FEATURE

By Karen Parker, Editor-in-Chief, ASI

With a strong foundation of clean and structured data, formulators can utilize artificial intelligence and machine learning to speed up the R&D process.

More and more, artificial intelligence (AI) is transforming the way we live and do business. The adhesive and sealant industry is no exception. With the help of software solutions, adhesive formulators can improve service to customers, improve efficiencies, and optimize material flows and workflows.

Jonathan Welch, head of Data Science at Albert Invent, answered some questions about artificial intelligence and machine learning models and how they can be applied in the formulation of adhesives and sealants. He also provided some industry insight into the Biden Administration’s new AI Executive Order. Welch has a PhD in Applied Physics from Harvard where he developed algorithms for Quantum Computation. His professional experience includes deep and statistical learning applications, developing modular data science pipeline architectures supporting very large datasets and cloud applications, and generative modeling in the image domain.

ASI: What is the scope of the Biden Administration’s new AI executive order and will it have an impact on the use of AI in the manufacturing and research and development sectors?

Jonathan Welch: The scope of the Biden Administration’s new AI Executive Order is broad and really focused on making sure very large-scale generative AIs are safe when deployed by requiring companies to demonstrate they understand the ways in which their AI tools can be potentially misused or present a potential threat to the safety and security of humans. What they are trying to prevent is someone making a very large generative AI or large language model (LLM) that has enough data and understanding that someone off the street or someone with ill intentions could use it to develop things that could harm people. The Executive Order is forcing anyone who builds these very large generative AIs to fully probe and consider all aspects of how it could be used. The key words here are “very large scale.” There are only a handful of companies on this planet that are capable of developing the kinds of generative AIs that fall within the scope of this Executive Order. So, as of now, we don’t believe the Biden Administration’s AI Executive Order will change very much for the adhesives and sealants manufacturing and R&D sectors. All the AI and machine learning (ML) models companies in this sector have in place today are too small and limited to be impacted by this Executive Order.

That being said, we do think Biden’s AI Executive Order should be a wake-up call for companies in all sectors to make sure our data is AI-ready and to consider the broad implications of this powerful tool. AI has the potential to be a game changer, bringing unprecedented speed and efficiency to the adhesives and sealants innovation process. However, for it to be an additive force, and not one that distracts or takes away from this industry’s important work, companies need to get their data in order now. And when it comes to R&D, many organizations in this industry are not ready.

ASI: How can formulators of adhesives and sealants harness the power of AI to improve and speed up their research and development?

Welch: There are lots of ways for formulators to harness AI/ML to help speed up their R&D process; however, most of them will require R&D organizations to have a strong foundation of clean and structured data — which they can do by using an end-to-end R&D platform, such as Albert. If they have this foundation in place, they can begin to build ML models that capture the relationship between formula composition (and potentially process parameters) and the final material properties that will allow them to start to study how changes in different parameters or components lead to changes in the final properties of a formula. They can also use more advanced techniques, like Active Learning, to leverage historical data to build models that help guide the selection of the next experiment to reach a particular design goal.

If you don’t have a clean and structured data foundation, there are still some things you can do. There are some general-purpose LLMs that could help teach you how to begin aggregating your data and build these types of models. Simply ask an LLM: “How can I prepare my formulation data for use in a machine learning model?” Then, dialog with it to learn how to clean and structure your data, and even write python code that can help to automate the process. Before you do this, however, make sure your organization doesn’t have any policies in place restricting you from interacting with LLMs for R&D, and be extra careful about keeping the data you share with it very broad and general, so you don’t risk any IP getting outside the organization. On the general research/knowledge acquisition side, there are AI tools like Elicit.org, which serve as natural language search engines for academic literature. These tools can help to quickly locate potentially useful papers/resources on research relevant to the types of chemistries encountered in the adhesives and sealant space by simply asking the AI the same questions you would typically ask a colleague helping you do a literature review.

ASI: What is generative design and how can it help adhesive and sealant formulators?

Welch: Generative design is a new approach to product design that uses AI to create multiple design options, which might meet specific performance criteria. For example, in the graphic design space, you might see the use of generative AIs like DALL-E or Midjourney being used to rapidly create and iterate on-concept art around a product ad campaign: “Give me an image of a happy family sitting around a campfire, enjoying a can of soda.” Within moments, you are presented with multiple variations on this scene, in potentially different styles to choose from. We believe the first instance of generative design in the formulation space will come in the form of finding novel candidate molecules to use as alternatives in a formula. There are already some initiatives happening to use generative models to propose individual molecules that exhibit a particular set of chemical or physical properties. While the potential for generative design in the formulation space could be enormous, it requires significant data to train a large foundation model that understands how to propose adhesives and sealant formulations directly. It may be hard for any company to do this on their own quickly, but over time if all R&D data is collected (including all successful and failed experiments) this could be achievable. That’s why Albert is so passionate about working with our customers to help them build a robust foundation of clean and structured data so they can be prepared to embrace generative design within their organization, once they accumulate enough data.

ASI: How can adhesive and sealant manufacturers employ AI to help them achieve their EHS (Environmental, Health and Safety – Regulatory) objectives?

Welch: If an R&D organization has enough clean and structured data on their chemicals, compositions, material properties, and EHS impacts, they can use this data to train AI models to predict the possible EHS impacts of a proposed new chemical or formulation. We generally see AI/ML used for new chemicals that aren’t tested and don’t have a toxicological profile, while following GHS rulesets is the most effective when the composition is fully known. Ensuring accurate EHS data comes down to having accurate data to follow GHS rules, along with enough data that is clean and structured for AI/ML models.

ASI: Can AI help manufacturers to enhance the production processes? And if so, how?

Welch: Yes, absolutely. Similar to EHS predictions, AI can help manufacturers predict the best ways to increase their yields and productivity before they step on the shop floor. The key is having enough clean and structured data to train the models on.