Asking AI questions: what's the best way to do it?

AI makes a lot possible, but it takes some getting used to how best to use it. There's no such thing as dumb questions, they say. But good questions do! And to get the right answers from AI, you have to ask the right questions. We at Pink Elephant are busy expanding our services around AI. Here, we are working closely with our colleagues at Finnish company Sulava on the one hand and Microsoft itself on the other. After all, our PinkPilot service is based on Microsoft Copilot technology.

We find that users sometimes still find it a bit difficult to define the right questions while using Copilot and other AI systems. Therefore, we have created an overview.

How do I ask the right questions to AI / Microsoft Copilot / ChatGPT?

Artificial Intelligence (AI) is currently revolutionising the way we interact with technology. ChatGPT and Microsoft Copilot are leading the way in this regard, but almost all other major technology companies are working on it and developments are moving fast. And that makes sense, because AI can make many things easier. By asking the right questions, it can write text for you, create or modify images and much more.

So asking the right questions is key to making the best use of AI. But how do you define those questions? What information should you provide for best results? Here are the key elements to a specific question. Of course, your questions can also be shorter and do not have to contain all four elements. The question "which film won the Oscar for best film in 1969" can also be answered by AI ("Oliver!")

You can also download this information: Questions-ask-AI-v1.pdf

The four elements of a good question:

  1. OBJECTIVE: What do you want from AI? For example, "I want a list of the 5 best films..."
  2. CONTEXT: What is it for and who is involved? For example, "... for a children's party for 12-year-olds who all love to dance".
  3. EXPECTATIONS: What should AI consider in the answer? For example, "List a brief description of the plot of each film..."
  4. SOURCE: What sources should AI use? For example: "... and focus on Dutch films from the past 10 years"

Incidentally, note that AI only knows what the system has "read". In some systems, the information read is at least a year old, for example. So then you won't get results from the past year!

Tips

  • An example of a delimited question: "make a summary of the presentation on the new products"
  • An example of a question with purpose, context and expectations: "create an outline for a training document on the new intranet. The audience is highly skilled professionals who work from home or at customer sites a lot and have a lot of virtual Teams meetings. The tone of the document should be informal and friendly. 
  • Microsoft Copilot is integrated into all kinds of Microsoft software, such as PowerPoint. For example, in PowerPoint you can ask, "Create a short presentation about the book "The Orphanage in the Azure Sea". In Microsoft Word, you can select a paragraph of existing text, then click on the Copilot icon and then choose "Rewrite with Copilot." 

But pay attention!

  • Check and verify AI results: if AI bases its results on wrong content, the result may be incorrect or nonsense. Check for facts yourself!
  • The same question can yield different answers: if the set of data AI relies on changes (gets bigger, or adds more recent information), the answer to the same question can change over time
  • Use AI in an ethical way: try to avoid hurting other people, or yourself, by the content you let AI generate.  

How does AI work? (a brief, incomplete and certainly not scientifically based explanation) 

You often hear the abbreviation LLM, which stands for Large Language Models. This means that the system has "read" a lot of texts and, based on that, guesses what the answer is to the question being asked. This guessing is based on probability. In X% of the texts it "read", the word "pink" was followed by the word "elephant". In Y% of the cases, it was the word "candy cane". Based on context and much more, the system decides what the next word will be. And so on. 

The other side of AI

AI makes a lot of things possible. But it also takes a lot of electricity and processor power to arrive at an answer. Countless data centres will be needed to support AI. Asking the right question as soon as possible therefore saves a lot of energy. 

share this message

Related articles