic institution around the world. The consequences of plagiarism in most university policies are universally severe, ranging from a mere failing grade in very generous cases, to outright expulsion. Countless tools and software have been developed for the automatic detection of plagiarism, from exclusive in-house solutions to commercially available products such as Turnitin and Grammarly. The market for plagiarism detection software even before the pandemic and generative AI was by no means tiny—Turnitin’s parent company was acquired for $1.8 billion on March 6, 2019 ^[s]. Immense time, manpower, and funding was invested just to limit plagiarism.

Universities and academic institutions around the world are now facing a similar—if not bigger—problem in generative AI. What used to require the already limited effort of finding suitable material and paraphrasing is now available with one well-written prompt. Generative AI serves as an all-purpose tool for the expert manipulation of language; it is a search engine, writer, and paraphraser in one package. It can also play the role of a teacher, a consultant, and a research adviser^[s].

Unlike plagiarism, preventing the use of generative AI is much less straightforward. Plagiarism detection software is built around “text similarity”^[s], an established and well-researched problem in natural language processing with many approaches depending on the application. Plagiarism detectors benefit from having a limited “search space”—that is to say, they only require a diverse database of texts, which are finite in nature. Text that passes a plagiarism detector is either an original piece of work, or a well-paraphrased and edited version of an original piece of work, both of which are nonproblems. On the other hand, generative AI is fully capable of automating the process of producing original text, rendering text similarity obsolete. Thus, a different approach must be taken. “AI detectors” such as Quillbot (which was previously a popular paraphrasing tool) gained popularity early on due to their claims of being able to distinguish AI-generated text and human-written text. However, several controversies generated question marks around their authority. False positives were rampant—genuine student-written texts were being falsely flagged as AI-generated. When fed text from the Bible or from scientific papers, those were also flagged as AI-generated. False positives in AI detectors are significantly more harmful than those in plagiarism detectors, because they cannot be manually verified^[s]. A plagiarism detector’s false positive can be manually rechecked against the claimed source material and left up to the professor’s discretion. The same cannot be said for AI detectors, which are based on statistics and heuristics, and therefore cannot serve as valid evidence.

With detection being established as an exercise in futility, it is safe to say that generative AI is here to stay. However, the educational system cannot bury its head in the sand. Avoiding change and retaining the same outdated practices for teaching will simply lead to said practices easily being taken advantage of by LLMs. The use of LLMs is not necessarily malicious; they have a myriad of other applications, which include helping in all parts of the writing process—suggesting ideas, drafting outlines, paraphrasing, and proofreading, as well as providing personalized instant feedback, something that is difficult to achieve with large class sizes. They can also explain concepts with a thoroughness that human teachers might not necessarily have the time nor patience for. All of these are capabilities that would add value to the pedagogical process, and it would be remiss to neglect them. Instead, it will be more beneficial for teachers and professors to modify their classes with these tools in mind. The calculator may have reduced the value of mental math skills, but no one is arguing that calculators ruin math education. Math is more than just arithmetic—it is about understanding abstract concepts and combining these concepts in a creative manner to solve problems and derive new mathematical ideas. Writing can be likened to arithmetic, where basic proficiency can easily be taught without the use of calculators. Beyond that, creativity and originality should reign supreme, and no amount of artificial intelligence trained to simply mimic human tendencies can ever surpass that.

The introduction of generative AI marks a possible paradigm shift in the manner with which we conduct education. Students will now have access to an automated teacher that will perform any task that they want. If this fact is avoided, education will grow redundant as outdated curriculums will be taken advantage of by LLMs. However, if schools and universities embrace generative AI and build the education system around it, then the benefits of generative AI can be enjoyed and education can evolve and push the boundaries of what humans are intellectually capable of.