Opinion: Note to B.C. schools: You don’t have to use generative AI

As a post-secondary educator, I can report that the “importance” of Gen AI is greatly exaggerated. What students can get in a classroom can be real and human, and doesn't have to be augmented by software whose output has only a tangential relationship to the aspirations of true education, writes Joel Heng Hartse. Photo by AntonioGuillem /Getty Images/iStockphoto

Article content

Several media outlets reported this month that the Vancouver school board and other districts in the Lower Mainland are planning to introduce AI chatbots into classrooms as soon as this academic year.

Article content

In one article, VSB associate superintendent Pedro Da Silva said that the VSB anticipates that AI is going to be “incredibly important” in supporting learning and in pathways to work and post-secondary education in the future.

Article content

Article content

Article content

I’ve been a teacher educator in B.C. for 15 years. So let me speak directly to VSB and other school districts considering “partnering” with Microsoft (and others) to incorporate generative AI into high school classrooms: You don’t have to do this.

Article content

Article content

As a post-secondary educator, I can report that the “importance” of Gen AI is greatly exaggerated. The main thing I have seen it do is help students to outsource the very things that people ostensibly come to university to do: reading, writing and thinking. I have seen AI produce false quotes from works of literature, introduce fake sources into bibliographies, fabricate people who don’t exist in reports and “edit” student work until it’s indistinguishable from advertising copy.

Article content

I may sound like an anti-technology crank (to be fair, I kind of am), but the most important thing that anyone who works in an academic institution needs to consider when they think about embracing generative AI is this: what these chatbots produce isn’t information.

Article content

It’s difficult to explain this to students (and administrators) who are used to “searching it up” on Google to find “answers” but what large language models produce are plausible sequences of words (not even words, but tokens — parts of words turned into numbers to make them processable) that are similar to the sequences they have been trained on. These large language models (LLMs) are very good at this; they’re so good that they accidentally produce things that resemble human-produced texts intended to communicate verifiable facts. But the relationship of the text produced by large language models and things that we think of as vitally important in education — truth, facts, information, knowledge — is incidental.

Article content

Article content

This is why Gen AI constantly and confidently produces text that is simply wrong. I ask it to list former members of my favourite band, and it adds a guitarist who never lived. I ask it for a list of relevant academic sources about a theory my student wants to learn more about and it gives me five imaginary articles with fake URLs. I ask it to generate a biography of a scholar I know, and it tells me they’re an expert in things they’ve never written about.

Article content

Article content

Some LLMs are better at avoiding this than others. Those that explicitly include footnotes from online sources are preferable to others, but all of them produce text that doesn’t correspond to reality, and only a person with deep knowledge of the subject matter that they’re trying to learn about has the acumen to tell the difference. The best description of an LLM still comes from a tweet by Jon Christian, the editor of the tech media company Futurism, who described it as “a virtual dumbass who is constantly wrong.”

Article content

The great thing about this technology is that you can just … not use it. No one is making you. And if your job requires you to, you can just learn as you go, kind of like I learned how to make Pivot Tables in Excel at my first office job after college.