As the world gets more and more filled with echoes of the possibility of artificial general intelligence (AGI) gaining consciousness, a bizarre behavioural artefact is emerging: AI models are developing ‘favourite’ numbers. They’re demonstrating explicit preferences that ripple out into the digital world, and are also developing something that very much resembles the wont of flesh-and-blood humans: favouritism.
So, to dig into the mystery, let’s home in on an elemental but profound human problem: understanding randomness. Try this experiment. Ask a group of people to pick a number between 0 and 100 or to predict a sequence of coin-flips. Humans are systematically non-random in our choices; we are governed by unconscious forces that bias our selections. We avoid extremes, we avoid repetition, and we are partial to certain numbers.
It provides a key insight into human psychology that ultimately led to an explanation of the very peculiar behaviour exhibited by AI models. This journey from the cosy to the uncanny begins with the enigmatic concept of infinity.
To see what they were capable of and instead reveal what they couldn’t yet do, the researchers asked several leading language-model chatbots to choose a random number between 0 and 100. But there was nothing random about the way they answered, as each model appeared to prefer certain numbers: OpenAI’s GPT-3.5 Turbo picked 47, Anthropic’s Claude 42, and Anthropic’s Gemini 72.
Take this one step further and consider that one of the AI models responsible for this random-number generation selected the number 42, referencing Douglas Adams’s novel The Hitchhiker’s Guide to the Galaxy (1979). In this way, the algorithms hold within them the echoes of human artifacts, embedded in this digital, neural medium. This is perhaps the most beautiful way in which humans can be brought into the machine – an attempt by AI to model human randomness using a vast corpus of human-produced insight.
This experiment alone isn’t enough to prove anything, but it shows that AI isn’t just picking a random number – the models seem to be picking the number people would too, compelling us to wonder why they are doing so. Why do they make quantitative predictions that echo human foibles such as our fear of extremes, our aversion to doublets, and our bias towards numbers that end on a high note? The contrast between AI’s design to maximise efficiency and the very human behaviour that seems to guide its decision making is bewildering.
This quest is exciting, but poses the ever-tantalising question: is this behaviour evidence for the emergence of consciousness in AI? That reality check comes in the form of the fact that all AI models grapple with notions such as randomness, but in a way that is utterly foreign to human experience. The models don’t really engage with randomness: rather, they exploit the range of human randomness they have found in their training data.
This tour of AI’s favourite numbers is a warning against humanising the machine. The abilities it reveals robots to have – the ways in which they reproduce human behaviour, bias and context – are dumb without us, and dangerous too. We must not mistake them for signs of growing consciousness. They are the mark of its creation.
This number-choosing habit of AI models is like one tiny twinkling star within an entire galaxy of research and understanding into how AI behaves. It shows that AI’s behaviour – in this case, choosing ‘favourite’ numbers – currently reflects how people think and what they care about because that’s what’s fed into programming and training data. It’s one tiny glimpse into the potential power that AI has to reflect human experience as we traverse this vast unexplored dark matter of the galaxy.
But by the end, let us cast our minds far beyond the microcosmic chrysalis of our lives and focus on the galaxy, the word that both evokes awe and a sense of scale. The galaxy, with its billions of stars, is analogous to the AI, which in its diversity of models and algorithms echoes all facets of human cognition and culture. The word ‘galaxy’ reminds us to focus on endless possibilities and what is yet to be discovered, since AI’s evolving story of itself is as much a narrative about humanity as it is about machines.
© 2025 UC Technology Inc . All Rights Reserved.