Chatbots: Here, There and Everywhere

Note: This essay originally appeared on LinkedIn on October 17, 2023.

Chatbots have been around since the 1960s when Eliza taught us that humans often lose their perspective on things when they converse with machines. The collective loss of perspective–a mass Eliza effect?-- generated by OpenAI’s decision to hook up a chatbot interface to its Generative Pre-trained Transformer has inspired a wave of chatbots, now sometimes called copilots, being added to existing software or launched as their own product. Walking around the Educause 2023 Exhibit Hall it was clear that the years of restraint in the aftermath of Clippy are over. Chatbots, as the Beatles song goes, are here, there, and everywhere.

Most of the chatbots and embedded AI tools I saw were pitched as labor-saving devices for teachers and administrators. This makes sense as the first use case for ChatGPT, as we discovered last December, was students using it to save time and effort on their homework. Performing mundane tasks–developing lesson plans, creating problem sets, producing meeting minutes, coding basic functions, and of course, writing essays–-are tasks LLMs can do well enough in a rising tide lifts all boats kind of way. So as the tide of generative AI rolls in changing our lives with each wave, what to make of it?

One of the themes of the conference was creating a frictionless student experience, but the instructional chatbots are no longer frictionless. We don’t want them giving the answers away. When interacting with students, they produce friction in the form of questions and interruptions designed to prompt reflection and work, all in the service of leading students to knowledge. The slow realization that LLMs are not enhanced internet search, but something else entirely is starting to get worked out in many of the products and services I saw. “Let’s save the time and effort of teachers and staff while forcing students to learn” is the general framework for how generative AI is supposed to improve education.

Most of what I saw at Educause was consistent with the feelings expressed by the Beatles and with Gartner’s notion that we are at the peak of the hype cycle for generative AI. What comes next is not hard to predict. The rush of emotion, the feelings of “But to love her is to need her everywhere,” turns to doubt that this is such a good thing. We will go from “Nobody can deny that there's something there” to “Someone is speaking, but she doesn't know he's there.” Along with the inevitable backlash stories asking “where is the return on the huge investments in AI?” there will be stories about the disappointments and failures of edtech to realize the high expectations created by the hype.

Those of us who have been through hype cycles before and know the history of Eliza and its effect on humans won’t be surprised by what comes next. I hope that as we get on with the work of figuring out what these new tools are good for, we will let go of the fears and anxieties that accompany those initial feelings of excitement. And instead, choose clear-eyed analysis of what we need these tools to accomplish and build machines for that purpose. There are plenty of problems to solve and challenges to overcome in higher education. Machines that help us teach and learn will make a difference in our work as educators, even if they won’t be everywhere.

Just like love songs don’t have to be sung by men about women, chatbots don’t have to be gendered female. Must we anthropomorphize chatbots? Does Khanmigo have a gender? Will the coming disillusionment be an opportunity to rethink how we design and when we need chatbots?

Let me know what you think in the comments. And please consider using the like button to help others find this post.

𝑨𝑰 𝑳𝒐𝒈 © 2024 by Rob Nelson is licensed under CC BY-SA 4.0.