by Kaveen Tennakoon, PhD Candidate in Chemistry
20 May 2026
AI for Smarter Materials and Self-Driving Labs
In Part 1, we explored how AI is already reshaping molecular chemistry by accelerating retrosynthesis, improving reaction prediction, supporting mechanism discovery, and transforming early-stage drug development. These advances show that AI is no longer just a computational add-on; it is becoming a practical research partner that helps chemists move faster from idea to experiment. In Part 2, we shift from molecules and synthetic routes to a broader frontier: materials discovery, self-driving laboratories, and AI-assisted physical and analytical chemistry. Here, AI is not only helping scientists predict what materials might work best, but also guiding experiments, interpreting complex data, and opening new ways to understand chemical systems in real time.
Designing Tomorrow’s Materials with AI
Materials science is another arena where AI is dramatically boosting the pace of innovation. Whether the goal is a better battery electrolyte, a more efficient solar cell material, or a new polymer, researchers must sift through an astronomical number of combinatorial possibilities. Here, AI’s ability to recognize patterns and optimize is invaluable. Predictive models can be trained on existing materials data to quickly forecast properties of untested compounds, allowing virtual screening of candidate materials at a scale that would be impossible in the lab. As a recent World Economic Forum report noted, “advanced machine learning models…can predict properties for numerous substances in minimal time, rapidly screening candidate materials against desired parameters.”. For example, if a materials scientist needs a molecule with high thermal stability and certain optical properties, an AI model can scan millions of candidates and flag a handful that best meet the criteria. This means fewer experiments wasted on dead-end leads, and a higher chance of finding that game-changing material.
Beyond property prediction, generative AI plays a growing role in material design. Instead of just evaluating given candidates, generative models can propose brand-new molecular structures or crystal architectures optimized for a specific target property. In one striking vision outlined by researchers, an AI platform might “instantly generate millions of unprecedented molecular structures, predict their properties and propose cost-effective synthesis pathways” for a desired materials application. Through iterative cycles of suggestion and evaluation, AI can effectively design novel materials from scratch. This approach is already taking shape. For instance, generative models have been used to design new catalysts and OLED materials by learning the underlying chemistry and then searching beyond known chemistries for better solutions.
What truly sets materials AI apart is the rise of self-driving laboratories. In these setups, AI algorithms are directly linked to automated synthesis and testing equipment, creating a closed-loop optimization system. The AI recommends an experiment, a robot or automated flow reactor runs it and collects data, then the AI learns from that result to suggest the next experiment. Pioneering work in 2025 demonstrated that such a self-driving lab can accelerate materials discovery tenfold: one system at North Carolina State University gathered 10× more data than conventional methods and identified optimal material candidates in a fraction of the time. By operating continuously and adjusting experiments on the fly, the AI-guided setup found the best-performing material on the very first try after training, a remarkable outcome. This closed-loop approach doesn’t just save time; it also saves resources and energy, since the AI can intelligently skip unpromising experiments. As one researcher put it, “it allows us to identify promising material candidates in weeks instead of years, while reducing both costs and environmental impact”.
Globally, there is a race to deploy AI in materials R&D. Major tech companies and national labs have launched initiatives like MatterGen, GNoME, and others, aiming to vastly augment the scale of materials research with AI. Startups are also innovating: some are coupling automated labs with AI “brain” software to create rapid discovery platforms, while others focus on specialized AI models (for example, large pre-trained neural networks that serve as general-purpose chemistry engines). An example is the development of machine-learning-based interatomic potentials, models like OrbNet from Orbital Materials, which can accelerate quantum chemistry simulations for molecules and materials. These AI-driven potentials enable researchers to simulate reactions and material behavior with near-first-principles accuracy. In practical terms, this means complex physical chemistry calculations (like molecular dynamics or electronic structure predictions) can be done much faster, allowing scientists to explore reaction mechanisms or material stability over longer timescales and larger systems than before. By blending domain knowledge with data-driven insight, AI is empowering materials scientists to tackle problems (like finding new superconductors or carbon capture materials) that were once deemed too complex or time-consuming, thus pushing the frontiers of materials design.
| AI Tool/Platform | Capabilities Overview |
|---|---|
|
Helps researchers search for known and predicted materials and compare crystal structures, formation energies, band gaps, phase stability, and other properties. It is especially useful as a starting point for virtual screening and for training machine-learning models. |
|
|
Converts chemical composition, crystal structure, and materials data into features that can be used to train ML models. It is useful for building property-prediction models from existing datasets. |
|
|
Predicts stable crystal structures at a very large scale. DeepMind reported that GNoME identified 2.2 million new crystals, including about 380,000 predicted stable materials, greatly expanding the pool of possible inorganic materials. |
|
|
Generates new inorganic material candidates and can be guided toward target properties such as bulk modulus, chemical system, or magnetic density. This makes it useful for designing materials rather than only screening existing ones. |
|
|
Predicts energies, forces, stresses, magnetic moments, and stability-related behavior much faster than many traditional quantum-chemistry calculations. It is useful for large-scale atomistic modeling, molecular dynamics, diffusion, and phase-transition studies. |
AI for Physical and Analytical Chemistry Insights
Not only is AI driving the creation of new molecules and materials, it’s also transforming how scientists analyze and understand chemical phenomena. In spectroscopy and analytical chemistry, AI algorithms are taking on tasks that traditionally required significant human expertise. A classic example dates back to the origins of chem-informatics: the DENDRAL expert system in the 1960s was an early AI that helped identify molecular structures from mass spectrometry data by encoding chemists’ heuristics. Today’s AI approaches have far surpassed those rule-based systems. Modern machine learning can tease out subtle patterns in spectra, helping assign structures or monitor reactions in real time. For Instance, in nuclear magnetic resonance (NMR) spectroscopy, researchers have used deep learning models such as DP4-AI to automatically elucidate the structure of complex molecules, achieving a 60-fold increase in processing speed over traditional methods. Another model can automatically detect signal regions in proton NMR spectra with accuracy rivaling that of expert spectroscopists, completing in seconds what used to be a manual, time-consuming step. These advancements hint at a future where much of the routine data processing in analytical labs is handled by AI, allowing scientists to focus on interpretation and discovery.
AI is also starting to guide experiments in real time. In spectroscopy or kinetic studies, this means an algorithm can analyze incoming data and decide how to adjust experimental parameters on the fly to gain the most insight. One vision articulated by Prof. Roland Riek (ETH Zurich) is using AI during NMR experiments to recognize which measurements are most informative and skip those that aren’t needed . His goal is to cut the time to fully characterize a protein structure from “six months to a few years” down to a far shorter timeframe by letting AI “make automated changes to the experiments on-the-fly” based on intermediate results . In essence, the AI would monitor the data and decide, for example, which pulse sequence to run next or when enough data has been collected for a particular analysis. This kind of intelligent automation could revolutionize physical chemistry experiments, much like self-driving cars do for driving, a “self-driving spectrometer” could maximize efficiency and minimize human intervention. While these adaptive experimental AIs are still in their early stages, proof-of-concept studies demonstrate that they can be effective. Even in quantum chemistry, AI is making inroads: machine learning models can predict the outcomes of high-level quantum calculations (such as DFT or CCSD) with near accuracy, effectively short-circuiting expensive computations. By combining neural networks with physical constraints (for example, ensuring predictions adhere to known conservation laws or symmetries), researchers have solved complex chemical physics problems significantly faster than brute-force methods. All these developments indicate that AI isn’t just doing chemistry for us; it’s enabling us to probe deeper into the fundamental workings of chemical systems, by handling the heavy data lifting and letting scientists ask more ambitious questions.
Together, these developments show that AI is changing chemistry at both the discovery and analysis levels. From designing next-generation materials to operating autonomous labs and extracting deeper meaning from spectroscopic and physical chemistry data, AI is helping researchers explore chemical space with greater speed, precision, and creativity. In Part 3, will look at the practical tools and platforms driving this revolution In more details, including open-source resources, commercial software, and user-friendly AI systems that are making these technologies accessible to everyday chemists. As these platforms continue to grow, the key question is no longer whether AI will influence chemistry, but how chemists can use it wisely, critically, and effectively in their own research.