Ai Assistant

3 Tips for Working with an AI Assistant

With artificial intelligent assistants on the rise, it is important to remember what AI is and what it isn’t in order to get the best work out of a virtual assistant. Most of the common AI chatbots and assistants out there (such as ChatGPT and Gemini) are language learning models (LLMs). This means that they are programs trained on large sets of data and use pattern recognition to answer questions. They are not generating original data; they are simply sifting through knowledge, aggregating it, and providing an answer that sounds right.

And therein lies the problem. AI sounds right. It knows how to craft a sentence that sounds right on the surface but can break down under scrutiny.  So here are some tips to transform your assistant’s responses from AI slop to AI-assisted gold.

Be Critical of AI’s Thinking

To get the best responses from AI, you need to understand its limitations. An LLM is trained on a specific dataset and cannot produce data outside of that dataset. It has no access to the most recent sources of information, such as journal articles or research papers. You might not know what data it was trained on, so you don’t know what type of biases it may have. If an LLM was trained on data that favors one group or ideology over another, then any response you get from it will also have that bias. Depending on your prompt, this bias may not be immediately clear. You may not have enough information to recognize the one-sided information.

A robot handing a lightbulb to a man sitting on a question mark
Image by Mohamed Hassan from Pixabay

To combat possible bias, it is important to ask questions about what AI is telling you:

  • Is there a missing point of view or important information?
  • Is there any inherent bias in the language or ideas?
  • What sources is the response pulling from?

This last question leads to the next point: checking sources.

Are You Hallucinating?

A known problem with AI chatbots is that they can make up sources. A chatbot can present you with a reasonable-sounding claim and even provide a source from a journal you’ve heard of. However, when you go to find the article yourself, you can’t find it. Are you hallucinating? No, but the chatbot is. The AI wants to provide an answer—and if it can’t find one, it will make one up. It is vitally important that you check the sources yourself before using them in an assignment. The fastest way to know someone used AI—not as an assistant but as an author—is to check hallucinated sources. So, what can you do to avoid this pitfall?

Turn AI Sideways

St. Catherine University encourages students to use the lateral reading strategy to evaluate AI claims. Lateral reading involves five steps: Break it down, search, analyze, decide, repeat/conclude.

Break It Down

First, look at the information and break it down into individual claims or arguments. This makes the next step easier.

Search

Searching means finding information to support previously identified claims or arguments. This way, you are using the AI response as an outline for your assignment. By finding (or not finding) information to support those claims, you are creating a stronger argument and avoiding hallucinated citations.

Analyze

Next, look back at your prompt. What did your prompt assume? What did the AI assume in response? Is there anything to challenge those assumptions? This is also a great time to investigate alternative perspectives. Look up who might be an expert in this topic and what they have to say regarding the claims AI has made.

Decide

Based on the evidence and information you have compiled from the previous steps, decide what is true, misleading, or requires further investigation. Ask the AI more about the response it has given. Try to poke holes in the argument and ask for more information.

Repeat/Conclude

Repeat this process until you are satisfied that what you have been given is what you need.

AI is the Assistant, not the Author

The purpose of AI is not to get a singular, correct answer. When used properly, AI acts as an assistant that helps you brainstorm, outline, and evaluate your own thoughts, ideas, and biases. It helps you find a starting point to dig deeper into a subject and proposes more ideas to explore. So, use AI to help with assignments—just remember to check your work.

Similar Posts