1.5: Where AI gets it wrong- hallucinations, bias, outdated info
Demystifying AI: A Practical Introduction for Instructors · LibreTexts
0%
Contents
Text
Size
From LibreTexts, under CC BY 4.0, as the source states it; its equations are written in TeX, which the PDF shows typeset. This page at LibreTexts.
1.5: Where AI gets it wrong- hallucinations, bias, outdated info
Opening the book…
Hallucinations, Bias, and Outdated Information
AI tools like ChatGPT can sound confident, fluent, and even brilliant—but they’re far from perfect. In fact, AI models make mistakes often. And not just small ones—they can invent facts, reinforce bias, or give you information that’s just plain wrong.
In this section, we’ll look at where AI stumbles—and why those errors matter in an educational setting.
???? What Are Hallucinations?
In the world of AI, a hallucination refers to when the model generates content that sounds believable but isn’t factually accurate.
Examples:
Citing a research article that doesn’t exist
Making up statistics
Describing historical events or people inaccurately
The AI isn’t trying to deceive anyone—it simply doesn't know what’s true. It’s predicting what a “correct-sounding” answer should look like based on its training data.
Instructor tip: Always double-check AI-generated content, especially when using it for assessments or instructional materials.
⚖️ What About Bias?
AI models are trained on real-world data—books, websites, forums, and more. But that data reflects all the biases, stereotypes, and inequities found in society.
That means AI outputs may unintentionally:
Favor dominant cultural narratives
Reinforce gender or racial stereotypes
Marginalize non-Western perspectives
These patterns aren't always obvious at first glance, but they can show up in subtle ways: through examples, tone, omissions, or even default assumptions in writing.
Instructor tip: Use AI as a starting point, not an unquestioned authority. Critique the output like you would any other source.
???? Why Is the Information Sometimes Outdated?
Most language models aren’t connected to the internet in real time. For example:
ChatGPT 3.5 was trained on data up to September 2021
GPT-4 (as of early 2024) may have limited web-browsing depending on platform
So if you ask an AI about recent laws, technologies, or events—it might give an outdated or incorrect response.
Instructor tip: When asking AI about time-sensitive topics (like COVID policy, AI regulations, or current events), check the model’s knowledge cutoff or supplement with recent sources.
???? Key Takeaways
AI sounds confident, but that doesn’t mean it’s right.
Always fact-check and consider potential bias, especially in student-facing materials.
