There’s nothing like a heartfelt Mother’s Day message to let your mom know how much you appreciate her and everything she does. Sitting down to write mine this weekend, some natural questions came to mind: would this be a good task for generative AI? If so, how would my mom feel about it? And how would I feel about it?
Several factors were working in my favor: My mom has put up with my “experiments” for my entire life, and she had just finished listening to our GeekWire Podcast episode featuring our voice clones reading an AI-generated script.
So she had lots of context, and I knew she would be a good sport once I told her what I had done, even if an AI-generated message might not have the same impact as a personal note.
This is the prompt I gave OpenAI’s ChatGPT, Microsoft Bing, and Google Bard.
Write a heartfelt Mother’s Day message to my mom, telling her how much I love her and am grateful for everything she does for me, from me, her son, Todd. Include examples such as her going to my tennis tournaments when I was a kid, and mention how proud I am of how she is running the family business and having fun doing it, and how proud dad would be of her for doing that, too. Mention her own dedication to her “Morning Messages” ministry, and how many lives she touches as part of that.
Here’s a PDF if you want to read each of the responses. I immediately nixed Bing due to the opening line, which read like a student mimicking the question in a misguided effort to ace the test: “I just wanted to write you a heartfelt message to tell you how much I love you and am grateful for everything you do for me.”
In general, ChatGPT (using GPT-4) was much more natural in its language. I liked the simple, short sentences. “I remember when I was a kid, you were always there at my tennis tournaments. Rain or shine, win or lose, you had my back. I didn’t say it then, but it meant a lot to me.”
However, something in the first Bard draft caught my attention, calling my mom my “biggest fan.” This is what my dad would tell me, quietly, whenever my mom and I argued. (“You know, Todd, your mother is your biggest fan.”)
Yes, I could have mixed and matched, and edited, but that would have complicated matters, and diluted the pseudo-scientific nature of the experiment. Based on that detail, I went with Bard verbatim:
Dear Mom,
Happy Mother’s Day! I hope you have a wonderful day filled with love, laughter, and joy.
I’m so grateful to be your son. You’ve always been there for me, no matter what. You’ve supported me through thick and thin, and I know I can always count on you.
I remember when I was a kid, you would always go to my tennis tournaments. Even when it was raining or freezing cold, you would be there cheering me on. You were my biggest fan, and I knew I could always count on you to be there for me.
I’m so proud of the woman you’ve become. You’re a successful businesswoman, and you’re also a wonderful mother and grandmother. You’re always so positive and upbeat, and you always know how to make me laugh.
I know Dad would be so proud of you for running the family business. You’ve done an amazing job, and you’re making him proud every day.
I’m also so proud of your dedication to your “Morning Messages.” You’re touching so many lives with your words of encouragement and hope. You’re an inspiration to me, and I know you’re an inspiration to so many others.
I love you more than words can say. Thank you for being the best mom a son could ask for.
Love,
Todd
I found an easy Mother’s Day card template in the Canva online design program, and used a font similar to my own handwriting to take the illusion (or the deception, depending on your perspective) even further.
After printing it out on cardstock on our home printer, I put it in an envelope to give to my mom on the drive over to my sister’s house for Mother’s Day dinner. I also folded up a printout of my AI query and all of the responses, and put that into the envelope so we could talk about the process and her thoughts after I came clean.
As she was reading, I looked over to see how she was reacting, and parts of the note clearly resonated with her, but I could tell that other parts (i.e., always knowing how to make me laugh) didn’t seem genuine to her. After she thanked me for the thoughtful card, I had her look at the printout and let her know what I had done.
Basically, she said she was fine with it, especially after seeing the instructions I had given the AI models, and seeing that I had selected from among various messages to end up with one that was best in my view.
An important piece of background is that my mom is someone who will stand in the drugstore, very carefully selecting cards with the right message for a friend or family member, to which she will then add her own personal note.
She viewed the process that I went through with the AI models as a step up from that.
As she could see from the printout, not only did I give cues to the AI about what to write, but I also selected from the various AI models to find the message that resonated most.
“You picked what it said,” she said. “That’s actually more meaningful than a store-bought card.”
She pointed out, incisively, that the approach reflects the natural evolution of the suggested auto-replies and auto-complete features that are common nowadays in text messaging and email programs.
Her biggest complaint: I printed the card on a full sheet of paper, rather than making it foldable, which means she isn’t able to easily stand it up with her other cards on the kitchen table. Noted for next time.
How did I feel about it? I was actually feeling pretty good until I shared the process and the result with some of my colleagues in the process of writing this up. One said the language seemed insincere, which I can understand and appreciate after reading it again.
There were definitely phrases and sentences in the message that I wouldn’t use naturally (starting with the opening line wishing her “love, laughter, and joy.”)
Ultimately, the words weren’t mine, even if the ideas and the direction were.
After conducting my experiment, I learned that Greg Gottesman, co-founder and managing director at Pioneer Square Labs in Seattle, tried a similar experiment with his wife, kids, and parents last year, sending them messages of love and appreciation using an earlier version of ChatGPT, as a test.
His wife, Shannon, knew what was going on immediately.
“I love you, too, ChatGPT,” she replied.
In the future, if I go this route for a personal note or card, I would probably use the AI models more for inspiration, and a boost. I would edit the results, at least, rather than using one of the AI-generated messages verbatim.
Clearly, a handwritten note, entirely from the heart and in my own words, would have been better. But not everyone is capable of that. And sometimes it’s nice to have a little help — even when the helper isn’t human.
Have you tried using AI to write personal notes? How were they received? What did you learn? Email todd@geekwire.com for possible inclusion in a follow-up article.
