Generative AI and Education
The winds of change are swirling
This is the third post in a series exploring the societal impacts of advances in AI, with an eye towards the future. In this post, we'll discuss the potential impact of AI on the education system, both in terms of its effect on the classroom and student experience, as well as how higher education should adapt to prepare students for future careers.
Generative AI’s Biggest Fans: Students
There are many debates about AI in education, but one thing is clear - students are actively using AI today.
According to an ACT research study conducted in 2023, 46% of American students in grade 10-12 reported using AI tools for school assignments. Interestingly, this study showed a high correlation between AI tool usage and academic achievement. Students that scored higher on their ACT were more likely to use AI tools than those with lower scores.
Another study from Pearson found that 51% of students used generative AI tools in the Spring 2024 semester. A separate study from Tyton Partners sponsored by TurnItIn showed that half of college students are regular users of generative AI and believe it has had a positive impact on their learning, and 75% of users would continue using it even if it was banned by their institution.
It's no surprise why this is the case. Schoolwork is often extremely pattern-based, involves reading large amounts of dense content, requires a clearly defined and structured output, and involves recall of specific facts and concepts. The Tyton Partners survey referenced summarizing text, answering homework questions, understanding difficult concepts, and assisting with writing assignments as some of the most frequent use cases, among many others.
Clearly, students need to be careful given generative AI's penchant for hallucination, especially around fact-based answers. But it's evident that generative AI can be used by students to complete assignments faster, better understand concepts, and generally be more efficient.
Generative AI’s Biggest Critics: Administrators
I acknowledge the header here is hyperbolic, but it's meant to highlight the dichotomy between the two sides of the education system. Despite the widespread student usage described above, there is significant teacher and administrative resistance to using generative AI in schools.
The same Tyton Partners survey showed essentially inverse opinion and adoption between teachers and students. Fifty percent of faculty thought that generative AI would have a negative impact on student learning, compared to just 22% of students, and only 22% of faculty were using generative AI, compared to 50% for students.
Despite faculty skepticism around generative AI's impact on learning, the Tyton Partners survey showed that faculty thought it was more important relative to students to know generative AI tooling for use in professional settings. 75% of professors that used generative AI and 47% of professors who did not thought generative AI writing tools would be needed for work, compared to 55% and 33% for the same groups of students.
Generative AI identification tools like TurnItIn, GPTZero, CopyLeaks and Originality AI have emerged to determine whether student submissions are coming from AI. The best information I could find on outright banning of generative AI showed that 27% of districts have completely banned generative AI for students, so it does not appear widespread, but certainly is happening.
Of course there are many teachers and administrators that are leveraging generative AI to make their jobs easier, so painting this picture as students vs. teachers is unfair. But generative AI undoubtedly makes administrator and faculty jobs more complicated if the status quo of education delivery is maintained.
How Do We Adapt?
Generative AI in education is complicated because both the pro and anti sides have extremely valid arguments. No one wants a system where students are just submitting AI-generated content and not learning any of the material. At the same time, restricting student access to generative AI may hinder their learning process and not prepare them for the working world.
This points us to the real question: Is the education system set up for a world where generative AI is prevalent? If generative AI can easily solve assignments and software-based detectors are necessary to catch them, are we really giving students the right assignments?
Thus far I’ve primarily treated the education system as a single entity. In practice, policies and approaches around generative AI should vary based on the level of schooling.
As an example, I am an adjunct professor for an Applied Artificial Intelligence Masters program. I encourage my students to leverage AI tools as much as possible to be more efficient and to enable better learning of the material. I have attempted to change the curriculum to adapt for generative AI, focusing about half the class on recent developments.
I plan to integrate some material that focuses on problems where humans still outperform AI specifically in the context of being a data scientist or machine learning engineer. I also plan to design an assignment around determining what specific tasks are conducive to using AI tools, and what tasks are better done manually.
All of my students are working professionals looking to advance their careers specifically in the AI field, so it would be a disservice to dissuade them from using generative AI tools. Additionally, since my students are working professionals and more practically focused, I think it is reasonable to treat my students as more autonomous drivers of their learning journey.
For younger students earlier in their academic journey, there are more consequences if they misuse generative AI to simply pass required assignments and not learn any of the underlying material. At the same time, if generative AI ends up being an important part of the workplace, it’s important that students learn in a way that makes them competitive professionally.
We should encourage creative ways to pivot homework and assignments that enables younger students to leverage generative AI to better understand concepts and skip out on busywork altogether. One opportunity is to make classrooms more focused on relating to personal experiences, and interpersonal dialogue and connection, rather than rote memorization.
Clearly, the way we teach and evaluate students will need to change at all levels of the education system. Students are using generative AI extensively, and school policy varies tremendously on its degree of acceptability. We should be focused on preparing students to the best of our abilities for their professional future, which likely means a balance between leaning into AI tools and preventing overuse.
This all becomes an especially acute issue when focusing on the higher education system, given the increasingly exorbitant costs and risk of knowledge work automation.
Higher Education Challenges Ahead
It’s not news that the cost of higher education have gotten so out of control that its causing people to question whether getting a college degree is really worth it. Companies seem to be steering away from using a four-year degree as a prerequisite for jobs. There have been recent pushes to cancel student debt as a pretty transparent acknowledgement of a broken system.
Personally, I've never understood why a sales job (or an equivalent) would require a four-year degree given the job responsibilities are extremely disconnected from the educational content of the degree. It makes the degree requirement almost a social-proofing exercise to show that a person can clear that societal bar, but that seems more and more misplaced when costs are so extreme.
Generative AI throws fuel on this fire. As discussed in the prior post, generative AI is commoditizing knowledge that was previously reserved in the higher education system, and should put downward pressure on the value of the knowledge obtained in a higher education setting. This may have a negative impact on both the volume of and salary associated with knowledge-based jobs.
Large language models passing the Bar Exam and USMLE doesn't make those models lawyers and doctors, but it must give us some pause as to how valuable those tests really are. If we are capable of automating those answers, does that suggest that we can make the process of becoming a lawyer and doctor more efficient? Can we shrink down the time and cost associated with schooling? Are there other ways of evaluating future lawyers and doctors outside of those tests that involve incorporating their own knowledge and available automated knowledge together?
If we do nothing, the combination of knowledge commoditization, the scale of people graduating college, and the cost associated with obtaining a degree paints a bleak picture. There are only so many high-paying jobs that justify the return on investment from a degree at current costs, so it will inevitably leave people in debt that is extremely hard to walk away from.
As stated previously this is not a new problem, as the promised gold-plated ROI from going to college is starting to show some cracks. If generative AI advances further squeeze graduates with high debt loads and limited prospects, where does that leave the higher education system?
Can we find alternatives to the four year bachelor’s degree as the primary prerequisite for positions that do not heavily leverage the content of the degree?
Conclusion
I don't mean for this post to disparage the purpose of education or the quest for more knowledge. As mentioned, I serve as an Adjunct Professor myself! I do think it's critical to objectively evaluate whether our education system is preparing society and graduates for the future.
Right now there is minimal pressure on higher education institutions to innovate to find solutions, because they are not feeling the pain of the consequences. School endowments have reached staggering highs, and individuals own all the risk in converting the knowledge they gain in school to downstream monetary gain. Some alternative approaches have emerged, like bootcamps, some even with income sharing agreements.
In a perfect world, academia would be confined to the pursuit of knowledge, and we wouldn’t need to talk about higher education with optimization and practicality in mind. Right now, reality dictates that higher education is primarily a ROI-driven decision for individuals.
These issues have been simmering for a while, but generative AI threatens to accentuate them further. Deliberately restructuring our education system to adapt to AI to better prepare students for future careers will be a steep but critical challenge in the coming years. It also might force us to more earnestly redistribute risk between students and institutions to hold all parties accountable for student outcomes.
To move forward, we need to focus on key areas:
Integrating AI literacy into curricula at all levels of education
Redesigning assessments to focus on skills that AI can't easily replicate, or how AI and manual work can be complement each other
Emphasizing critical thinking, creativity, and interpersonal skills
Exploring new models of higher education that are more flexible and cost-effective for students, and making institutions more responsible for student outcomes
Be prepared for more change, and pivot to the needs of society rather than get stuck in the status quo
It's clear that the intersection of AI and education will be a crucial area of focus for society as a whole. Our goal should be to harness the power of AI to enhance learning while ensuring that our educational systems continue to produce graduates that are prepared for the jobs of the future in an economically sensible fashion.