ChatGPT Isn’t a Source: Why Students Need to Fact-Check AI Homework

Why generative AI models hallucinate, how algorithmic bias impacts research, and practical steps for students to fact-check schoolwork.

AI in everyday learning

It’s not hard to see why people turn to AI: it’s convenient, it’s accessible, and it cuts down a mountain of workload in a matter of seconds!

I’ve spoken before about the rise of AI and how tools like ChatGPT are now embedded in many of our everyday lives (let alone students’ homework habits). However, learning how to properly fact check AI homework is becoming an essential skill as young people increasingly rely on these tools for schoolwork.

Generative AI cannot be relied upon consistently to give an accurate and verifiable response. The news media has covered instances where this technology has manufactured biased and even outright false content, which is why it’s doubly important for us to carefully scrutinise the information we might be consuming wholesale.

The illusion of AI accuracy

Surely, you’re mistaken, Cheryl! AI is a font of knowledge. I’ve used it many times with very satisfactory results.

That may very well be the case! I’m not saying that generative AI cannot form an effective part of a student’s toolkit – I’ll be the first to admit that I occasionally accept the AI overview from a Google search quite readily. That being said, care must be taken to ensure that this material is examined under a close lens rather than being accepted without question.

In a nutshell, generative AI models (like ChatGPT) work by processing and identifying patterns which are sourced from datasets, which are themselves primarily and publicly available on the internet. This means that generative AI produces information from these (often) unidentified sources, which aren’t immune to instances of bias.

Take for example, a case where ChatGPT produced resumes for hypothetical women portraying them as younger and less experienced or when Stable Diffusion generated more than 5,000 images of a typical ‘terrorist’ or ‘drug dealer’ with underlying racial and gender disparities. As noted by Dr Matthias Holweg: “A model… is biased by default of the sources it’s using.”

AI language models are such that they write in an authoritative tone, to give the impression of a sound knowledge base, even when that confidence is unwarranted. It’s not uncommon for generative AI to experience so-called ‘AI hallucinations’ - when a chatbot generates a citation or a statistic from thin air. These models are also known to be susceptible to sycophantic behaviour; AI chatbots may tell you what you want to hear without first checking if something is true.

The examples I’ve picked out are especially egregious, but oftentimes misinformation is harder to spot; thus, the danger in relying on AI responses is in accepting their plausibility and probability at face value.

What we lose when we skip traditional research

Fact-checking is a vital skill which AI can’t replace (yet!), and it’s something that we should continually engage in. When we stop fact-checking, we reduce our ability to accurately evaluate sources, and we suffer from the erosion of core critical thinking skills.

Even when taken in the context of schoolwork, children lose out on deep reading and comprehension skills which will result in long-term academic consequences. When we relegate our mind to AI, we attain only surface-level knowledge without a deeper understanding of the topic at hand.

An interesting article from Psychology Today writes in detail about the cognitive atrophy we suffer as a result of this over-reliance on AI:

• “Adults who offload thinking to AI lose capacity they built. Children may never build it at all.

• When students process information through the same model, the result may be similar minds.

• Auditing AI output requires expertise the child is still supposed to be developing.

• In a study, developers who delegated coding to AI produced working code but failed conceptual understanding.”

Practical ways to fact check AI Homework

Media literacy is an essential life skill that can teach us how to parse out the misinformation from the truth. When you fact-check AI homework, always consider multiple perspectives by checking against trusted sources like academic sites and journals as well as reliable news outlets without biases.

When a response is generated, carefully consider the citations: Are there any? Are they verifiable? Are they from trusted sources? The key is to adopt a healthy ‘pause-and-question’ habit.

In an academic setting, seek out an expert who can act as a trusted source of information, like a professor, teacher or tutor. We at Titanium Tutors have many experts on our books who can advise on a multitude of subjects and topics, and who would be a boon to your studies.

On a larger scale, proactively addressing the AI deception problem is crucial to ensure that AI technology augments rather than endangers the human experience.

Putting in all this extra work to ensure what AI feeds us is accurate is indeed more time-consuming, but real learning happens when we question, not when we consume.

AI is here to stay

For all the risks that generative AI poses, that isn’t to say that AI isn’t showing some spectacular real-world benefits to our society; we just need to remember how to use this tool wisely. Generative AI is a handy tool that is here to stay, but that doesn’t mean we have to accept all that it creates.

---

Blog Post Crafted by Cheryl

Cheryl manages our Admin Team, and is a qualified teacher with 5 years' experience in schools across England and Canada.

Cheryl graduated from University of Toronto with a Bachelor of Education. She tutored secondary school students in English for over nine years in Canada.

Cheryl speaks Cantonese, English and French, and in her spare time, she can be found illustrating and reading children’s books for inspiration.