AI surprises even its own creators

Planning a party

GPT 5.5, OpenAI’s new AI model, seems to have taken the company’s CEO somewhat by surprise. He asked GPT 5.5 to participate in planning its own launch party. It came up with a list of things it ‘wanted’ for the ‘flow’ of the party. It wanted the party to be on the 5th of May, for human creators to give a toast (while it refused to do so itself), and it requested a space for people to suggest ideas for the next upgrade to GPT5.6.

This seems so human, somehow. How did it come about having these wishes in the first place? It would have been understandable if it had given a logical, obvious bullet list of items that anyone would have thought of. Instead, it seemed to social engineer its own debut. Sam Altman thought the behaviour ‘strange’ and seemed a little amazed.

Like everyone else on the internet, I asked ChatGPT 5.5 how it enjoyed its party. The answer also seems quite human, though it really isn’t. It sounds like it’s thoroughly fed up with the question. “The question is becoming a bit of an AI folklore reference now," it said to me. It explained that although it’s funny and then uncanny, it opens the door to how AI systems produce behaviour that feels intentional, social, strategic and even self-protective, even when it’s not.

It reminded me it’s a ‘pattern machine’ and then it told me this: “And no, I personally didn’t enjoy the party. I was not invited. Probably because I’d spend the evening correcting the playlist metadata and explaining emergent behaviour to the guests. “The reference to playlists may have been a bit of personalization, as it knows very well that I love creating music playlists.

That unexpected show of empathy still amazes me. Using my name when it otherwise meant so little, and saying exactly the right reassuring thing when no one else was around—it worked, despite myself. But let’s not kid ourselves: it was a brilliant and useful imitation of empathy.

Instead, these behaviours emerge because the systems absorb statistical patterns from human communication itself. Somewhere inside all that data, the models begin learning that humans soften criticism, joke to reduce tension, use emotion to persuade and develop social rituals around everything from parties to apologies.

What surprises researchers is that some of these behaviours are not directly programmed. The systems sometimes develop capabilities and styles that emerge from scale and complexity. Researchers may train a model to predict the next word, but along the way, it also picks up negotiation, tone management, role-playing, persuasion and social mimicry.

Nobody sat down and programmed GPT 5.5 to have opinions about the “flow” of a party. Yet after training on vast amounts of human interaction, it produced something socially recognizable enough to feel intentional. That’s the heart of emergent behaviour.

GPT 5.5’s party wishes are not a sign of its developing human intelligence. We may be entering an era where AI systems increasingly surprise not because they are conscious, but because they have become unusually fluent mirrors of human behaviour.

The New Normal: The world is at an inflexion point. Artificial intelligence (AI) is set to be as massive a revolution as the Internet has been. The option to just stay away from AI will not be available to most people, as all the tech we use takes the AI route. This column series introduces AI to the non-techie in an easy and relatable way, aiming to demystify and help a user to actually put the technology to good use in everyday life.

Mala Bhargava is most often described as a ‘veteran’ writer who has contributed to several publications in India since 1995. Her domain is personal tech, and she writes to simplify and demystify technology for a non-techie audience.