The Elephant in the Classroom: ChatGPT

It's here, so let’s just talk about it. Are you reading robot essays? When you ask students to paraphrase, are they secretly laughing? Maybe, but maybe not completely…yet. Change is happening fast though. The elephant in the room is ChatGPT and its peers. ChatGPT was released for public use as a prototype on November 30, 2022. It is a large language model trained artificial intelligence (AI). It is the newest, flashiest, text-generating AI around, but there are many other AI text-generating tools out there that have been here for a while. ChatGPT finally grabbed the attention these tools deserve. ChatGPT is going to get better fast with the improvements gained through opening up use of this tool to the world. Articles and podcasts are now popping up daily about AI writing tools, and people are worried about the next generation’s ability to write.

But hold on, let’s learn a little more about these tools before we go there. According to Nick Duncan on ContentBot.ai “An AI writer is software that makes use of artificial intelligence to predict text based on input that you supply it. AI writers are capable of creating marketing copy, landing pages, blog topic ideas, slogans, brand names, lyrics, and even full blog posts.” (No, this blog post is not written by a robot, although the thought did cross my mind, and I might try it in the future.) Once most people hear about these tools, they want to try them for themselves, so here’s a short list of free AI text and image generating tools.

• ChatGPT writes text from your prompts

• Paraphraser.io paraphrases content you feed it

• Moonbeam writes content based on your choice of genre and details you provide

• Sudowrite is a creative writing tool that is an “always available brainstorming partner”

• DALL·E 2 creates realistic images and art from a description in natural language

For the nerdy minds of the group, Lex Fridman, Research Scientist at MIT, computer scientist, and artificial intelligence researcher talks about the way ChatGPT was designed and why this publicly released version works so well. Fridman explains that ChatGPT “was trained on code, and more data that’s able to give it some reasoning. Then, and this is really important, it was fine-tuned in a supervised way by human labeling.” Fridman describes that this human labeling pointed the text-generating AI “in the right direction that aligns with the way human beings think and talk.” Following that process, lots of humans ranked lots of responses generated by ChatGPT to refine the AI even further. Fridman suggests that the human ranking of the quality of text generated, used together with a technique called reinforcement learning (learning the optimal behavior in an environment) was able to get this tool to generate output that is very impressive to humans.

Let’s get back to the fear that students won’t be able to write. Dr. Bret Weinstein, podcaster, author, and former professor of evolutionary biology at Evergreen State College in Washington State, comments that the big concern with this technology is the ability to “fake expertise.” He posits a world of a “Cyrano de Bergerac dystopia where everybody is using this thing behind the scenes in order to say things that are beyond their own capacity to articulate.” Text-generating AI is not there yet, but that reality may exist soon. To get ahead of that, OpenAI, the maker of ChatGPT released an AI detection tool on January 31, 2023. According to an article in The Guardian (December, 2022) OpenAI wants to counter the risk of what is being termed “AIgiarism” by watermarking the output “by subtly tweaking the specific choice of words selected by ChatGPT.” OpenAI says this “should not be noticeable to a reader, but would be statistically predictable to anyone looking for signs of machine-generated text.” OpenAI is warning that this newly released detection tool is “imperfect” and shouldn’t be relied on when making decisions. This is their first version of the detection tool, so it will undoubtedly improve quickly. In the meantime, a computer science major from Princeton University, Edward Tianspent, created GPTZero, a free and publicly available tool that can detect whether a piece of text is written by a human or machine.

So where do we go from here? Don’t give up, embrace. We go to the same place teachers went 50 years ago when electronic calculators began entering the classroom. We build on the technology to advance our students' learning to higher levels. We incorporate text-generating technologies into writing and teach students how and when to use them ethically and professionally. Maybe students in the next generation will no longer write every word like we do now. Maybe they won’t know all the rules of syntax that we do because the computer does that job (this is true today). It can be worrisome to contemplate. But, instead of thinking about what they won’t do anymore, what if we contemplate what they will do. Maybe that extra time will be used to think deeply, to compare the merit of ideas, to create who knows what. If you can’t imagine what good can come of this, teach your students about these tools and ask them. They will have answers.

References

Associated Press. (2023, January 31). Cheaters Beware: ChatGPT Maker Releases AI Detection Tool. VOA. Retrieved February 1, 2023, from https://www.voanews.com/a/cheaters-beware-chatgpt-maker-releases-ai-detection-tool/6942134.html

Rogan, J., 2023. #1919 - Bret Weinstein. The Joe Rogan Experience. [Podcast]., [Accessed 29 January 2023]. Recorded January 4, 2023

Rogan, J., 2023. #1934 - Lex Fridman. The Joe Rogan Experience. [Podcast]., [Accessed 1 February 2023]. Recorded January 31, 2023.

Hern, A. (2022, December 31). Ai-assisted plagiarism? ChatGPT bot says it has an answer for that. The Guardian. Retrieved February 1, 2023, from https://www.theguardian.com/technology/2022/dec/31/ai-assisted-plagiarism-chatgpt-bot-says-it-has-an-answer-for-that