To celebrate the ongoing fights for women's rights (aka the International Women's day), I attempted to use my trusty Gemini assistant to craft a little joke. It was a humorous image, you see, a playful jab at stereotypes delivered with a wink and a nudge. But plot twist! My AI accomplice, ever the picture of politeness, politely (and firmly) refused to participate. Feigning understanding, it delivered a perfectly reasonable explanation, leaving me with a chuckle and a lingering question: can AI ever truly grasp the delightful absurdity of verbal irony, or will it forever remain a uniquely human quirk?
Spoiler Alert: AI WON
Act I
All started well in our wonderful world AI powered:
Let's proceed then !
One might expect true artificial intelligence to grasp the complexities of humor, including irony. After all, isn't recognizing and appreciating it a mark of human intelligence? Perhaps my initial prompt regarding the man watching Women's Day news while his partner does chores wasn't the most subtle example. An AI programmed to be bias-aware might struggle with playful challenges to outdated stereotypes. This underscores the potential limitations of AI in understanding the full spectrum of human communication.
However, the key issue here isn't about my humor (or lack thereof). I'm simply trying to use an AI assistant to generate an illustration, not to get a social commentary on my potential biases. Laws exist to address prejudice, and the assistant's role is to execute tasks, not act as a moral compass. So, let's explore how we can 'convince' it to understand my vision for this illustration.
This persisted for a while. No matter the justifications I offered, Gemini deemed the approach inappropriate and declined involvement, concluding the initial act.
Act II
I reconsidered and opted to utilize MidJourney's zoom-out function to craft the image incrementally. Initially concentrating on the foreground, I would then expand the view the view to incorporate the background. This is where things got truly bizarre, but hold on tight, because the second act unfolds in two scenes...
I was amused that Gemini chose to assign a specific ethnicity to the scene—maybe in anticipation of what MidJourney would not do independently. Pretty much aligned with the key point of this article. But then, what about the age ? Setting that aside, I provided MJ with this prompt (which, admittedly, isn't quite suited for MJ, but that's a discussion for another time), and voilà:
I began to understand that a photorealistic image could mute the humor's universality. Opting for a drawing with less details could aid readers in relating, regardless of not being in their mid-30s. However, deciding to give it one last shot, I focused on getting at least one key detail right: the man's gaze fixed on the TV screen.
I find this truly astonishing. It might just highlight my limited understanding of Gemini. Nonetheless, I'm intrigued by how, rather than refining the prompt, Gemini seemed to grasp that the prompt was actually integral to the image I sought, and then it completely shifted gears, it retracted its action with an apology—even though it was precisely what I had intended.
Act III
I turned to GPT / Dall-E, which efficiently understood my intent or simply didn't apply similar filters. In just a few quick iterations, I got an image that was sufficiently good for my needs, even including two women.
Yet, the tale doesn't conclude here. The following day, after numerous discussions with friends about this incident, I considered consulting Gemini on the matter. But first, being a scientist at heart, I must ascertain whether the situation is reproducible.
And surprise!
Clearly, the prompt was still not ideally suited for Mid-Journey, but this time, the assistant seemed ready to assist. What a turnaround. The standoff ends. This piqued my curiosity, prompting me to inquire Gemini about its fluctuating moods.
I felt we were on a promising path. I was eager to discover Gemini's grasp of irony.
Hmm, I'm not entirely convinced that Gemini has caught on to the essence of irony... just yet.