No competitive research without AI

At HUN-REN, the goal is to push that figure significantly higher. The network has already put a range of concrete measures in place to drive AI adoption, and in his presentation Jakab outlined three distinct levels of AI use — all of which are already represented within HUN-REN.

Three levels of AI

The first is augmented intelligence, where AI supports specific tasks, such as drafting a document. HUN-REN promotes this through its institutional AI ambassador network, multi-tiered training programs, and accessible AI infrastructure.

The second is automated intelligence, where entire workflows can be handed off to AI. One in-house example is a grant pre-evaluation system running on secure internal servers, which uses proprietary models and workflow-based solutions to pre-screen requirements and publications.

The third is autonomous intelligence, where users define only the desired outcome and the system independently carries out complex tasks using multiple tools. Human involvement is largely limited to validating the final output — though that validation process can itself be far from straightforward.

Jakab also noted that HUN-REN's AI strategy is currently being developed and will set out ambitious targets for the network. He suggested that within a year, it may be possible to identify five scientific publications in which AI could justifiably be credited as a co-author, reflecting its deep integration across all phases of the research process.

The need for human oversight

In the panel discussion that followed — where he was joined by Kathleen Walch, Director at the Project Management Institute — Jakab argued that AI delivers real value when organizations deploy it to address well-defined problems while keeping humans in the loop, particularly in decision-making. Both research institutions like HUN-REN and private-sector companies can make AI work for them only if they take an "AI-first" approach at the organizational level, building their workflows around artificial intelligence rather than bolting it on.

He also cautioned that AI-generated outputs — whether documents or code — do not create value in themselves, which brings human validation and accountability back to center stage. As AI tools grow more sophisticated and their outputs increasingly convincing, there is a real risk that users stop scrutinizing results critically. While the pace of development is remarkable, organizations would do well to set up dedicated "exploration teams" — groups tasked not only with spotting emerging solutions, but with rigorously evaluating whether those solutions actually fit their specific needs.

The Chinese example

Gergely Szertics, Director of the HUN-REN AI Service Center, took part in a roundtable titled "AI, innovation and results — when will the next leap occur?" He pointed out that given the size of the network, HUN-REN needs a distinctive approach to AI: rather than betting on a single large-scale system, the focus is on running numerous parallel micro-projects, with the emphasis firmly on building organizational capabilities.

Those capabilities are supported by the same initiatives Jakab described — AI ambassadors, automated tools, and related programs. Szertics stressed the importance of a support structure that works at multiple levels, serving both advanced users and those who are only just getting started with AI.

He also held up China as a key reference point, noting that the country is making rapid strides in fundamentally rethinking how research is conducted. Entire research ecosystems there are being rebuilt around AI, with artificial intelligence woven into every stage and every discipline.