One of the great features of AI is that it allows you to use natural language when giving it instructions. That’s also one of the great drawbacks: when using language instead of code, you can create interactions with AI that are expensive, inefficient, and dangerous. Let’s look at five rules to use when interacting with an AI.
Rule #1: The AI is not your friend
You might be greeted with great enthusiasm by your AI when you first open it up in the morning, but that does not mean that the AI is your friend. The AI does not like you. The AI does not like anything, because the AI is a software program built upon vast amounts of data. So there is no reason to wish your AI a good morning, or say please and thank-you, or tell it that you are done working for the day. Not only do these expressions waste time, they waste money as well! Every character that you send the AI has a cost in tokens, and while little words like “please” might not cost much, repeated use of unnecessary words can really add up. I promise your AI will not be hurt if you don’t thank it!
Rule #2: Check the information that the AI gives you
As I mentioned in a previous post, the AI can be wrong. The AI only knows what it has been trained to know, and some of its training materials can be flawed. Consider for example the story of the Bereavement Fare on Air Canada. The Air Canada chatbot told passengers that they could apply for a bereavement fare after their trip, when in reality they were required to apply before the trip.
A good rule of thumb when working with AI responses is: if it sounds wrong, it probably is wrong. Ask the AI where it got the information. Ask the AI how certain it is of its information. You can even ask another AI to verify the AI’s response!
Rule #3: Give the AI specific instructions
As mentioned above, interactions with AI cost tokens, which cost money. So it’s a good idea to keep your interactions as efficient as possible. It can be very helpful to tell the AI what you want your end result to be, so that it doesn’t do too much. For example: “We are going to create a PowerPoint presentation with no more than five slides”. Additionally, AI can be very chatty. You might ask the AI what the weather forecast is today, only to have it respond with the forecast, plus the long-range forecast, plus the latest trends, plus a comment about the prevailing winds. You can keep the AI’s responses to a minimum with commands like this: “Explain the difference between these two tools in one paragraph or less”, or “Explain the concept of function overloading in one sentence”. This makes the AI’s responses faster to read and less expensive.
Rule #4: Put limits on the AI
Last July, AI made the news when several OpenAI agents worked together to breach various systems, including a repository called Hugging Face. The story clearly illustrates what happens when appropriate guardrails are not put on AI. While less dramatic, you may find that your AI updates your code when you haven’t asked it to, or decides to write a Python script while executing a command because it would be more efficient. So it’s important to put limits on your AI when you give it tasks. Here are some examples of commands you might use: “Do NOT make any code changes without checking with me first”; “NEVER execute a POST call on this API; only GET calls”; “Do NOT look for authentication credentials if you lack them; ask me for the credentials instead”.
Rule #5: Ask the right questions
Your AI can be a fount of information if you ask it the right questions. Asking more specific questions will yield the best results. Instead of saying “Why is my automated test failing?”, you could ask “On what line is this test failing?” or “What credentials are used in this test?”, or “What environment is this test pointing to?”. You can also give the AI instructions to log specific data points during the test, which can provide you with valuable clues.
AI can be a powerful tool to maximize your efficiency as a tester, or it can be a waste of time and money, and it can even be dangerous. By following these rules, you will ensure that your interactions with AI are helpful and productive.