How AI is being harnessed to make packaging more sustainable

In a rush? Click here to read a short summary of this report

Introduction

The application of artificial intelligence (AI) to polymer design is rapidly gaining traction as the polymer industry seeks to meet growing sustainability demands. By harnessing AI’s ability to predict properties, optimize formulations, and accelerate material discovery, the industry can create more efficient, sustainable polymers—critical for product protection in modern applications.

Key Takeaways

• Accelerating Polymer Discovery: AI can predict polymer properties and identify promising candidates faster than traditional methods, cutting down the time required for experimentation by 50-80%.
• Tailored Polymer Functionality: AI allows for the creation of polymers with specific properties such as strength, flexibility, and thermal resistance, helping manufacturers meet precise product requirements.
• Biopolymer Innovation: Machine learning is being used to design sustainable, bio-based polymers, advancing the transition from fossil-based to renewable carbon sources.
• Faster Time-to-Market: AI models can reduce the time to bring new polymers to market, with some processes now taking weeks instead of years.
• Data Challenges: For AI to work effectively, high-quality, accessible data is essential. The industry faces challenges with inconsistent datasets and intellectual property barriers.
• Skills and Collaboration Gaps: There is a need for cross-disciplinary expertise, combining polymer science with machine learning, to fully realize the potential of AI in polymer design.

Conclusion

AI’s role in polymer design is poised to revolutionize the development of more sustainable and functional materials, but several barriers remain, including data quality and the need for specialized skills. The industry must overcome these hurdles to unlock the full potential of AI.

We have heard about and indeed reported on many of the emerging uses of artificial intelligence (AI) tech when it comes to helping the packaging industry meet its sustainability goals, whether that’s retrieving ocean plastic waste, monitoring recyclability, improving supply chain visibility or veering into the vaguely apocalyptic. Here, however, we are going to focus on one particular application – that of using AI to create more sustainable and more efficient polymers.

The demand for ever-more advanced polymers that can fulfil the demands of product protection for modern applications while meeting vital environmental goals is growing. In future the traditional routes to polymer design, which rely on rigorous empirical experimentation, may not longer be able to keep up.

It may be fair to say that the polymer industry has been relatively slow to pick up on the possibilities for AI in materials discovery compared to sorting, recyclability and so on. But it’s now gaining momentum.

To put this in some context, the AI in Materials Discovery Market Size and Forecast 2025 to 2034 report found that ‘the demand for AI in materials discovery is increasing due to the growing need for rapid investments and innovations’. Within this, by material type, the polymers segment accounted for the highest revenue share of 28% in 2024.

Benefits and applications

AI can support polymer design by using machine learning to predict properties and optimize synthesis. Researchers train AI on large datasets of existing polymers to identify structure-property relationships, enabling them to identify promising candidates more quickly and efficiently.

Ellen de Ruiter, Consultant, Circular and Biobased Plastics at TNO, the independent Netherlands-based research organization, elaborates on how AI can be employed in this context: “AI is used to actually predict the properties of a certain polymer structure and combination of monomers. And if you have a model that can do that, you can then also go the other way around and actually identify the properties you need first and then design for that. An example of where this might be used would be very specific application areas such as medical polymers.”

Hannah Melia, Head of Marketing at generative AI specialists Citrine Informatics, also highlights the increasing significance of AI driven predictive experimentation models and how these work in real-world settings: “The product expert sets objectives and constraints on final properties and the models then rank candidate polymers that have the highest likelihood of meeting the constraints.

“The expert chooses from these candidates which to take to the lab. They test the resulting polymers and feed the test results back into the system and use that data to further train the models and get new candidate polymers. This iterative loop has been shown time and again to get to target properties with 50-80% fewer experiments than trial and error by an experienced formulator.”

What follows are some examples of the specific ways in which polymer discovery and development can be improved through these approaches:

• Tougher, more functional plastics: As we have already touched upon, AI modelling can allow developers to create polymers with more targeted functionalities – all those vital properties we’re all well aware of, including strength, flexibility, thermal resistance.

We’ve already seen real-world examples of this. Nestlé’s R&D team announced this year it has collaborated with IBM Research to develop new AI tools, including one ‘capable of proposing new high-barrier packaging materials for shielding products from moisture, oxygen and temperature changes’.

Scientists from both companies leveraged AI-based processing techniques to construct a knowledge base of known materials from public and proprietary documents. The team then fine-tuned a fit-for-purpose chemical language model, enabling it to learn the representation of the molecular structures.

Over in the US, researchers at MIT and Duke University have used machine learning technologies to identify ‘crosslinker’ molecules that can be added to polymers to enable them to withstand more force before tearing, resulting in more durable materials.

• Biopolymer discovery: There is of course an increased onus on developing novel biopolymers to replace, where possible, fossil-based varieties. Just one example of where AI is currently being used here is provided by Ellen de Ruiter, who tells us about TNO’s PolySCOUT, which ‘offers a highly effective Machine Learning solution that accelerates the design of novel, bio-based polymers’.

“As part of the sustainability transition, we need to move from fossil-based to renewable carbon, and we can use either biomass or CO2 as feedstocks for these polymers. PolySCOUT is a proprietary machine learning model that can help us to design new biopolymers for speciality applications.”

According to the company’s website: ‘The PolySCOUT suite of models supports novel polymer development in two key ways. Primarily, polymer developers, material scientists, and brand owners can input design criteria for novel polymers that encompass mechanical and physical properties, and eventually also factors like environmental impact, safety, feedstock availability, or economic feasibility. PolySCOUT then suggests polymer compositions that meet those criteria.

But as Ellen mentioned above, this can also be done in reverse: ‘Scientists who have already developed affordable, available, bio-based polymers can use the PolySCOUT models to uncover potential new applications for it.’

• Bringing new polymers to maturity faster: Up to now, developing a new polymer and bringing it to market has typically taken decades. But what if this process could be speeded up?

“Polymer scientists, based on their years of knowledge and experimentation and experience, know where to look for new materials but it’s a process of experimentation and intuition and the lab testing process can take years,” says Ellen. “Working with AI accelerates this process – even down to weeks in some cases.”

Aside from PolySCOUT, there is growing interest in advanced AI algorithms to provide this. Georgia Tech researchers, for example, are using and adapting AI algorithms to accelerate materials discovery. The team has developed ‘groundbreaking’ algorithms that it says can instantly predict polymer properties and formulations before they are physically created.

Barriers to adoption

With all the above, one might wonder why AI is not already employed more widely for polymer discovery. Wonder no more – as with any technology, there will inevitably be barriers and drawbacks along the way, such as:

• Data quality and availability: AI prediction models are only as good as the quality of the training datasets. If data is collected from a number of labs with varying measurement methods there’s a danger of inconsistency. Furthermore, even if good quality data exists, it may be difficult or impossible to access due to intellectual property issues. In future, to address this, there is a need for improved data sharing and perhaps greater standardization across the field.

Hannah Melia, however, is keen to stress that these barriers are not insurmountable – and in some cases there may be some misconceptions around them. “Contrary to popular belief, we don’t always need a lot of data to get going. Most companies that use the Citrine Platform have less than 100 datapoints when they start. In fact, companies that get distracted by trying to digitize all their data before doing AI, delay getting the value of AI.”

• The ‘black box’ question: Complex AI models increasingly function as ‘black boxes’, or rather, their internal processes are not transparent: while what is inputted is known and understood, it can be difficult to understand their decisions or outputs. Whether this is down to their essential complexity or the above-mentioned need to protect intellectual property, the result is the same – it is hard or impossible to understand the scientific reasoning behind them and this can limit the amount of ‘new’ scientific knowledge to be gleaned. This, in turn, can lead to a certain amount of scepticism in the scientific community.

But again, Hannah sounds a note of mitigation – not all techs are the same or present the same challenges. For example, “Our kind of AI at Citrine Informatics is not a black box that hallucinates – we are not built on the same technology as Chat GPT. Our platform shows its thinking, highlighting which factors are having an impact on final properties. This is one of the reasons product experts like using it. And every single prediction comes with error bars.”

• Skills gap: Clearly there is a need for more training in AI for traditional polymer researchers and this is something that still needs to be addressed. But if we’re looking at this in a positive light, it could also be argued that the growing intersection of technologies such as AI with traditional manufacturing and materials science has opened up interesting new possibilities for collaboration / connection across different fields.

Ellen de Ruyter is able to give a direct example of how this is happening within TNO right now: “In our PolySCOUT team, for instance, we of course have our polymer scientists who really understand the materials, but we also have machine learning specialists that do the modelling work. These people are learning from each other’s fields to develop these solutions.”

What’s next?

Moving into the future, there are still understandably plenty who are cautious about how far we should embrace AI. We don’t need to go all-out and picture a Terminator-style ‘Rise of the Machines’ scenario to know that complete reliance on AI at the exclusion of the human factor is an unsettling prospect.

But in this context at least, says Hannah Melia, there is probably no call for panic. “Our kind of AI always has a human in the loop. It is also not using the internet as a data source and so isn’t racially biased etc. It does accelerate people’s work, but we have not seen lay-offs at our customers, instead they seem to choose to do more projects simultaneously with the same number of employees instead. So, I can’t see anything to be concerned about.”

And furthermore: “I would argue that as we use AI to remove forever chemicals from plastics and reduce greenhouse gas emissions etc., we absolutely have a moral obligation to use the best, quickest tools possible for these tasks.”

To conclude, the worlds of AI and polymer design will almost certainly have to get used to the idea of being bedfellows moving forward – or at least there is going to be a great deal more bed-hopping. But rather than summarizing all we have already said above, we generally like to end these pieces with some practical advice for readers. To that end, we asked Hannah what practical advice she would give to polymer scientists looking to incorporate AI into their research. She offers these three points.

1. From now on make sure you clearly record all inputs (raw materials, processing parameters) and all outputs (final properties, cost, etc.) for every test sample you make and test.

2. Just as you wouldn’t try and code up Excel if you needed a spreadsheet, don’t try and DIY AI.

3. Pick an AI provider who, a) has an easy to use, no-code platform, b) has many years’ experience rolling out AI in polymer companies and understands change management, c) does not need lots of data to get started, d) has ISO 27001 certification or equivalent so you know your data will be secure.

If you liked this story, you might also enjoy:

The ultimate guide to the Packaging and Packaging Waste Regulation in 2025

How are the top brands progressing on packaging sustainability?

Everything you need to know about global packaging sustainability regulation in 2025

The key to increasing the use of reusable packaging in supermarkets