Artificial intelligence is transforming research at a pace few institutions anticipated, and many are still learning how to respond. Agentic AI can analyse vast data sets, simulate complex phenomena and generate novel insights at a scale and speed that are entirely new. To dismiss AI out of fear would be shortsighted; we would be forgoing a technology that could solve problems which humans alone cannot. And we also need to equip our students to use technology they’ll be expected to master when they enter the workforce.
When it comes to generative AI, I am strongly in favour of its use. I am equally insistent, however, that we use it with care, intention and clear responsibility.
- GenAI as a research assistant: transforming start-up due diligence
- Not faster but fairer: tame ‘techno-solutionism’ to build inclusive futures with AI
- Spotlight guide: GenAI as a research assistant
The challenge is not in AI itself but how we use it. In the research context, this means designing ethical guardrails and governance systems that balance safely with maximising potential.
AI as an opportunity, not a threat
AI, particularly in its agentic form, has enormous potential to accelerate discovery. It can model chemical reactions, predict material behaviours and analyse biological systems at scales and speeds far beyond human capability. Consider the challenge of a shifting environment: AI can evaluate millions of potential materials for carbon capture and water purification – tasks impossible for individual researchers to do manually.
But we must ensure that our use of AI is safe and ethical.
Designing effective safeguards
Safety must be built into AI architectures from the start. They should include:
- operational limits: clear boundaries on what an AI system can and cannot do, including constraints on autonomous actions and decision-making authority
- ethical parameters: built-in rules to prevent harmful outputs, biased recommendations or misuse of sensitive data
- verification mechanisms: systems to validate outputs before they influence research decisions, such as cross-checking results against physical laws, known benchmarks or human expert review.
These safeguards are not barriers; they are enablers. In research, the real risks include over-reliance on opaque models, propagation of bias from training data, hallucinated or irreproducible results, and unintended use of AI-generated outputs in high-stakes decisions. By defining operating conditions carefully, we allow AI to tackle complex problems with confidence while minimising these risks.
Governance at every level
Governance includes model validation protocols, access controls, audit trails, version tracking and mandatory human-in-the-loop review for consequential decisions. Universities and research institutions need policies that guide responsible AI use. These should address data privacy, intellectual property, reproducibility and appropriate human oversight. Researchers must understand which tasks can be AI-assisted and which should remain firmly in human hands.
On a broader scale, collective governance is essential. Just as cybersecurity relies on shared standards and threat monitoring, AI requires community-driven frameworks to prevent misuse. Monitoring, auditing and regulatory systems can detect unintended behaviours, safeguard sensitive research and prevent malicious application.
Regulation will also play a central role. Rather than blanket restrictions, this should take the form of risk-based regulation: lighter oversight for low-risk applications such as exploratory modelling, and stricter requirements for high-impact or sensitive domains. The future of AI safety will depend on both preventive design and active oversight. As models grow more capable, so will the need for detection systems that identify bias, data leaks or harmful use. The aim is not to slow progress but to channel it responsibly.
Ethical and practical data stewardship
AI depends on data. Protecting IP and ensuring responsible use of research data sets is critical. Researchers should clearly define what data is used, how it is stored and how it will inform AI outputs. They should also disclose AI use to those affected by research, in line with transparency principles. Ethical oversight helps ensure that AI supports the public good rather than reinforcing biases or creating unintended harms.
This emphasis on stewardship is both ethical and practical. Properly managed data allows AI to operate at full potential, producing insights that can transform science and society. Neglecting these measures risks not only safety but the credibility and utility of AI-assisted research.
Cultivating collective expertise
No single individual or lab can manage these complexities alone. Institutions can catalyse communities of practice around AI governance, encouraging collaboration among researchers, data scientists, ethics boards and IT teams. Shared standards, continuous training and transparent communication build trust and accountability.
Researchers also need literacy in AI principles – not just in how to run models but how to interpret outputs critically. Knowing the limitations and assumptions of AI systems is essential to prevent errors and maximise impact.
Optimism through responsibility
The most exciting aspect of AI is not that it can do what humans already do faster but that it can explore what humans cannot yet perceive. Capabilities to test thousands of hypotheses, simulate chemical structures and map complex systems at speed could unlock breakthroughs in medicine, energy, environmental science and materials engineering.
Yet the promise of AI is inseparable from responsibility. By designing safeguards, ethical frameworks and governance systems, we ensure that AI’s power is harnessed safely, reliably and for the collective benefit. Responsible deployment is not a constraint on innovation; it makes AI a tool for progress rather than risk.
The path forward
Higher education has an opportunity to lead. By building AI governance frameworks, investing in training and empowering collaboration across institutions, universities can ensure that AI delivers maximum benefit.
AI will not replace researchers; it will empower them. The key to unlocking its full potential lies in our ability to pair ambition with safeguards, curiosity with ethics, and speed with thoughtful oversight. The future of research depends on AI being both powerful and responsible, and on the human commitment to guide it wisely. Harnessed responsibly, AI offers a path to breakthroughs in medicine, energy, sustainability and countless other fields. To ignore such technologies would be a lost opportunity.
Hongliang Xin is professor of chemical engineering at Virginia Tech.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
comment