Tutor, TA, Talent Scout: The Coming GenAI Revolution in Higher Education
GenAI can make artificially intelligent assistants available to every student, teacher, and employer — if we implement it right.
“I’m optimistic about universities broadly…There was a great book by Clark Kerr called “The Uses of the University,” and … in that book, and he said that when you looked 500 years ago at Western Europe there were 66 institutions that were in roughly the same place that they had been 500 years before, roughly doing the same thing that they were doing 500 years before. So there were 2 religions that were not so changed over those 500 years, there were 2 governments that had not been so changed over those 500 years, and there were 62 universities that were in the same place doing what they had been doing for that half a millennium.” — Michael Drake, President of Ohio State University, March 2020.
Introduction — A Perfect Storm of Disruption in Higher Ed
Higher education is currently at a convergence of disruptive forces, from the rise of MOOCs (e.g., Outlier) and increasing tuition costs, to the shift to online learning during COVID-19. Although higher education is “on the edge of the crevasse,” as Clay Christensen predicted in 2013, these universities are prospering so much financially that they do not see the impending threat in the data (Smith 2023). Now on top of these swirling headwinds, generative AI (GenAI) tools like ChatGPT have entered the fray. The dominant narrative in the media has been that Generative AI tools like ChatGPT will undermine higher education, because it makes it so easy for students to cheat on their assignments and exams. However, if implemented correctly with the right safety measures in place, we can mitigate the downsides of GenAI while benefiting greatly from the greatest positive transformation of education in human history.
In this paper, I will argue for three Generative AI (GenAI) use cases in higher education:
- an AI tutor for students,
- an AI teaching assistant (TA) for educators, and
- an AI hiring assistant for recruiters.
The primary benefits of these use cases will be enhancing outcomes for three A’s: individual apprehension, assessment, and access. To address the problems of enforcement, thought diversity, and bias, we propose (1) outcompeting plagiarism problems with a better learning experience, (2) intelligent assignment design, and (3) ensuring equal access to GenAI tools for all.
Use cases for ChatGPT as an educational tool
AI tutors for students
In a 1984 study, educational psychologist Benjamin Bloom observed what he called the “two sigma problem” in education: students who received one-on-one tutoring using mastery learning techniques performed two standard deviations (hence, “two sigma”) better than students who received traditional classroom instruction, outperforming 98% of the traditional classroom students (Bloom, 1984). The “problem” is how to economically scale 1–1 tutoring across millions of students, which is the holy grail pursued by edtech companies like Duolingo. AI tutors can help address this problem by providing personalized instruction that adapts to each student’s pace and needs, thus reducing the “teaching to the mean” approach necessitated by traditional classrooms. AI tutors can benefit students who find the traditional pace either too fast or too slow, augmenting learning by offering tailored feedback (VanLehn 2011).
As such, an AI tutor can be a great complement rather than substitute for education. For example, Khan Academy’s “Khanmigo” AI tutor has been trained by human education experts to be more “socratic” and better at math than ChatGPT (Khan 2023). Its benefits can be summarized by 3 C’s: context, conversational characters, and coaching. In terms of context, instead of simply telling students the answer to a question, Khanmigo will ask them where they are getting stuck. Not only can it detect mistakes, it will ask students to explain their reasoning, and even predict the misconception that is causing the mistake. It can ask students why they care about a topic, and motivate them based on their response (e.g., how learning biology can make you a better athlete). It is already more effective than human tutors in teaching coding, which can help solve the problem of the computer science teacher shortage (Khan 2023). As for conversational characters, students can talk to literary characters like Jay Gatsby to improve their reading comprehension, bringing the material to life far better than reading dry academic papers. Lastly, Khanmigo is great at coaching students. For example, Khanmigo is a great writing coach that can write with the student instead of for them. It can help students draw an outline, construct a draft, and then provide personalized feedback. In one writing exercise, the student writes two sentences of a story, the AI writes the next two sentences, and then they alternate. In another exercise, Khanmigo can debate with a student to help her refine their arguments. The key is to fine-tune the AI to education. All these benefits are possible because on the back-end, Khanmigo put in the hard work of prompt engineering and training the AI to verbalize its thoughts before it speaks, which made it a better tutor in subjects like math, a concept Khan calls “AI thoughts.” Incorporating AI memory and voice can further improve this experience, especially for subjects like language learning that can use voice recognition.
AI TAs for teachers
Teachers often face challenges in addressing the needs of each student due to time constraints and increasing class sizes. AI TAs can alleviate some of this burden with at least three use cases: providing teacher guides, grading, and individualizing learning plans. This will help teachers focus on higher-impact activities (Graesser et al., 2001). GenAI apps like Khanmigo can already instantly generate tailored lesson plans, complete with hooks, outlines, and learning goals. Khanmigo can also provide not only answers to student questions, but can also give different ways to explain the material. As for grading, ChatGPT can already assess essays based on substance and style, and provide targeted feedback on how to improve. Lastly, GenAI can easily generate individualized student goals and progress reports to help teachers tailor teaching to each student.
AI hiring assistant for recruiters
The increasing prevalence of online education has largely solved the scarcities in both classroom seats for receiving instruction and faculty seats for providing instruction. However, what online education has not solved is the credentialing problem, as employers often struggle to assess the qualifications and skills of candidates with non-traditional educational backgrounds. As such, recruiters typically refer to imperfect heuristic signals of aptitude, such as university degrees and internal referrals. To address this issue, a generative AI assistant can be used to assess candidates based on skills and personality traits. AI-based assessments can be customized to suit specific job requirements and can use various testing modalities, such as coding challenges, problem-solving tasks, and situational judgment tests. These assessments can be administered asynchronously, allowing candidates to complete them at their convenience, thus ensuring a more inclusive and accessible recruitment process. Benefits of AI-driven assessments include improvements in: unbiased evaluations, equitable pipelines for sourcing talent, time and cost efficiency, candidate experience, job-fit, and retention. Past AI efforts have failed mainly due to biases in the training data (facial expressions, language, tone of voice), and proxies for protected characteristics such as race and gender (e.g., Hirevue, Amazon’s AI resume screener, etc.). However, GenAI could use a different assessment mechanism — it can evaluate candidates’ communication skills, for instance, based on a more diverse training dataset (Web Text) compared to other companies’ biased training datasets that underrepresent minorities. GenAI can also explain its evaluations to human recruiters, resulting in more explainability and transparency compared to other AI algorithms. Although GenAI may reduce the power of intermediaries such as universities, schools like CMU should get ahead of this threat by adopting GenAI tutors that better prepare students for employees who evaluate them based on skills rather than credentials.
Implementation Risks and Mitigation Strategies
In the long term, universities should form an industry consortium task force to monitor the impact of GenAI tools on student and teacher outcomes so they can adapt to their evolving capabilities. They can track metrics such as student test scores, GPAs, and engagement, as well as teacher workload and satisfaction. In the short term, a phased approach is recommended to address the following issues arising from GenAI.
Enforcement problem
To tackle potential misuse of AI tools like ChatGPT, universities must use a carrot-and-stick approach. On the stick side, schools must regulate their use with updated academic integrity policies at both the school level and the class level, requiring professors to set clear guidelines about bounds of use. For example, professors may prohibit the use of AI tools to prevent plagiarism for specific assignments (e.g., computer science modules, labs, or tests). However, like digital piracy, it cannot be eradicated by regulations alone. We have already seen the beginning of an arms race between AI writing tools and plagiarism detection softwares like Zero GPT, leading to an endless cycle of “whack-a-mole.” Therefore, on the carrot side, we must “outcompete” plagiarism by making the educational experience so engaging that AI-driven plagiarism is unattractive. We can do so with GenAI tutors that guide users through modules and remove friction points, much in the same way that Netflix and Spotify saved the film and music industries by providing a more convenient user experience than piracy (e.g., torrenting).
Diversity of thought problem
Some scholars theorize that GenAI can reduce creativity because most writers and creators will use GenAI as a “first draft” that they will then refine. This reduces the overall originality of works produced. As such, educators must design assignments with prompts that require critical thinking and creativity beyond a simple prompt they can input into ChatGPT. For instance, ChatGPT did not give very effective answers for my capstone project on use cases for generative AI for video game platforms, which required more primary and secondary research to synthesize insights. For these assignments, GenAI will simply be another tool in the toolbox that complements human intelligence, freeing up time for students to tackle more advanced challenges.
Biased access problem
Ensuring equal access to AI tools is crucial to avoid exacerbating existing inequalities in education. Institutions must address the “digital divide” by providing all students with access to AI resources and support them in using these tools responsibly.
Conclusion — GenAI for EdTech is a Matter of When, Not If
By leveraging AI tutors, TAs, and hiring assistants, GenAI can enhance outcomes for individual apprehension, assessment, and access. Adoption is not a matter of if, but how. The successful integration of GenAI in higher education requires a thoughtful approach that mitigates risks, including enforcement, diversity of thought, and biased access. By doing so, higher education can harness the power of GenAI to create a more personalized, inclusive, and effective learning environment, ultimately driving the greatest educational revolution in history. For the first time, if we get GenAI right, every child on earth can finally receive a world-class education.
References
Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6), 4–16.
Bostrom, N., & Yudkowsky, E. (2014). The ethics of artificial intelligence. In K. Frankish & W. M. Ramsey (Eds.), The Cambridge handbook of artificial intelligence (pp. 316–334). Cambridge University Press.
Graesser, A. C., Chipman, P., Leeming, F., & Biedenbach, S. (2009). Deep learning and emotion in serious games. In Serious games: Mechanisms and effects (pp. 81–100). Routledge.
Kerr, C. (2001). The uses of the university. Harvard University Press. (Original work pub. 1963)
Khan, Sal. “Sal Khan: The Amazing AI Super Tutor for Students and Teachers.” Sal Khan: The Amazing AI Super Tutor for Students and Teachers | TED Talk, 1 May 2023, https://www.ted.com/talks/sal_khan_the_amazing_ai_super_tutor_for_students_and_teachers/c/transcript?language=en.
Lederman, D., et. al. (2020, October). To Chat, or Bot to Chat, Just the First Question: Potential legal and ethical issues arising from a chatbot case study.
Luan, H. (2021). A review of using machine learning approaches for precision education. Journal of Educational Technology & Society, 24 (1) (2021).
OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat
Pappano, L. (2012). The Year of the MOOC. The New York Times, 2(12), 2012.
Shneiderman, B. (2020). Human-centered AI: Reliable, safe & trustworthy. International Journal of Human-Computer Interaction, 36(6), 495–504.
Smith, Michael D. (2023). The Abundant University. Unpublished manuscript.
Lederman, Doug. (2020). The best way to stop cheating in online courses? Teach better. Inside Higher Ed. https://www.insidehighered.com/digital-learning/article/2020/07/22/technology-best-way-stop-online-cheating-no-experts-say-better
VanLehn, K. (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist, 46(4), 197–221.
Follow our Social Accounts- Facebook/Instagram/Linkedin/Twitter
Join AImonks Youtube Channel to get interesting videos.
