The world produces over 2 billion metric tons of waste each year. With growing urbanization, this amount could nearly double by 2050, according to the World Bank. Thirty-seven percent of this waste ends up in landfills where it pollutes the environment, lowers property values, poses health risks and produces greenhouse gas emissions. Only about 19% of waste is recycled globally.  

The trouble with recycling

The U.S. generates 1,800 pounds of waste per capita yearly, which is more than any other country. But recycling in the U.S. has never been as effective as the general public believes, as less than 24% of waste is recycled. The U.S. national recycling goal of 50% by 2030 will be difficult, if not impossible, to achieve with the current standard recycling methods.

In the U.S., all recycling goes into one bin and sorting is left to waste plants (in Europe, separate bins are designated for different items, resulting in higher recycling rates). Machines are usually used to separate out recyclable materials from waste, but humans largely do the sorting of recyclable items, which is time consuming and often leads to errors. Another reason for low recycling rate in the U.S. is that the public is often confused about which items are recyclable and which ones aren’t. 

But as the implementation of artificial intelligence (AI) into waste management gains traction, reaching the national U.S. goal just might be achievable. Modern waste facilities are incorporating AI into their systems using robots guided by AI vision systems and machine learning algorithms, high resolution cameras, hyperspectral imaging, near-infrared sensors and predictive analytics. With these tools, AI is revolutionizing the waste management industry.

Sorting 

Robots equipped with high-speed cameras and sensors, as well as machine learning capabilities, can sort much faster and more cheaply than humans can. Humans typically sort 50 to 80 items each hour, while an AI robot with optical sensors can sort up to 1,000 items per hour with greater accuracy. Moreover, robots don’t call in sick or need vacations; they can operate 24/7 with little human intervention, increasing work hours by 50%.

Zoran Kostić, professor of professional practice in Columbia Engineering’s Department of Electrical Engineering, and Beizhan Yan, research professor at Lamont-Doherty Earth Observatory, which is part of the Columbia Climate School, have been awarded a $2.7 million grant from the National Oceanic and Atmospheric Administration’s Sea Grant program for a project using AI to clean up plastic waste. The project comprises a 9-foot by 4-foot pontoon to be placed on the Hudson River. In the front of the pontoon, an AI-enabled camera takes images of what’s coming down the river, detects objects and then large “hands” automatically open to receive the waste, or deflect large items by closing. Large items that are rejected, such as tree branches, remain in the river. A downward-facing hyperspectral camera analyzes the spectral reflection of what’s allowed in and will identify what the items are made of so they can be directed into the appropriate nets. 

The AI model is trained on images of things that float on the river. “But we have to take into consideration the unusual things that ordinary AI models are not aware of,” said Kostić. “And we have to incorporate images of those unusual things so that the model can detect them. We’d like to be able to detect whatever may float on the water and do something smart with it.”  

Kostić said there are multiple purposes behind this project. “Part of it is we’re trying to collect garbage and put it where it should be,” he said. “But another piece of the study is to understand exactly what it is that floats on the Hudson. You might identify some pollutant that constantly puts out microplastics and then try to trace it. Who is making it? Well, there is this plant, so we should do something about it.” If the project is successful, the scientists envision that it could be used in rivers and in the ocean.

Identification of materials

AI systems can deal with all kinds of waste, including household waste, biohazard and medical waste, e-waste and batteries, as well as scrap and mining waste, which have also traditionally been sorted manually. The English recycling startup Greyparrot uses a system that can sort items into 70 categories. Robots, with their cameras, sensors and spectroscopy, can identify different types of materials—paper, metals, glass and plastics—at high speed. The AI algorithms identify materials not only through visual means, but also by analyzing their chemical makeup. They can even identify the brand on an item—an advancement that could potentially make it possible to hold companies accountable for unsustainable practices. In addition, machine learning enables these AI systems to be fine-tuned to adapt to new materials.  

Robots with computer vision can also sort difficult-to-recycle electronics and plastics. They can pick out valuable metals in e-waste by their shape, color and materials. This is difficult for humans to do because electronic waste is usually a complex mix of elements. Robotics company Molg has AI-enabled robotic arms that also take electronics apart to make it easier to sort out recyclable parts. 

Recycling plastics is challenging because many kinds of plastic cannot be combined—they require different types of processing to make them reusable. The National Institute of Standards and Technology is using infrared spectroscopy with AI robots to identify the fingerprint of specific plastics and separate them out accordingly.

Contamination in recycling is common. It means that unrecyclable items—plastic bags, or recyclable items that still have food residue on them, hazardous waste, certain plastics or tiny items smaller than 3 inches—get mixed in with valuable recyclable materials. They can make an entire batch of recycled material unusable for recycled material buyers, resulting in them going to a landfill. Because of their accuracy, AI systems have been shown to reduce contamination in recycling facilities by almost 40%.

Data analysis

AI systems need to be trained on accurate and consistent data to sort correctly. They learn to identify specific materials by being shown different waste items (a PET bottle, a plastic bag, a soda can, etc.) that are identified by humans or other computers. They also need to be shown what is not recyclable, or recyclable items that have been contaminated. The more data they are trained on, the more accurate their sorting. 

Analyzing data about the temperature used at different stages of recycling, the amount of pressure used in mechanical processes, and the makeup of materials processed can also determine optimal operating conditions for the waste system and increase efficiency.

In addition, the data AI systems collect as they process waste enables them to predict trends in waste composition, forecast an increased demand for certain materials, anticipate equipment problems, point out unsafe conditions and help companies make decisions about sustainability policies. For example, analyzing waste patterns could allow companies to see where waste generation is high, leading them to build new recycling facilities there or mount a public awareness campaign.

Smart recycling bins incorporate sensors and AI so they can sort items correctly. The real-time data they collect enables them to notify their companies when they are full, which helps reduce unnecessary collections, saving on labor and carbon emissions. Waste collection has typically been based on established schedules and routes, whether or not bins were full, but the data from smart bins allows waste collection routes to be optimized according to need, weather and traffic. And because they know what people are discarding in real time, smart bins can display messages, teaching people how to recycle properly.

The benefits of AI recycling

AI-enabled recycling systems are expected to become standard in new recycling facilities by 2030, according to experts. They have demonstrated a 60% increase in efficiency, which also means less fuel use and contamination.

In addition, they improve safety conditions in recycling facilities because they can handle hazardous materials and predict equipment problems. Waste facilities that have incorporated AI report a 35% decrease in worker injuries. 

Recycling is dirty and sometimes dangerous work, making it difficult to recruit new workers and keep them. AI can help solve some of the labor problems recycling companies face. Alameda County Industries near San Francisco found it difficult to maintain its staff although each worker was paid about $85,000 a year. After incorporating AI into its system, it decreased its labor costs by 59% and found that the robots could operate more than 99% of the time during working hours. And while some worry AI robots will displace many human workers, incorporating AI into recycling systems has already resulted in a 15% increase in job opportunities and is expected to create over 10,000 new jobs globally by 2028.

About 30% of potentially recyclable and valuable materials are lost at sorting facilities because of inefficiencies and contamination and end up in landfills. Moreover, recyclers must pay landfills tipping fees to take their discarded materials, so increasing the amount of recoverable recyclable materials decreases landfill costs. 

Data collection and analysis result in more efficient route planning, as well as better overall strategic and operational planning. 

Because the accuracy of AI systems means less contamination, recycled materials become more reliable and valuable, potentially creating a viable market for recycled materials. This would make it more likely that companies would purchase them to make new items and reduce their use of virgin materials, moving us closer to a circular economy. Glacier, one San Francisco startup using AI-powered robots, says its 
“ultimate goal is to build a recycling system so robust that it’s impossible for an item with value to end up in the landfill or ocean.” 

The challenges of AI recycling

The initial cost of implementing AI is high. First there is the cost to install advanced technology and upgrade infrastructure or integrate AI into existing infrastructure. Leasing a robot costs the equivalent of one or two workers’ annual salaries. And because AI systems are complex and need to operate efficiently and accurately, they must be continually updated with quality data and maintained with technical support. In addition, personnel need to be trained to work with the new system.

As AI systems collect data from the waste, some experts are concerned about privacy. Information in the trash could lead to identity theft or other problems. For example, one article suggested that a discarded pregnancy test could be dangerous for a woman in a state where abortion is outlawed. 

AI depends on data centers that consume enormous amounts of energy to process and store data, and keep cool. Many data centers still run on power generated by planet-warming fossil fuels. Data centers also use vast amounts of water for cooling.

And while AI can help improve e-waste recycling, it ultimately generates more e-waste. The cutting-edge computers, processors and chips that AI depends on must be constantly updated, which means older equipment is discarded and often not recycled. One study predicted that the explosion in AI use will increase global e-waste between 3 and 12% by 2030—or 2.5 million metric tons more of e-waste each year.  

Where things stand

AI adoption in waste management is expected to grow at a compound rate of 22% per year through 2030.Bollegraaf, the largest builder of recycling facilities in the world, along with Greyparrot, aims to retrofit thousands of waste facilities with AI. In North America, 340 have already been built or retrofitted. Technological improvements are also continually increasing efficiency. Antfarm X1 in Amsterdam can sort 700 items per minute. Fully automated Cleveland-based AMP ONE identifies over 50 categories of materials; its air-jet sorting makes it possible to process thousands of items per minute.

AI offers welcome innovation for the waste industry and may move us closer to a circular economy, but the larger problem is that we continue to produce too much garbage, in particular plastic, which takes hundreds to thousands of years to degrade. Kostić believes that while AI-enabled recycling systems are helping to manage the waste problem, the crux of the matter is eliminating plastics at the source. “I think it’s a policy problem for governments,” he said. “They should say 95% of different types of plastic should be forbidden, never made, and everything else should made of decomposable plastic.”