4 min read
How to manage cognitive offloading as a tester
I use AI daily, and a couple of months ago I started to get worried, since I was doing less thinking and more prompting, and it started affecting me as a tester.
I use AI daily in my job, which involves a lot of high cognitive tasks such as testing and programming, and a couple of months ago, I started to get worried, not about AI becoming sentient and going full Terminator on humanity, but about the fact that I found myself offloading some of my mental processes to an AI agent.
It started with a test scenario that I needed to set up. I just had to create a page in a CMS with different components, and I thought, "Why don't I save some time? I use an MCP and let the agent create the page for me with what I need?"
And as if it were raw sugar, in a matter of days, I incorporated that into my daily testing routine when I was too busy, and slowly but surely, I kept using AI more and more for setting up some test cases.
At one point, I noticed that I was not fully sure how to do the setup of the test cases by hand, I mean I knew what I needed but I started needing AI more to explain how to actually do it step by step.
That is when I started to get worried, because as a tester I am suppose to be the person who notices things, and I was actually outsourcing the noticing to the AI.
This outsourcing has a name, and is cognitive offloading. We do this on a daily basis, like when we go to the supermarket and write the list of things we need to buy, which is cognitive offloading.
In the tech world, we have been offloading to notebooks, calculators and Stack Overflow for years, but the difference now is that the tool we are offloading to can do almost any cognitive task we throw at it, which sounds great until you look at what happens to the brain underneath.
While I was reading to learn more about cognitive offloading, I found a 2021 paper in which researchers asked people to do a simple copy task, half with an external tool to store information, half without.
The half with the tool finished faster, and then on a surprise memory test afterwards, they did much worse.
The researchers wrote something that resonated with my experience: "while cognitive offloading accelerated task processing, it interfered with the formation of memory for the processed information."
That resonated with me because I was feeling the same effects. I was faster, but I did not feel I was becoming better.
Now, why am I concerned about this? Because I think a tester is the worst possible role to be letting the AI take the reins and to stop thinking. The job of a tester is to learn, notice, gather information, and investigate.
To spot things nobody asked us to spot, to read a stack trace and feel that the symptom doesn't quite match the message, and to communicate our ideas and the information we gather.
So what can a tester actually do about it? I do not have a clever framework for this, and I do not think you need one.
What I have are a few personal rules that, for me, are starting to help, and that might also help you, too:
- Don't open the AI for the first 15 minutes of a problem
- Explain the problem before you prompt it
- Pick one test case or scenario a week and diagnose it fully solo
- Write down what you used AI for at the end of the day, you will be surprised how much offloading happens
- Read the documentation before asking the AI for a shortcut
- Try to break the AI's answer before accepting it
- Pair with another tester instead of with an AI once a week and discuss your ideas
- Spend one hour a week reading the codebase without any help
Well, I hope you enjoyed the article, and remember, as tempting as it might be, don't stop thinking!
Link to the article mentioned above: https://pmc.ncbi.nlm.nih.gov/articles/PMC8358584/
It started with a test scenario that I needed to set up. I just had to create a page in a CMS with different components, and I thought, "Why don't I save some time? I use an MCP and let the agent create the page for me with what I need?"
And as if it were raw sugar, in a matter of days, I incorporated that into my daily testing routine when I was too busy, and slowly but surely, I kept using AI more and more for setting up some test cases.
At one point, I noticed that I was not fully sure how to do the setup of the test cases by hand, I mean I knew what I needed but I started needing AI more to explain how to actually do it step by step.
That is when I started to get worried, because as a tester I am suppose to be the person who notices things, and I was actually outsourcing the noticing to the AI.
This outsourcing has a name, and is cognitive offloading. We do this on a daily basis, like when we go to the supermarket and write the list of things we need to buy, which is cognitive offloading.
In the tech world, we have been offloading to notebooks, calculators and Stack Overflow for years, but the difference now is that the tool we are offloading to can do almost any cognitive task we throw at it, which sounds great until you look at what happens to the brain underneath.
While I was reading to learn more about cognitive offloading, I found a 2021 paper in which researchers asked people to do a simple copy task, half with an external tool to store information, half without.
The half with the tool finished faster, and then on a surprise memory test afterwards, they did much worse.
The researchers wrote something that resonated with my experience: "while cognitive offloading accelerated task processing, it interfered with the formation of memory for the processed information."
That resonated with me because I was feeling the same effects. I was faster, but I did not feel I was becoming better.
Now, why am I concerned about this? Because I think a tester is the worst possible role to be letting the AI take the reins and to stop thinking. The job of a tester is to learn, notice, gather information, and investigate.
To spot things nobody asked us to spot, to read a stack trace and feel that the symptom doesn't quite match the message, and to communicate our ideas and the information we gather.
So what can a tester actually do about it? I do not have a clever framework for this, and I do not think you need one.
What I have are a few personal rules that, for me, are starting to help, and that might also help you, too:
- Don't open the AI for the first 15 minutes of a problem
- Explain the problem before you prompt it
- Pick one test case or scenario a week and diagnose it fully solo
- Write down what you used AI for at the end of the day, you will be surprised how much offloading happens
- Read the documentation before asking the AI for a shortcut
- Try to break the AI's answer before accepting it
- Pair with another tester instead of with an AI once a week and discuss your ideas
- Spend one hour a week reading the codebase without any help
Well, I hope you enjoyed the article, and remember, as tempting as it might be, don't stop thinking!
Link to the article mentioned above: https://pmc.ncbi.nlm.nih.gov/articles/PMC8358584/