Can Generative AI Make Education Less Diverse?

Generative AI is creating waves. I am hard-pressed to be in a meeting where AI doesn’t come up. There is hype, but there is no smoke without fire. GenAI has generated a reaction in our collective consciousness that very few tech innovations have achieved in the past 100 years. Good tech is like magic, and GenAI is a textbook example by that yardstick. If you missed it, ChatGPT-4 can pass the freshman class at Harvard .

I have little doubt that the potential of AI is immense. But there is a BUT. A significant BUT concerning the use of AI in education, something near and dear to me. No, I’m not all doom and gloom about students cheating on papers or exams, though some of that will happen. My concern is deeper. It’s solvable, but we need to be vigilant as it is more nuanced than the typical doom and gloom narrative surrounding AI.

I want to caveat that many of my observations are based on personal experiences with AI and some emerging research studies. Nothing is set in stone, and we’ll continue to observe and adapt to ensure that the practices proving beneficial for our students are scaled up.

My Learning from ChatGPT

I am a big believer in “learning by doing.” When Facebook’s Oculus came out, rather than engaging in theoretical discussions, our VP of Engineering, Brian Hurlow, suggested we start using Oculus in our team meetings. We bought it for the tech team, and within a month, we had our answers. It’s good tech but far from the maturity needed to go mainstream. I’ve promised myself to write a blog about VR in general, especially in education.

Returning to AI, we have been using Grammarly in our company since 2020 as an AI-driven English writing aid. If you have heard of or used Grammarly yourself, you probably know that it was one of the early implementations of AI as a language aid that received mass appeal. ChatGPT took it to a whole new level. I started using ChatGPT from January 2023 for various daily tasks—email editing, task lists, or even conducting research on topics in minutes that used to take hours. I could even request ChatGPT to make my writing sound a certain way with simple prompt engineering. It made me sound more professional, considerate, and clear. I was easily saving a few hours a day using AI, and I still do.

But things started to change. All my emails began to sound the same, and I worried if they would elicit the right responses from the recipients. I was concerned they became less readable and engaging.

Looking Ahead

Looking ahead, it wasn’t hard to imagine a future where most people use ChatGPT/GenAI. Wouldn’t my entire inbox start sounding the same? Would I even be excited to open and read emails? I might perhaps have a chatbot to respond to my emails. This idea started making me jaded, as I could extrapolate all sorts of red flags when it comes to education.

In our last company conference, I said, “Education is not just the transfer of knowledge but the transformation of the learner.” This is why we are building Yellowdig. But when the entire world uses GenAI to create and transfer knowledge, can it still engage learners effectively?

ChatGPT vs. Human Generated Content

I recently came across a study conducted by the University of Illinois, in which Yellowdig was involved. The research compared online learning discussion forums where students were divided into two groups—one using Yellowdig for organic, human-to-human conversations on course topics and another using ChatGPT to generate ideas. The research found that although students wrote longer posts with the help of AI, they received less engagement. This supported some of the issues I was personally encountering. Yes, it doesn’t hurt that Yellowdig was beating ChatGPT in engagement!

There are many interesting observations in the research, but what stood out to me is the indication that the diversity of conversation was lower when AI played a more significant role. The author of the study Unnati Narang stated, “Our analysis of the text of the posts shows that even though AI content tends to be more verbose, it lacks diversity, ranks lower in readability, and is less personal.”

Diversity and Education

Why is diversity important in education? I am specifically referring to diversity in thoughts and experiences. Arthur Schopenhauer wrote, “In consequence of the nature of our intellect, concepts should arise through abstraction from intuitive perceptions; hence, the latter should exist before the former.” We learn concepts through experience, trial and error, and exposure to diverse points of view, ideas, beliefs, and experimental results to develop our perception of the world around us.

Path Forward

What are the concrete steps we can take concerning AI? Here are a few of my suggestions:

Firstly, use tech solutions firsthand to fully understand their intended and unintended consequences. Rapidly evolving tech requires an experimentation, piloting, and impact measurement approach. This advice isn’t limited to AI but applies to all technology adoption. Many central initiatives fail when decision-makers don’t use the tools daily and rely on second-hand information.

Secondly, ensure an iterative mindset. We’ll go through a rapid innovation cycle. Build applications with a plug-and-play architecture so each piece can be upgraded independently. Many leading tech companies work this way, leveraging best-of-breed products and often switching between providers. We must adopt this mindset in learning design.

Finally, instructional design, or more aptly termed, instructional engineering embedded in learning science, will be increasingly important. Institutions starting to build capability in this area will leverage new tech effectively and ensure we don’t fall into the trap of using tech that doesn’t create the right learning impact.