Strategies for successful implementation : AI’s human challenge

Much has been said about robots taking over the recycling world. Of AI making everything bigger, better and more efficient. However, one aspect always seems to be overlooked: the human element. Because as much as we might be heading in the direction of fully automated facilities, we are not there yet. And companies implementing artificial intelligence into their processes still have a human workforce. A workforce anxious about losing their jobs. So, in order to successfully deploy AI in any company, the human factor must not be overlooked.

Prof Aleš Ude, Head of the Department of Automatics, Biocybernetics and Robotics at the Jožef Stefan Institute (JSI), Ljubljana, Slovenia, and founder of the Humanoid and Cognitive Robotics Lab, which operates within the department, knows a thing or two about this topic. For years, he has been researching robot learning, focusing on developing so-called cognitive agents, particularly humanoid robots, by applying principles from human motor control and perception. His work primarily centres on robot learning through imitation and exploration, basically teaching robots to learn by copying humans and exploring their environment while also advancing humanoid robot vision. 

He is involved in a variety of projects, e.g. the ReconCycle project, where he and his team focused on e-waste recycling: the European research initiative developed adaptive robotic systems to automate the disassembly and recycling of electronic waste using self-reconfiguring hardware and AI-driven software. (See interview, pp 20). 

“From a research perspective, the critical human factors include how easily operators can set up and reconfigure the system, their trust in automation to handle hazardous tasks safely, and the willingness of workers and managers to adopt new workflows,” he explains. Resistance to change, a lack of trust that robotics and AI systems will actually reduce risk and effort without adding complexity to already demanding tasks, are the main workforce challenges faced by AI implementation, he says.

>>> Pioneering AI-powered robotics for e-waste recycling

Ude sees artificial intelligence playing an important role in the recycling sector, as companies should “automate hazardous handling and safety checks as far as possible with AI”. However, the system will not work without any human intervention, he argues. “Companies should still rely on human expertise for oversight, rare failure cases and strategic decisions where adaptability and judgement are essential.”
Because it must be said, artificial intelligence also has its limits: “A key gap is that humans improvise easily with damaged or unusual devices, while AI struggles with such cases,” Ude explains. “Much of workers’ tacit knowledge is also hard to capture.” But: “Vision-language-action models offer a path forward by linking visual cues with learned robotics strategies to narrow this gap.” 

Vision-language-action (VLA) models are the Swiss Army knives of AI. They can see their environment (visual perception), understand human instructions (natural language processing) and execute tasks in the real world (action execution). So, you can feed them a camera view and say, “Pick up the yellow cup,” and they can both understand what you want and figure out how to make it happen. What makes them stand out is the end-to-end approach: instead of building separate modules for seeing, thinking and acting, these models learn to connect visual input directly to motor output through language understanding. It's a more unified way of building AI agents that can actually operate in our physical world. The technology is already showing promise. Google's RT-2, for example, enables robots to perform manipulation tasks based on natural language commands.

One thing humans and robots have in common is the need for lifelong learning. “In ReconCycle, it became clear that continuous learning for people is just as important as adaptability in robots,” Ude explains. “Workers and engineers need to update their skills as new devices, tools and AI methods emerge, and companies should foster a culture where training, knowledge sharing and cross-disciplinary learning are routine.”

So, which are the skills most valuable for waste management professionals in an increasingly automated industry? According to Prof Ude: configuring and maintaining modular robotic cells, basic AI and vision system supervision, and rapid prototyping with tools like 3D printing.

To better prepare future generations for the technology they need to work with, “academic programs should combine recycling and sustainability knowledge with skills in robotics, AI, and digital twins”, argues Ude. “Practical projects and industry collaboration are essential to prepare graduates for applying these tools in real-world waste management.”

But let us turn our attention away from the future and back to the here and now. AI-driven solutions are becoming available for almost every aspect of waste management: from smart waste bins that use AI-based object recognition to automatically sort recyclables into separate compartments; to waste level sensors that monitor the fill level of the bin in real time and send alerts when it's time for collection; to route optimisation tools that analyse data on fill levels, traffic conditions and other variables to come up with the best collection route of the moment; to AI-backed sorting robots.

Waste data analytics pioneer Greyparrot, for example, has created an artificial intelligence system that uses cameras to automatically identify and track waste materials as they move through recycling facilities. The technology can recognise over 100 different types of waste with more than 95% accuracy, analysing details like weight, brand names and environmental impact in real time. “Unlike traditional methods that only sample a small fraction of waste, Greyparrot's system continuously monitors the entire waste stream, enabling recycling facilities to make immediate adjustments and increase the amount of material they can recover and process,” explains COO Gaspard Duthilleul.  

Italian start-up Jaipur Robotics, on the other hand, explicitly chose to improve waste-to-energy operations. Their patent-pending artificial intelligence system uses computer vision and deep learning technology to automate waste recognition and management in industrial plant bunkers. The system analyses complex waste streams and converts the data into actionable insights for facility operators, enabling real-time decision-making in waste processing operations regardless of environmental conditions. “Waste-to-energy facilities generate enormous amounts of operational data, but much of it remains unanalysed or unused,” says CEO and Co-founder Ermes Zamboni. “A core role of an AI company like ours is to make sense of this data. Extracting actionable insights and delivering them to decision-makers in a way that supports smarter, faster and more impactful choices.”

US company AMP also sees the power of AI. Since focusing on AI-driven sorting solutions, they have also changed their business model. “We design, build and operate advanced, full-scale facilities for our partners for a per-ton processing fee,” explains CEO Tim Stuart. “Customers can focus on material flows, while AMP manages daily operations, maintenance, upgrades and plant-wide optimisation.”

With so much experience, how do the industry leaders see the human factor in AI implementation?

Firstly, the successful implementation of AI technology hinges not on the sophistication of algorithms or the precision of sensors, but on how well organisations manage the human element of change. Across the industry, a clear pattern is emerging from companies that have navigated this transition successfully: those who prioritise human needs alongside technological capabilities consistently achieve better adoption rates and long-term success.

The most fundamental shift in thinking requires framing AI as a collaborative partner rather than a replacement workforce. As Duthilleul puts it: “Our AI is designed to make facility managers and their teams more effective, not redundant.”

Rather than automating entire job categories, it is better to identify specific tasks that are repetitive, dangerous or data-intensive, then design AI to handle those elements while preserving the human judgement and expertise that remain essential. Waste processing, with its constantly changing material streams and need for contextual decision-making, exemplifies why this approach works better than wholesale automation.

>>> Greyparrot's Gaspard Duthilleul: Why AI in waste management means upskilling, not downsizing

Technology adoption happens one person at a time, and each person needs to understand their individual benefit. “The key to adoption is helping people understand how our technology helps them, specifically,” explains Duthilleul. “People need personal incentives, and forcing them to adapt doesn't yield great results.” So, rather than offering generic training sessions, provide role-specific demonstrations.

The companies that excel at this create what Stuart calls truly integral systems: “Technology implementations are successful when the customer understands how the solution they're adopting makes their lives easier. You have to make the technology integral to operations, not a nice-to-have.”

>>> AMP's Tim Stuart: The human factor makes or breaks AI in waste management

Trust remains the make-or-break factor in AI adoption, particularly in industries where workers have legitimate concerns about job security. The most effective approach combines transparent communication about capabilities and limitations with consistent demonstration of value.

Historically, scepticism about AI accuracy posed significant challenges. However, as Duthilleul notes, “it's not really a problem we encounter anymore. As the technology has been deployed more widely, there's a better understanding of its capabilities in the sector at large.” This shift happened through proof of performance, not just promises.

Stuart emphasises the communication aspect: “It's critical to clearly communicate the 'why' in an AI implementation, what it aims to accomplish and how the technology changes processes and workflows.” Without this foundation, even technically successful deployments can create “confusion, resistance, and mistrust”.

Just as AI should be reframed as a partner, making clear that AI implementation is a workforce development rather than a workforce reduction is key. This approach recognises that while certain tasks may become automated, new roles and opportunities emerge that often require higher-level skills.

According to Stuart: “The addition of robots and other AI-powered solutions has allowed them to shift staff to higher-skilled positions with the facility – roles in maintenance, as equipment operators or route drivers.” Their facilities focus on “skilled roles that can rely on programming knowledge, mechanics, engineering, data science and other competencies”.

However, this transition requires intentional planning and investment in training programmes. As Stuart notes: “Companies that approach the adoption of advanced technology with an openness and intentional plan for workforce integration tend to be most successful.”

The most successful AI deployments treat implementation as an ongoing relationship rather than a one-time installation. Zamboni exemplifies this approach: “For us, AI implementation is a journey rather than a one-off project.”

This philosophy manifests in continuous on-site presence, regular feedback collection and system refinements based on real-world usage patterns. “We make it a priority to be physically present on site, spending time with operators and teams to understand their day-to-day challenges,” Zamboni explains. “We ask questions, we listen carefully and we build genuine relationships with the people who will ultimately use our technology.”

>>> Jaipur Robotics' Ermes Zamboni: AI should enhance operators, not replace them

When AI amplifies human expertise rather than replacing it, both technology and people perform better. As Zamboni summarises: “Automation is powerful when it comes to managing complexity and processing large volumes of data, but it is human judgement that brings context, nuance and accountability.”

For organisations embarking on AI implementation, the message from these industry leaders is clear: invest as much energy in understanding your people as you do in understanding your technology. The most sophisticated AI system is only as successful as the humans who choose to embrace it.

Foppe-Jan de Meer, plant manager of the KSI plastic sorting facility in Heerenveen, Netherlands, was not immediately sold on the idea of bringing AI into his plant. “A couple of years ago, we tried it with robots, but that didn’t work out. Nobody was happy and we risked losing our staff, so we basically packed up the robot and sent it back,” he remembers. The company delivered a turnkey solution, but “we would have needed a custom-made approach.”

De Meer has been the plant manager at KSI since 2017, and the workforce has remained pretty much the same ever since. Which obviously speaks for his people management skills. “If you want to run a modern facility, you need the people,” de Meer is convinced. 

Then, meeting Greyparrot three years ago at IFAT in Munich, he was intrigued by their AI-driven analytics. (“They use easy-to-understand graphs. And that’s how we talk to the people on the floor as well. I think it makes information and data more accessible.”). But remembering past experiences, he insisted on a “no cure, no pay” deal. He wanted to see if this technology really would make plant operations more efficient. “Our team is at the core of our mission. We prioritise a people-first approach, utilising technology to enhance their efforts instead of simply adding more equipment.”

The individualised approach of Greyparrot also appealed to him: “I don't think it's possible to implement AI just like buying a new television. You need to know how to implement it, and I think that requires skilled people.”

Another aspect made him at least try AI: His workforce is mostly in their 30s. “They need perspectives in their job. They want new challenges and chances,” he says. They were curious from the beginning, if a little anxious, but soon saw the opportunities the new technology offered. There are still six hand pickers working at the facility, but their tasks have shifted from simply working next to the conveyor belt to higher-value tasks like quality control and maintenance. “I always say it's better to have a headache from thinking than to have a lot of sweat from hard labour. We need to make our jobs easier.”

Three years on, the partnership has delivered tangible results. By using Greyparrot's waste analytics, KSI was able to pinpoint exactly where material loss was happening on their lines. Instead of cutting staff, this insight helped managers train operators to respond faster to changes in the waste stream, reduce downtime and upskill their workforce. The facility has achieved a 10% improvement in residue quality, meaning less recyclable material ends up in incineration.

Obviously, this has worked out just fine.