Way back in late November 2022, when ChatGPT was released by OpenAI, it was immediately clear that the implications of an easy-to-use and brilliant AI model on education were going to be profound. Recognizing the newfound ease with which students could now evade even the keenest of academic integrity watchdogs, K-12 schools and postsecondary institutions scrambled to understand what these new tools were all about. Then they began to consider and plan for what they would mean for classrooms, a feat that is proving to be more difficult.
Among educators and policymakers, the views on ChatGPT are relatively mixed. Some have tried to accommodate it as a new tool, while others have resisted it through bans or engaged in arms races to root out cheaters, sometimes using other AI tools. South Park tried using a falconer. Regardless of whether you think AI in education is a positive or negative development overall, it is here to stay, and it is changing education much more quickly than most anyone (critics or proponents) anticipated.
Since its launch, every day seems to bring a new development, whether a new use for ChatGPT itself, a new tool built on top of it, or a separate technology from another company. Many are text based, but some use or generate other media such as video, images, or audio. Any one of these tools could have been game-changing all by itself just a few months ago, but now there are so many that it seems almost impossible for society to simply keep track of them, much less take the time necessary to consider, digest, and adapt to their implications.
Just this week, OpenAI announced that ChatGPT is being superseded by GPT-4, which can receive input from both text and images. A demo showed incredible capabilities to code games or powerful applications without actually needing to know how to code. But a whitepaper issued by OpenAI along with the update showed some clear implications for education. Whereas ChatGPT (sometimes referred to as GPT-3.5) could reliably pass many of the nation’s most rigorous and path-defining examinations, GPT-4 seems able to exceed its predecessor’s capabilities at nearly all of them, registering as a top performer. As the chart below demonstrates, the technology has rapidly improved in its ability to pass tough exams in math, science, coding, law, and knowledge of wine.
Of course, this does not mean that AI will literally be able to take the test for students. Most of these testing organizations have checks in place to prevent cheating, such as requiring students to show up in person without their phone or laptop at hand. However, high performance across such a wide range of domains will likely impact what students are expected to know in ways that we cannot even begin to imagine.
On the same day as the GPT-4 announcement, Google announced that its “Workspace” applications (e.g., Gmail, Docs, Sheets, etc.) will have ChatGPT-like AI tools built in. Users will soon, per the video, be able to write an email or have the AI summarize a long email thread that takes too long to read. It will design and write entire presentations, documents, or spreadsheets with limited direction. Within just hours of Google’s announcement, Microsoft announced Microsoft 365 Copilot, a similar AI tool for the tech giant’s own applications. While OpenAI’s tools are just now breaking into classrooms, Google’s software is already the default for 150 million students worldwide, and Microsoft’s is used by a similar amount. Presumably the companies will build in curbs or transparency on the use of AI to help teachers, but one expects that students will quickly discover ways around those.
It is tempting to extrapolate from all of this and assume that these capabilities will continue to accelerate indefinitely (what is sometimes called the “singularity”). However, as is the case in some of AI’s other applications (self-driving vehicles, for instance), certain important edge-case problems are proving very costly and time-consuming to solve. The same may prove true in education. A much larger problem remains society’s ability to adapt and determine those things which humans should know how to do, which are better offloaded to computers, and which we would prefer to keep human-generated no matter AI’s capabilities.
The post With GPT-4, Disruptive Changes Continue to Wash Over Education appeared first on American Enterprise Institute – AEI.