Original title: Spook the Machine: Gamified Exploration of Human Imagination of Machine Fear
Levin Brinkmann, Hiromu Yakura, Sonia Nicoletti, Mar Canet Sola +8 · September 2026
Researchers wanted to know how people react when artificial intelligence displays fear, and how to design systems where humans and AI interact through emotion. They built a game called Spook the Machine. Players made images to try to scare AI bots that had specific phobias. The bots reacted either calmly or with intense emotion, such as begging for mercy. Other players could also see a gallery of images that successfully scared the bots. When machines showed high emotion, players stayed engaged longer after failing. They thought longer about their moves and learned faster from the gallery, though the creativity of their pictures did not change. Rewarding new, novel ideas kept players trying diverse concepts. Without rewards for novelty, people just copied previous winning images. How machines express emotion pulls players in, while the reward system guides how people explore ideas.
Imagine telling spooky stories around a campfire. Normally, you stick with the one that made everybody shiver last time. But if the listeners get bored with repeats, you have to invent new ones. In this paper, the AI machines are like those listeners: they react emotionally, and the gallery shows which stories have worked before, while the reward system either pushes you to repeat successful scares or to try something fresh.
This matters for anyone designing AI that interacts with people—from video game characters to digital assistants. It shows that making AI expressive does not automatically make users more creative, but it does make them more persistent. To guide what users actually create, you need to adjust the incentive structure.
The abstract does not state specific limitations, but the study is confined to a Halloween 2024 deployment with 832 participants, specific AI phobia personalities, and a gamified image generation task. Generalizability to other forms of human-AI interaction, longer-term effects, or different creative domains is not established. The measures of creative output and engagement are specific to the platform's metrics.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Spook the Machine: Gamified Exploration of Human Imagination of Machine Fear — Levin Brinkmann, Hiromu Yakura, Sonia Nicoletti et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.