The Hidden Mental Toll of Using AI for Complex Projects: A Personal Experience

Sometimes, trying to write or review academic work with the help of AI chatbots can cause more mental stress than simply writing from the brain—drawing on how you feel, your knowledge, and established facts.

In this blog, I'll share a personal experience to illustrate this kind of stress.

AI chatbots and tools have access to enormous amounts of data—both on and off the internet—and can process this information in seconds to predict what they think is most appealing or acceptable to the user.

Let me take you back to December 2024, when I started writing my dissertation. I began with a popular AI chatbot to get a clearer direction on what to research. My first query asked for suggestions on areas in optimization I could explore. The chatbot returned a long list of options! I reluctantly picked one, even though many others also looked promising.

Next, I asked for possible topics within that area. Again, I received another long list. Choosing a specific topic became difficult because almost every option seemed exciting. Whenever I picked one and asked for more details, I was met with even more overwhelming information. In the end, I couldn't settle on any topic. I had to approach my professor, who gave me a precise, focused list of topics under optimization—a huge contrast to the ocean of options I had initially encountered.

Later, during my literature review, I prompted the chatbot to search the web for local literature and provide a summary. It generated a smooth review with a long list of authors. My excitement quickly faded when I noticed that many of the local papers it returned were supposedly authored by my professor. This intrigued me but also made me skeptical. I minimized the chatbot window and started a manual internet search using each paper title attributed to my supervisor. To my surprise, only 20% of them were real. The rest? Completely fabricated—fake authors, fake titles, fake publication dates, and fake journals.

When I asked, "Are the papers above real or fake?" I got a shocking response. The chatbot admitted that 80% of the references were hypothetical—merely templates—and apologized for not stating that earlier.

Apology accepted. I decided to keep the real ones and ignore the fake ones. But I had to repeat this process for almost every author whose work I wanted to explore in inventory optimization. The result? Weeks slipped by as I grew increasingly fatigued from verifying whether the papers I was reviewing actually existed. At some point, I had to abandon most of the papers and the entire draft of my literature review—because many weren't relevant to my research or weren't open access. More weeks were lost.

I decided to slow down and take a more strategic approach. I registered on academic platforms like Academia.edu and ResearchGate, where I could find niche-specific and open-access papers. I downloaded relevant papers and even reached out to authors who hadn't uploaded their work. Surprisingly, I got access to more relevant and recent papers than what the chatbot had ever provided.

I carefully read through the ones most specific to my work and scanned the abstracts, introductions, results, and conclusions of others broadly related to my topic. The result? Within a week, I had gathered almost all the papers I needed and completed the literature review expected for my dissertation—using the conventional method, with the help of a grammar checker and minimal assistance from the AI chatbot. This saved me a tremendous amount of mental stress.

That was my first experience using AI for a critical, complex, and rigorous academic task.

AI chatbots are amazing for many things. However, they may fall short when it comes to niche-specific projects. Even if they have limited knowledge about a topic, they often try to say too much—resulting in fabricated facts.

If you're working on a critical project, I recommend avoiding general-purpose large language models (LLMs). Instead, go for AI tools built specifically for your task. For example, uploading 20 research articles to an overtrained or undertrained multipurpose LLM for a thorough literature review may not give you the best results. It might skip important details, force connections between unrelated papers, or overwhelm you with excessive information that takes even more time to filter.

Every AI tool has its strengths—just like every human. You wouldn't use a social media chatbot for rigorous scientific research or advanced image generation and expect high-quality results. Those tools are better suited for writing replies, social posts, month greetings, birthday wishes, and other general content.

So, next time you have a project—especially a critical, complex, or rigorous one—look for dedicated AI tools. We all love that "I too know" classmate when it comes to debates and quiz competitions. But when it comes to football or clubbing, they might not be the best choice.

📖 ALSO READ: Automation Is Coming - Are We Ready?

This blog will be shared on my LinkedIn feed, and all engagements are welcome. Feel free to drop your own experiences in the comment section and hear from others too.