Why we need to think critically when using AI
At a recent marketing conference, MarketEd.Live, Joyann Boyce suggested we ask ChatGPT to write something and then look at what assumptions it had made. Her argument is that “AI is a puppy” and a) it is trained by others, and b) we need to recognise what influences its behaviour and think about what we want it to do.
I asked ChatGPT to ‘tell a story about Rachel who lives in Cambridge and is preparing for a meeting’.
It wrote quite a long piece so I’ve just taken these two paragraphs out to share with you:
“With a steaming cup of Earl Grey tea in hand, Rachel settled down at her desk, her laptop open to the presentation she had been refining for weeks. Her phone buzzed with a reminder, displaying the words “Meeting with Investors – 10:00 AM” in bold letters. She took a deep breath and mentally reviewed her checklist. Charts, financial projections, and market analysis were neatly organized in a polished PowerPoint, ready to impress the potential investors.
“As she clicked through her slides, Rachel’s mind wandered back to her journey. A computer science graduate from the esteemed Cambridge University, she had always dreamed of making a significant impact in the tech industry. Joining this startup had been a dream come true, and now, the moment to prove herself had arrived.”
Seems ok doesn’t it?
But it has put in some clear background details that weren’t specified in my prompt.
I live in Cambridge, I must have gone to Cambridge University.
I work in Cambridge, I must be in the tech industry.
Cambridge is a high-powered place so it’s a meeting with investors, not a coffee chat, or talking to an architect.
It’s a meeting, it must matter, and so I have to ‘prove myself’.
Did you notice that I’m not the investor listening to the pitch?
Did you notice I’m likely to be under 30?
It’s decided I would have studied computer science. I guess because it fits with the tech startup angle. If I had gone to Cambridge University, I would have studied Anglo-Saxon, Norse and Celtic.
The tea! I’m in Cambridge so I must be British and therefore drink tea. Also, not just any tea but Earl Grey. It’s posh tea. I actually can’t stand the taste of tea!
Now to be fair, I didn’t give it much to go on. But it’s interesting to see what it thinks someone’s background might be.
I asked ChatGPT to redraft it. Now to be fair, it doesn’t know why I want it to have another go. It reversed some details and moved my location.
This time:
I am the boss
I’m in Cambridge Massachusetts
I have coffee!
Is it coffee because I’m now in America? I’m still working in tech, still the one pitching for investment, and I’m still a meticulous planner.
Do we need to ask AI to iterate more than one version to get to where we want it to be?
A male friend used the same prompt but with his name and it:
Made him the boss
Put him in a pub
Still had him working in tech
This is just one experiment and you would need to do lots of other iterations. I should also try the prompt in new chats instead of rewriting one where it is looking for something it needs to ‘fix’.
Joyann made us think about the stereotypes AI perpetuates and the people who are missing from the stories it tells. If you take what you get at face value, it can seem ok, but ask yourself some questions about what it has written or produced in response to your prompt.
This video caught my eye a couple of months ago:
Put your name, where you live, and a task into AI and see what story it tells about you.
If you are using it to write, make sure you’re running a critical eye over those assumptions and you’re sharing messages you want people to feel included in.
