We talk about artificial intelligence both too much and too little: too much because of the frequent, spectacular promises about what the technology can do; too little because the real societal impacts often appear quietly and without fanfare. Szertics Gergely, director of the HUN-REN AI Service Center, argued that one of the main difficulties is that technology is advancing faster than society, companies and institutions can adapt.
The podcast’s regular expert guest is Aczél Petra, communication researcher and professor at Széchenyi István University. Both emphasized that the question is not whether AI is useful — research already shows substantial gains. AI accelerates data processing, pattern recognition and analysis for tasks that would be too slow or too large-scale for human effort alone.
Concrete research applications at HUN-REN
At HUN-REN, AI has already been applied in several research and development areas: cleaning virology databases, estimating lightning lengths, image analysis of animal experiments, and developing models that support research processes. These examples demonstrate AI’s practical benefits beyond demonstration projects.
Three levels: augmented, automated and autonomous intelligence
Szertics distinguishes three levels of AI use:
- Augmented intelligence: humans ask and the AI answers, with humans retaining control of the process.
- Automated intelligence: a workflow is divided into smaller, verifiable steps and AI assists in automating those steps.
- Autonomous intelligence: autonomous agent systems, much discussed today, but not yet the primary source of robust value in everyday organizational operations.
According to Szertics, the most tangible benefits today come from automated intelligence: this does not mean handing everything over to AI, but automating selected steps within a well-designed process (for example, downloading an email attachment, extracting invoice data, categorizing it and advancing the workflow).
This approach is less flashy than autonomous agents but far more controllable, reducing the risk of responsibility being washed out.
AI adoption as the "most human IT project"
Szertics argues that AI implementation is not a classic IT project: success depends chiefly on people’s understanding, responsibility and culture of use, rather than on how many software licenses a company buys. Organizations that think AI adoption is solved by buying a few hundred Copilot subscriptions may be disappointed.
For that reason, HUN-REN is building AI use strategically: through education, an ambassador system, technology selection protocols and a service-center model that helps decide which AI solution fits a given research or organizational task.
Which model should run where?
HUN-REN’s experience is that different situations require different solutions: sometimes an open-source model suffices; other times frontier models are needed; for sensitive data, locally-run, closed systems are required. This is especially important in science, where certain data cannot be transferred to external service providers. Szertics noted that in genomics, energy or nuclear power plant development, sending data to external servers is not an option, so local infrastructure must be used even if the local model is less capable than the latest international systems.
Shadow AI and changes in everyday life
The discussion also highlighted that the role of Google search is increasingly being replaced by ChatGPT-style interactions. When people turn first to an AI for advice about a sick child, workplace problems or everyday decisions, that change is more than a shift in search habits — it represents a new kind of human–technology relationship.
The so-called shadow-AI phenomenon is also spreading: in many organizations AI is not officially adopted, yet employees use it on personal devices and accounts, create text, summarize materials, interpret data and then bring results back into workplace systems. Management often behaves as if AI is not present, while in practice it already works in the background.
Szertics warned that this trend could have social impacts comparable to those of social media. Moreover, new devices, laptops, phones and software increasingly ship with built-in AI support, so AI is becoming embedded in the consumer environment.
The core question: are we shaping AI use consciously?
For HUN-REN experts, the decisive question is no longer whether AI will be part of our lives — it already is — but whether we actively shape how it is used, or let it reshape us invisibly.
Main topics of the conversation (timestamps)
- How does AI reshape society? (00:27)
- What positive impacts can AI bring to research? (04:07)
- Who is responsible if AI errs in an automated process? (10:17)
- Which research projects at HUN-REN benefit from AI? (14:36)
- Which jobs can remain human-centered alongside AI? (21:47)
- Why are not all fields equally automatable? (25:41)
- Which concrete AI projects is HUN-REN working on? (27:48)
- What is the bigger challenge: using safe AI or smart AI? (31:29)
- Why is AI a sensitive question for scientific data? (33:38)
- Why are AI and digital-world issues missing from political programs? (40:20)
Further details are available in the HUN-REN podcast episode referenced in the conversation.
