Initial efforts in crafting hardware specifically for AI capabilities have faced their fair share of criticism for falling short. However, Finnish startup Binit is making strides in an interesting direction by leveraging large language models’ (LLMs) prowess in image processing to monitor residential waste.
The concept of applying AI to classify and sort waste to enhance recycling processes at both local and commercial scales has captured the imagination of innovators for some time now (including startups like Greyparrot, TrashBot, and Glacier). Yet, according to Binit’s founder, Borut Grgic, the domain of tracking domestic garbage remains largely unexplored.
“We’re venturing into creating the first domestic waste tracking system,” Grgic shared with TechCrunch, drawing an analogy between his AI-driven invention and a sleep tracker, but focused on analyzing garbage disposal behavior. “It’s based on camera vision supported by artificial neural networks, diving into LLMs to identify everyday waste items.”
This burgeoning venture, birthed amid the pandemic and bolstered by close to $3 million in angel investment, aims to design AI-enhanced hardware that not only serves a functional purpose but also adds aesthetic value to kitchen spaces—preferably positioned on a cabinet or against a wall close to the garbage can. This battery-operated device includes cameras and sensors to detect proximity, activating to scan items as they’re being disposed of.
Grgic discusses their reliance on interoperating with established LLMs, mainly OpenAI’s GPT, for image recognition tasks. The data gathered on discarded items then enables Binit to deliver analytics, feedback, and a ‘trash score’ through an app, incentivizing users to minimize waste output.
Initially, the team attempted to train its own AI for waste identification but struggled with accuracy (around 40%). Switching to OpenAI’s advanced image recognition improved their success rate dramatically, achieving near 98% precision, according to Grgic.
Grgic is astounded by the device’s high level of accuracy, which may partly stem from OpenAI’s extensive training dataset’s capacity to recognize a vast range of items, making it especially adept at identifying common household objects.
The system requires users to present the object to be disposed of in front of the camera briefly, allowing it to capture the item from multiple angles for accurate recognition— including nuanced details like brand-specific features on a coffee cup.
User data on disposed items is collected and analyzed in the cloud, where Binit offers insights and suggests behaviors for reducing waste. While basic analytics are free, the company plans to introduce advanced features through a subscription model.
Binit also aims to be a valuable data resource on consumer disposal habits, which might interest packaging companies, provided the startup can achieve significant user adoption.
The necessity of high-tech gadgets for waste awareness might seem superfluous—an echo of common sentiments on consumer habits surrounding waste production. Grgic, however, believes that habit formation is key, drawing parallels between their product and the psychological effects observed with sleep trackers.
Preliminary testing in the U.S. revealed about a 40% reduction in non-recyclable waste output among users, underscoring Binit’s belief in data transparency and gamification as effective motivators for habit change.
Binit’s app intends not just to report but to actively guide users in reducing waste, utilizing LLMs to provide localized suggestions tailored to the user’s disposal patterns and available alternatives.
Furthermore, Binit envisions collaborations that promote sustainable habits, positioning the product in opposition to unchecked consumerism and championing a shift towards more sustainable, conscientious consumption habits and waste management.
Grgic views the current cultural shift as an opportunity to rethink our relationship with disposables, advocating for a culture of repair and reuse over discard.
While a smartphone app could serve a similar function, Binit argues there’s tangible value in a dedicated, hands-free device designed for kitchen use, though they also plan to offer a scanning feature via their app at no cost, catering to diverse household preferences.
Binit has conducted pilot testing in a number of U.S. and European cities and is gearing up for a commercial launch in the U.S. this fall, aiming for a retail price that makes it an attractive addition to the smart home ecosystem.
This report was corrected to state that Ljubljana is in Slovenia, not Slovakia. We apologize for the mistake.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
