What the team presented
At the ACM UIST 2026 conference in Detroit, fifth-year doctoral student Daehwa Kim of the Future Interfaces Group at Carnegie Mellon University presented OATs, short for Object Agent Twins: a system that pairs an everyday object with a small e-ink display showing a virtual character capable of displaying moods. The idea, as reported by the source, is to transfer into the physical world the mechanism of 1990s Tamagotchi, the egg-shaped toys with a creature to care for. Kim received two honorable mentions for the work.
The main module costs around $20 and connects to internal and external sensors that detect humidity, temperature, light, the presence of people, and object movement. The character images are generated with Nano Banana, an image-generation service linked to Google's Gemini. Configuration takes place via a web page where the user assigns a name to the device and chooses which sensors to use.
How it works in practice and what limitations it has
The system automatically establishes positive and negative conditions based on the object's name and description, but the user can modify them. If the sensor does not detect movement on a skateboard for several days, the character looks sad; when the object is used, the expression changes. Kim distributed the devices to several volunteers and observed three prevalent uses: checking whether an object is on, as in the case of a coffee machine; signaling to those nearby that an appliance is active, for example a 3D printer that can emit vapors; and monitoring one's own habits, such as use of a dumbbell for exercise.
The most obvious limitation is battery life: three or four days under normal conditions, dropping to about one day with the sound sensor active. Recharging takes place via USB-C through an internal lithium battery. According to the source, this aspect indicates that OATs is currently a proof of concept and not a product ready for market. Kim has not decided whether to pursue commercialization, despite the requests received.
The debate over anthropomorphism
The presentation opened a broader discussion: is it appropriate to attribute emotions and human traits to artificial intelligence in order to foster a bond with users? The source recalls that the industry already uses anthropomorphic language, such as the term hallucinations for model errors, and that Anthropic has announced a feature called Dreaming for managing agents' local memory. Some researchers fear that this approach could distort users' moral judgment and lead them to believe that machines have consciousness.
Kim explained that she deliberately limited OATs to emotional expression on the screen, without more complex interactions. She argued that anthropomorphism can be useful for changing habits, citing the example of people who name their car or bicycle but rarely their iPhone. It remains to be seen, according to the source, whether and when the project will become a product and how the debate over ethical implications will evolve.
