Where AI meets the real world
Inside the Institute for Strategic Artificial Intelligence
An AI-generated hypothesis only becomes worthy of a real-world experiment by being digitally or experimentally tested and refined in a closed-loop optimisation, many times over. That premise sits at the core of the Institute for Strategic Artificial Intelligence (ISA), working in deep partnership with UNE's LabNext70.
Founding Director Professor Amir Karton, a computational chemist whose career spans the intersection of chemistry, physics and computer science, leads ISA's research agenda alongside Founding Deputy Director Associate Professor Aaron Driver, UNE's Chief AI Officer and Director of LabNext70.
"While there is no shortage of AI centres and institutes globally," says Professor Karton, "what sets us apart is our integration of generative, predictive, and agentic AI into a single pipeline. We apply this architecture well beyond the physical sciences to override research bottlenecks across agriculture, clean technology, and regional resilience."
Prediction without validation is guesswork
ISA's core design principle is that AI output requires structured verification. The institute builds human-guided workflows where AI autonomously handles complex, repetitive and data-intensive tasks, while researchers define objectives and retain judgment at every decision point.
"We are not building systems that operate in a black box without expert oversight," Professor Karton explains. "Our architecture is designed around human-guided AI. The researcher remains at the centre of the pipeline, defining the strategic objectives and injecting expert judgment at every critical decision point to ensure our workflows are transparent and auditable."
This translates into a research pipeline built on two feedback loops. The first is digital: AI generates candidates - whether materials, molecules or hypotheses - and these are vetted through computational simulations. Results feed back to the AI, improving the next round of candidates. This cycle operates with high-throughput capabilities, allowing it to run rapidly and repeatedly.
"In scientific research, standalone gen-AI outputs are insufficient," says Professor Karton. "The system requires rigorous feedback and verified empirical data to ground its predictions and continuously improve its performance."
The second loop is physical. Candidates that survive digital vetting are passed to automated laboratories for real-world synthesis and testing. Those experimental results feed back into the AI in turn, further refining its output.
"After you go through these feedback loops many times, you really start to exhaust the chemical space," Professor Karton says. "This way we are systematically optimising the candidates, e.g., refining the materials or molecules until they are ready for real-world applications."
Solar recycling as proof of concept
ISA's solar panel recycling project, covered in detail in our earlier article, demonstrates this pipeline in practice. The digital feedback loop is now operational, and Professor Karton is integrating the experimental loop through an automated robotics platform at the University of Wollongong.
"This is a highly sophisticated deployment," he notes. The automated lab is digitally integrated with the AI pipeline, and once operational, will provide the physical validation layer that completes the pipeline.
A framework that travels
While the materials science workflow involves physical laboratories, the underlying architecture applies well beyond chemistry. The real-world validation step could be a database query, a web-based test, or any form of external verification.
"This validation architecture is fundamentally domain-agnostic," Professor Karton says. "The concept itself is something that we are going to adopt across many projects."
ISA's focus areas span physical sciences and research translation, digital agriculture and regional innovation, circular economy and clean technologies, sovereign AI and digital infrastructure, and community resilience. The institute currently has active deployment pipelines operating across these domains.
The institute operates as a premier technical asset, delivering deployable capabilities for government at all levels, industry partners, and the New England North West region.