The AI Wrote My Homework

Recently, there has been a lot of news about the text-based chatbot called ChatGPT. The tool has taken the world by storm, and its servers are having trouble keeping up with the demand. The service had over a million users register a week after launching in late November 2022. Over the next few months, that number will grow significantly. ChatGPT has already sparked fascinating discussions amongst Atla members speculating about its impact on students. If you are wondering what ChatGPT is, let’s dig deeper.

ChatGPT, which stands for Generative Pre-trained Transformer, is a large language model chatbot.  The service was launched by an artificial intelligence (AI) research and deployment company called OpenAI. The capped-profit company aims to build safe and beneficial artificial general intelligence tools. Along with ChatGPT, OpenAI is the creator of DALL·E, an AI system that creates images and art, only needing a natural language description as the input. OpenAI continues to work on various projects, some commercial and others free. ChatGPT is currently in a free preview.

Over the years, AI systems were typically trained in specific tasks, such as a question-and-answer AI (IBM’s Watson), AI-powered devices, and conversational AI (Amazon’s Alexa). These systems perform specific tasks well but fall short when they stray from what they were programmed to do.  OpenAI attempts to push the boundaries of AI, not focusing on one area but building systems with artificial general intelligence (AGI). This type of AI allows for greater flexibility in how the system responds.

GPT History

Open AI’s first natural language processor, GPT, was released in 2018. GPT was primarily trained on the BookCorpus dataset, which consists of around 7000 works by unpublished authors. The following year, OpenAI released GPT-2. In addition to the data used in GPT-1, GPT-2 was trained on an OpenAI dataset called WebText. The content in WebText was retrieved by scraping data from web pages and their outbound links with an emphasis on quality and was approximately ten times the size of the BookCorpus data. The trend continued with GPT-3 utilizing five datasets — Wikipedia, Common Crawl, Books1, Books2, and WebText2. These datasets account for about 45TB of text, which is equal to hundreds of billions of words. That finally brings us to GPT-3.5, also known as ChatGPT. ChatGPT is a less powerful variation of GPT-3 that specializes in natural-sounding conversation responses.

If you want to learn more about the models these systems were based on, see Improving Language Understanding by Generative Pre-Training, Language Models are Unsupervised Multitask Learners, and Language Models are Few-Shot Learners.

Landscape

Microsoft saw OpenAI’s potential a few years ago and invested one billion dollars in the company.  In the last few weeks, Microsoft announced it would increase its investment to ten billion. Microsoft plans to bring ChatGPT functionality to its Bing search engine platform and other applications. This prospective partnership has the existing search engine giant, Google, very concerned, and Google is investing billions into its own AGI platform to compete.

Both Microsoft and Google are actively deploying AI already. Have you noticed AI trying to finish your sentence as you write an email? Over the last twenty-plus years, AI has been slowly integrated into our lives, and AGI is just another step forward. Smaller players can also build on these platforms, bringing new capabilities to existing products and services.

Practicality

Generally speaking, the use of ChatGPT is entertaining. My daughter and I experimented with it by requesting the creation of stories and poems using a few random topics.  She felt the tool’s responses were sometimes strange and unexpected and other times silly, fun, and creative. When asking ChatGPT simple questions, it does a pretty good job of answering. Unlike a search engine that provides 20,000 websites that may answer the question, ChatGPT gives a single response. That is the joy and drawback of the system; right or wrong, it confidently gives you an answer, sometimes providing a spectacularly wrong answer.

The AI can be helpful with administrative tasks; for instance, it has the ability to write emails with a few simple directions. When a student needs to write a report, the AI tool will attempt to take care of grammar, general research, and repetitive tasks, allowing the individual to focus on the advanced research required to make the document shine. This ease of use has many in the education field concerned. 

Future

In this new AGI world, what does it mean for higher education? The overarching bans on ChatGPT at a few institutions appear to be excessive. However, the concerns about students using ChatGPT for assignments are valid. Fortunately, OpenAI and others have released tools that allow you to detect the probability of AI output. Some faculty are embracing ChatGPT by using it to generate ideas, quiz questions, lesson plans, and handling frequently asked questions.

Students have a massive amount of information at their fingertips. Even with these powerful tools, what has not changed is true learners still need to seek and synthesize information. Answers to questions in the humanities are typically nuanced. This is where AI starts to break down, and librarians must be ready to step in. There will be more tools like ChatGPT, so it’s essential to understand its application and look for ways to leverage AI to support students, scholars, and staff. If you haven’t already, I encourage you to talk with colleagues to learn how they use ChatGPT. Their responses may surprise you.