Should We Replace TAs with AI?
In the previous section, I explained whether or not you should learn computer programming in the AI era.
The question of whether AI will replace human jobs is largely rhetorical at this point, with the answer being a resounding “Yes.” This replacement is an inevitable progression, as history has shown through previous industrial revolutions. Identifying which types of jobs will be displaced first is also not so significant; any jobs susceptible to digitalization for AI models to learn are at risk, including ones associated with intelligence, like medical diagnosis and higher education, as well as those regarded as creative, such as music production and book writing. The important question you should be asking is, “What will be the consequence of replacing humans with AI?”
Every year, I work with multiple teaching assistants (TAs). I cannot imagine teaching any class with more than 24 students without their invaluable support. Despite the excellence of these TAs, recruiting and training them to a high level can be difficult. Moreover, when there are several TAs for a course, ensuring uniform performance among them becomes challenging; as a result, some TAs end up bearing a greater workload, as students demand their assistance more frequently than the others. This imbalance is even more pronounced across different academic years, where the performance of TAs can vary considerably, directly impacting my teaching evaluations. This raises the question: can we replace TAs with AI and still expect the same level of effectiveness?
Foundational courses with stable content, like Introduction to Computer Programming, can benefit greatly from AI models, which can proficiently address most student queries and provide detailed evaluations of their programming assignments, helping students understand any missed concepts or programming details. In a sense, and with all due respect to my wonderful TAs, these AI models often offer clearer and more precise explanations than many of the TAs I have experienced such that they are perfectly capable of serving in the TA roles. What about elective courses teaching specialized topics, like Introduction to Natural Language Processing? Even for these courses, the content often remains stable, focusing on fundamental concepts rather than the very latest advancements, in which cases, AI can still effectively take the roles.
When it comes to advanced courses revolving around cutting-edge subject matter, the situation becomes more complex. These courses often demand a level of expertise that is challenging to find in human TAs and typically have smaller enrollments due to their specialized nature, often negating the need for TAs altogether. Assuming AI will continue to evolve at an accelerated pace (which is highly likely), it may be more practical to train AI models to act as TAs even for these advanced courses rather than continually seeking out human TAs with such expertise. By opting for AI over humans, you eliminate the need for TA recruitment and save a substantial amount of money. With all these advantages, it might seem imprudent not to replace human TAs with AI. Nonetheless, the long-term implications of entirely replacing TAs with AI warrant consideration.
We hire TAs not only for their immediate roles in courses but also to help develop their teaching skills. This is why most Ph.D. students are required to fulfill TA duties as part of their training; after obtaining Ph.D. degrees, they should possess the ability to teach proficiently. Those without prior teaching experience may struggle in a classroom setting, which differs significantly from presenting research work at seminars or conferences. This early lack of teaching experience can impact their performance as professors, particularly in the initial years when they already have a multitude of other unfamiliar responsibilities to manage; as a result, it can erode their confidence in teaching.
What do people do when they lack confidence in something? They tend to avoid doing that thing. As this hesitancy reaches a critical mass, it may lead to the exploration of alternative teaching methods, eventually giving rise to AI instructors teaching entire courses, rather than merely serving as TAs. Given this momentum, it is conceivable that some universities will adopt AI instructors to deliver courses, particularly in certain majors (Computer Science being a likely early candidate). If this approach proves effective, ensuring students learn the course materials well and embark on successful career paths, more majors and educational organizations will follow suit. What is the issue with this shift? If students excel in their learning and achieve successful careers through AI instruction, shouldn’t we be content with the outcome?
The concern lies in the increasing loss of the human element in our society and the risk of conditioning our students to function more like machines. Teaching is not merely the transmission of knowledge and ideas; it extends far beyond that. As an educator, you share your experiences with your students. You understand the aspects of course materials that can be challenging and offer valuable insights on how to overcome those hurdles when you notice students grappling with similar issues. You remember the benefits of forming study groups to tackle difficult tasks, allowing you to recommend such groups to your students. AI may excel in training students to become highly proficient in their disciplines, but it cannot cultivate the profound teacher-student relationships that humans can.
Here is another angle to look at. Students exclusively educated by AI are unlikely to possess the capacity to instruct other individuals in those subjects, primarily because they lack real-life examples of teaching. Their best course of action would likely be to recommend the AI systems they have used to others instead of attempting to teach fellow students; consequently, we may face a decline in the number of qualified human teachers. At this juncture, employing AI for teaching will no longer be an option but the only option we may have.
Even after considering these arguments, it is highly probable that TAs, particularly in Computer Science, will be supplanted by AI in the near future. However, it is crucial that we anticipate the repercussions of this shift and establish new avenues for our students to learn the art of teaching. Failure to do so could diminish a fundamental aspect of our society, for teaching has been a cherished human legacy since the beginning. Replacing humans with AI before nurturing these alternative opportunities for any profession could result in an irreversible loss of essential human qualities. Therefore, we must cultivate the human capacity for teaching while embracing AI as a tool for education ensuring a harmonious coexistence of the technological and human aspects of education.
