Over the past couple of years, generative AI tools have gained significant traction in the mainstream. Millions of people are now chatting with ChatGPT to write essays or code, using DALL·E to create art from text, and even trying new search engines like DeepSeek that use AI for direct answers. As one researcher puts it, this “rapid rise and widespread adoption” of generative AI has “transformed our lives”.
One such researcher is Chengzhi Zhang, a PhD student at Georgia Tech specializing in generative AI literacy – essentially, how people learn to use tools like ChatGPT effectively. In Zhang’s words, she focuses on “how to help people understand generative AI and how to use it more responsibly, critically, and effectively.” Working with Dr. Brian Magerko and the Digital Media department, Zhang is exploring ways to educate the public about these technologies. She emphasizes that generative AI should be seen as a helpful assistant or tool, not an all-powerful oracle. After all, these systems can make mistakes. For instance, a chatbot might “exaggerate wrong info” it finds on the internet, so blindly trusting every output isn’t wise. The goal is to utilize AI effectively and ethically, leveraging its benefits without falling prey to its pitfalls.
The Eliza Effect
As generative AIs like ChatGPT get more impressive, it’s easy to forget their limits. Many users start to feel like the computer programs truly understand them or are smarter than it is. There’s even a name for this tendency: the “Eliza effect.” The term originates from ELIZA, a 1960s chatbot that deceived many people into believing it could empathize. Today, the Eliza effect can be used to describe how people often attribute more intelligence to AI programs than they actually have. We see this whenever someone pours out their emotions to a bot, believing the software genuinely grasps their feelings, or when people take an AI’s confident-sounding answers at face value.
Zhang warns against falling into this trap. No matter how fluent or caring an AI’s response seems, “it’s important not to view AI as a person,” she says. At the end of the day, tools like ChatGPT lack human understanding and emotions – they generate text by matching patterns in massive datasets. Overestimating their intelligence can lead to real problems. For example, a student might assume, “ChatGPT helped me write this essay, so it must be correct,” and then trust everything it outputs as reliable learning materials, even if it is not always true.
Understanding the Eliza effect is a crucial step toward achieving better AI literacy – the fundamental knowledge and healthy skepticism we all need when interacting with these intelligent machines.
New Guidelines for AI Literacy
Zhang and Magerko are determined to bridge the gap between what generative AIs can do and what people understand about them. They argue that “a lack of robust generative AI literacy” is preventing people from using tools like ChatGPT “effectively and responsibly.”
In a recent paper, Zhang and Magerko proposed 12 guidelines for generative AI literacy to tackle this problem. These guidelines cover key points, including crafting effective prompts, staying vigilant for misinformation or bias in AI outputs, and understanding the broader social implications of AI. The goal is to help people use generative AI in an “efficient, ethical, and informed way”.
For example, one guideline emphasizes how you phrase your request to the AI. Even minor tweaks in wording can make a massive difference in the answers you get. Another guideline is to be aware of the AI’s limitations: just because a chatbot sounds confident, it doesn’t mean it’s correct, so always double-check important facts instead of assuming the AI is right.
An Evolving Field
Zhang and Magerko acknowledge that their guidelines are not the final word but a starting point. Generative AI is “a relatively new and rapidly evolving field”, so our approach to AI education must also evolve. AI literacy guidelines must be continually updated to meet the new challenges that each generation of AI models presents.
The researchers see their work as just the beginning. “Our work aims to construct a holistic guideline that serves as a starting point for… future discussions,” they write. As more educators, companies, and policymakers join the conversation, these initial guidelines can be refined and expanded into more detailed teaching resources and strategies.
They also emphasize the importance of staying flexible and informed. Generative AI technology changes week by week—one software update can suddenly enable a chatbot to browse the web, and each new image model is more realistic than the last. With the “fast-paced evolution of generative AI,” any educational approach must be updated accordingly.
Zhang hopes that improving public understanding of AI’s capabilities and limitations will empower people to navigate the ChatGPT era with curiosity and caution. With the right literacy skills, we can enjoy the benefits of generative AI, like help with writing or creative inspiration, without falling for its illusions or hype. In an era when AI is ubiquitous, a basic understanding of it goes a long way.