Early attempts to build specialized hardware to house artificial intelligence brains have been criticized as, well, a bit trash. But now there's an AI gadget in the works that's literally full of trash: Finnish startup Binit is applying the image-processing capabilities of large language models (LLMs) to tracking household trash.
AI that sorts the stuff we throw away in order to improve recycling efficiency at the municipal and commercial level has been garnering entrepreneurial attention for a while now (see startups like Greyparrot, TrashBot, and Glacier), but Binit founder Borut Grgic believes tracking household waste is unexplored territory.
“We're building the first household waste tracker,” he told TechCrunch, likening the upcoming AI gadget to a sleep tracker that can record your trash-throwing habits. “It's camera vision technology backed by a neural network. So we're using LLM to do general household waste recognition.”
Founded during the pandemic and with nearly $3 million in angel funding, the early-stage startup is developing AI hardware that can be used in kitchens (and looks cool) by attaching to cabinets or walls near where trash-related activities occur. The battery-powered gadget is equipped with cameras and other sensors that wake up when someone is nearby and can scan items before they go in the bin.
Grgic says it relies on integration with commercial LLMs (mainly OpenAI's GPT) to do image recognition. Binit then tracks what's being thrown away in the home and provides analytics, feedback and gamification via the app (such as a weekly trash score) to encourage users to throw away less.
Initially, the team tried to train their own AI model to recognize garbage, but the accuracy rate was low (about 40%). That's when they came up with the idea to use OpenAI's image recognition capabilities. Grgic claims that after integrating LLM, the accuracy rate for garbage recognition was nearly 98%.
Image credit: Binit
Binit's founder says he has “no idea” why it works so well. It's unclear whether OpenAI's training data contained a lot of junk images, or whether the sheer volume of data used in training simply allows it to recognize more things. “It's incredibly accurate,” he claims, suggesting that the high performance it achieved in tests with OpenAI's model could be because the items scanned were “common objects.”
“It recognises the brand and can also tell relatively accurately whether a coffee cup has a lining or not,” he continues, adding: “So essentially what we ask the user to do is pass an object in front of the camera, so the user has to hold the object steady in front of the camera for a second, and at that moment the camera takes images from all angles.”
The data of the garbage scanned by users is uploaded to the cloud, where Binit can analyze it and generate feedback for users. Basic analysis is free, but premium features will be introduced through subscriptions.
The startup also aims to become a data provider on what people are throwing away, which could be valuable information for organisations like packaging companies if it can scale up its use.
Still, an obvious criticism is: do we really need a high-tech device to tell us when we're throwing out too much plastic? Don't we all know what we consume and that we need to make an effort to produce less?
“It's a habit,” he argues, “and we think we're aware of it, but we don't necessarily act on it.”
“I know sleep is a good thing, but when I started wearing a sleep tracker and getting more sleep, it didn't really teach me anything I didn't already know.”
Binit also says that during testing in the US, users expressed interest in the waste transparency the product offers, resulting in around 40% less waste in mixed bins, so the company believes its transparency and gamification approach can help people change ingrained habits.
Binit wants the app to be a place where users can get both analytics and information to help them waste less — for the latter, it also plans to leverage LLM to get suggestions, and will personalize recommendations taking into account the user's location, Grgic said.
“The way it works is, for example, with packaging, for each package that you scan it creates a little card within the app that says, ‘This is something you threw away.’ [e.g. a plastic bottle]”…and in your area, these are alternatives you can consider to reduce your plastic intake,” he explains.
He also sees room for partnerships with food waste reduction influencers and others.
Grgic claims that another novelty of the product is that, in his words, it “counters unregulated consumption.” The startup is in tune with the growing awareness and action towards sustainability: the need to abandon disposable consumer culture and replace it with more conscious consumption, reuse and recycling in order to protect the environment for future generations.
“We are now [something]”I think people are starting to ask themselves: Do I really need to throw everything away? Or can I start thinking about repairs? [and reusing]? “
But surely Binit's use case is more than just a smartphone app, and Grgic argues that it depends on the situation: some households, for example, are happy to use their smartphone in the kitchen when their hands might get dirty while preparing food, while others see value in having a dedicated hands-free trash scanner, he says.
Notably, the company also plans to offer the scanning feature for free through the app, meaning both options will be available.
So far, the startup has tested its AI litter scanners in five U.S. cities (New York; Austin, Texas; San Francisco; Oakland; and Miami) and four European cities (Paris, Helsinki, Lisbon, and Grgic's hometown of Ljubljana, Slovakia).
He said they're working toward a commercial launch this fall, likely in the U.S. The target price for the AI hardware is around $199, which he described as the “sweet spot” for a smart home device.