Researchers at UNESCO AI ethics forum emphasise need for human judgement

Artificial intelligence must accelerate scientific discovery without weakening the human judgement, scrutiny and integrity that underpin research, speakers told the third day of the UNESCO Global Forum on the Ethics of AI in Riyadh.

During the session “Rethinking Sciences in the Age of Artificial Intelligence”, Dr Tran Tien Cong, Vice Dean of the Faculty of Artificial Intelligence at Vietnam’s Posts and Telecommunications Institute of Technology (PTIT), argued that education systems need to prepare researchers for a rapidly changing AI landscape rather than simply teaching them how to use today’s tools.

Dr Cong noted that PTIT established Vietnam’s first Faculty of Artificial Intelligence two years ago, raising a fundamental question about what students should learn when the AI tools available when they graduate may be very different from those used today. He said the priority should be developing people who can continue to think, learn and exercise scientific judgement even as the technology changes.

Drawing on Vietnam’s national strategy for science, technology, innovation and digital transformation, Dr Cong highlighted the need for two complementary layers of AI capability: broad AI literacy among professionals such as doctors, economists, lawyers and engineers, alongside highly trained researchers capable of understanding the mathematics, algorithms and scientific principles behind AI and developing new technologies.

He also stressed the importance of introducing students to research early, including through collaboration with research groups, laboratories and industry. PTIT is extending this approach to younger students, reflecting Dr Cong’s view that research culture begins before a student writes their first scientific paper.

The challenge, he said, is that AI can increasingly generate explanations, conduct experiments and produce research papers, creating a risk that researchers become better at obtaining answers while becoming less capable of questioning whether those answers are correct.

Dr Cong identified three capabilities that will become increasingly important: strong foundations in mathematics, statistics, algorithms and scientific methods; the ability to work across disciplines; and intellectual independence. He highlighted an “AI+X” approach, arguing that future breakthroughs could emerge from combining AI with fields such as biology and medicine rather than developing AI in isolation.

Researchers must also be willing to ask why an AI system produced a particular result, whether it can be reproduced and whether the evidence supports it. As AI becomes more capable, he argued, qualities such as curiosity, scepticism, ethics, imagination and responsibility become more important, not less.

“The objective is not simply to train the people who use AI,” Dr Cong said. Instead, the aim should be to develop a generation able to “understand AI, create AI, challenge AI” and use it to expand human knowledge without surrendering human judgement.

The importance of embedding responsible AI throughout the research process was reinforced by Dr Maad Alowaifeer of King Fahd University of Petroleum and Minerals. He described collaboration with Saudi Arabia’s National Center for Artificial Intelligence on Saudi Sign Language recognition, including the development of a large dataset while addressing privacy concerns. The project led researchers to consider how much information is contained in facial expressions and whether sign-language systems could work from representations of hand and body movements without capturing identifiable facial features.

Dr Alowaifeer argued that ethics should not be treated as an additional layer applied after an AI system has been developed. Instead, responsible AI needs to be considered throughout the process, as addressing ethical and privacy issues only after development can be too late and potentially more expensive.

The discussion also examined how AI-enabled science can be made more equitable. Liviu Știrbăț, Head of the Unit for AI in Science at the European Commission’s Directorate-General for Research and Innovation, described the EU’s efforts to pool data, computing resources, funding and talent across member states. He highlighted collaboration between AI specialists and researchers in fields including materials science, physics, social sciences and biotechnology.

Prof Anicia Peters, CEO of Namibia’s National Commission on Research, Science and Technology, highlighted the uneven distribution of AI resources, noting that access to computing infrastructure remains a significant challenge across Africa. She also stressed the importance of protecting indigenous knowledge and ensuring that AI development does not turn valuable local data into resources extracted without appropriate safeguards.

The environmental implications of AI were addressed by Carlo Angeles, Director of Impact at C-Minds and an inaugural Senior Fellow of the Science Panel for the Amazon. He pointed to AI’s energy, water, infrastructure and materials footprint, while also highlighting its potential to support environmental research and ecosystem protection.

Denisse Albornoz, Deputy Director of The Engine Room, meanwhile, examined the risks associated with open data and AI, particularly where indigenous and local knowledge is incorporated into AI systems without meaningful consent or community involvement. She argued that openness is not automatically neutral or beneficial and that greater attention is needed to who controls data, who benefits from it and how communities can influence its use.

Across the discussion, speakers returned to the need for science to remain open, trustworthy and inclusive as AI becomes more deeply embedded in research. The opportunity to accelerate discovery is substantial, but so too is the need for verification, transparency, ethical safeguards and human oversight.