Carnegie Mellon’s OATs give household objects Tamagotchi-style AI faces that sulk when ignored

carnegie mellons oats give household objects tamagotchi style ai faces that sulk when ignored A houseplant has no way to say it needs water. A $20 module attached to its pot does: it shows a small sad face on an E Ink screen until someone fetches the watering can, and the face brightens once the soil gets wet.Image Source: WiredImage Source: Wired

A houseplant has no way to say it needs water. A $20 module attached to its pot does: it shows a small sad face on an E Ink screen until someone fetches the watering can, and the face brightens once the soil gets wet.

The modules are called OATs, which stands for Object Agent Twins. Each one is a micro-agent AI avatar on a tiny E Ink display, designed to be paired with an everyday household object. They were created by Daehwa Kim, a fifth-year PhD student in Carnegie Mellon’s Future Interfaces Group.

In early November, Kim showed the project at the ACM Symposium on User Interface Software and Technology in Detroit, where academics and industry experts gather to explore human-computer interfaces. She came away with two Honorable Mention awards.

A Tamagotchi for the coffee machine

Anyone who grew up in the ’90s will spot the inspiration. Tamagotchi were egg-shaped toys containing digital pets that demanded real-time care and attention. Kim wanted to bring that feeling to physical things people already have at home.

“We bring this joy of caring from virtual to physical objects,” Kim said.

Her argument begins with screens. According to Kim, people look at mobile devices all day, and AI agents, including those built to complete tasks for you, keep dragging attention back to the smartphone. OATs put those interactions out in the room instead.

The use cases are straightforward. Attach one to a skateboard you never get around to riding, and after a few days without detected motion, the avatar turns sad and lonely. Hop on the board and it flips to happy, goofy expressions. “These kinds of inanimate objects can be more expressive,” Kim said.

Three patterns from testers

Kim said she handed OATs to many people and noticed three kinds of use.

The first responds to the question “Am I being used?” A coffee machine is one example: its avatar gets excited when the grinder starts on the beans. The second is alerting people who are near an object. Because 3D printers give off toxic fumes while running, an OAT on a printer can let people know it’s in use. The third is habit tracking. One tester put an OAT on a dumbbell to log how often they lifted, and kept exercising regularly so the AI pet wouldn’t be let down.

The demo at the symposium attracted a crowd, and Kim made it into a contest for the most interesting object pairings. Her website hosts a gallery of sprites people created, including an animated conference badge and a panda avatar for an Owala water bottle.

Low cost, short battery life

Each OAT contains an E Ink display connected to internal and external sensors that can measure moisture, temperature, humidity, light, occupancy and motion. The main module is priced at around $20.

Users set it up through a web portal, giving the OAT a name and choosing the sensors that suit the object it will sit on. The sprites are generated by Google Gemini’s Nano Banana image-generation service, which creates a range of emotions for different states depending on the sensor data.

The web app also drafts “Positive” and “Negative” behaviors based on the object’s name and description. For a plant that prefers dry conditions, positive triggers could include “consistent air temperature readings” and “low humidity levels.” Owners can change these to fit their particular plant.

Power is the main drawback. On its lithium-ion polymer battery, an OAT runs for about three to four days, dropping to a single day when used with the sound level sensor. Charging is via USB-C. That limitation is a large reason the project remains a proof of concept and isn’t a product.

Kim has been urged to sell OATs kits online, but she hasn’t decided whether to build a business around it.

The trouble with cute AI

Giving AI a physical form has been tried before. Companies keep hunting for “the next smartphone,” from the failed Humane Ai Pin to OpenAI’s upcoming hardware developed with former Apple designer Jony Ive to Meta’s Muse Charm, a pocket-sized device running Meta’s Muse AI agent that is expected to launch in December. Increasingly, these companies are relying on cuteness to win people over.

The demo also prompted a lengthy debate about anthropomorphizing AI, and whether it’s appropriate to give these sprites human-like traits and emotions so people feel attached to them.

The AI industry does this all the time. Chatbots “hallucinate.” This summer, Anthropic announced a “dreaming” feature in which its AI agent tidied its local memory as a background process. Human language like this can skew moral judgments about AI and push people to trust the tools more than they should.

Kim said this is the reason OATs are restricted to expressive emotions on an E Ink display and offer no other interactions.

“Anthropomorphism can be very useful in certain ways, like helping people to change their behaviors,” Kim said. “People name their car, and they name their bike. I rarely see people who name their iPhone. I feel like in the future of computing, we will have more of a relationship with AI, but we should definitely be careful not to abuse this.”