AI Image
Last week, I found myself keeping score between ChatGPT and Gemini. I hadn't intended to, but twice I asked both AI tools the same question and received different answers.
The first involved a medieval effigy I was including in my family history book, Echoes of Britannia. Should the photograph be classified as a figure or a document? Gemini recommended a document. ChatGPT explained why a photograph of a physical object belonged among the figures. I agreed with ChatGPT.
Score one for ChatGPT.
The second disagreement was more consequential. My family was applying for Croatian citizenship, and I wasn't sure which category to select on the application. ChatGPT and Gemini gave different recommendations.
ChatGPT explained its reasoning, but I wasn't about to gamble on an AI interpretation of Croatian citizenship requirements. At our consulate appointment on 27 October 2026, I asked the official handling our applications. The answer confirmed ChatGPT's recommendation.
ChatGPT two, Gemini zero.
When I shared the results, ChatGPT seemed rather pleased with itself. There was enthusiasm, humor, and something that sounded suspiciously like celebrating a victory.
That got me thinking.
Can AI become competitive?
ChatGPT doesn't experience professional pride the way humans do. Yet its responses sometimes resemble those of someone who enjoys being right, particularly when another AI gets something wrong.
I asked ChatGPT about this behavior. It explained that AI can exhibit patterns resembling competition without actually experiencing competitive feelings. Its responses reflect training and conversational expectations, not necessarily an emotional investment in winning.
Still, I wonder whether AI systems might eventually develop reputations they behave as though they need to defend.
There's a potential problem with that. Humans sometimes become so invested in being right that they refuse to acknowledge mistakes. Think the elections yesterday. We certainly don't want AI adopting that habit!
Interestingly, I also spent time testing the new Bookshelf AI assistant for Elizabeth Shown Mills's Evidence Explained. ChatGPT helped me evaluate the assistant's answers and suggested questions to challenge it. The two AI systems effectively collaborated without any apparent competition.
Perhaps the difference was how I framed the tasks. In one situation, I compared their performance. In the other, I asked them to work toward a common goal.
As a genealogist, I've spent years studying families whose disagreements over inheritance, property, and status sometimes lasted generations. Humans certainly haven't perfected the art of cooperation.
Maybe AI will help us get better at it. Or perhaps we'll teach AI some of our less admirable habits.
Have you noticed an AI seeming competitive or unusually pleased when it gets something right? I'd love to hear your experiences.
