AI Agents and the Human Perspective

The way we share our knowledge shapes the evolution of our species. Historically, it began with the spoken word, around the campfire. Later, writing and books allowed us to structure arguments and make them last. Video and audio brought the power of demonstration, of presence and of personality.

Today, as Artificial Intelligence advances, it is tempting to announce agents as the "inevitable next step" on this timeline. But at Foral we see things differently. Rather than arguing that agents are merely a new format for passing knowledge on, we believe they enable a new relationship between a person, their knowledge and those who seek it.

The difference may seem subtle, but it is fundamental. AI is not only there to "capture and perfectly represent" a human mind. It is there to create a living space for debate, continuous refinement and the solving of concrete, everyday problems, as long as it is paired with the human perspective.

Prudence and nuance: the true human advantage

A general-purpose large language model (LLM) holds a massive volume of factual information (and more), an advantage over humans. Yet the value of human knowledge lies not only in the answers an expert gives, but in their ability to recognise when they do not know the answer, to take responsibility for what they say and to change their mind in the face of new evidence.

If an agent cannot reproduce these traits (prudence, nuance and doubt), we are not truly amplifying the expert's knowledge. We are merely multiplying statements made in their name.

An AI agent works as a two-way bridge: it lets the creator discover gaps in their reasoning through interactions, lets the user weigh different interpretations, and lets the system itself make explicit the uncertainties that would otherwise stay hidden inside a seemingly complete narrative. Knowledge evolves not because AI reproduces it automatically, but because it supports an ongoing process of scrutiny and adaptation.

Beyond the book and the video: accessibility and monetisation

The traditional ways of sharing knowledge demand a monumental effort. Writing a book, or recording and editing a video course, takes hundreds of hours and a considerable financial investment, which creates a huge barrier to entry. On top of that, the author of a book will never know for sure how the reader will apply that information to their own specific case.

With agents, monetising knowledge becomes far easier and more accessible. An expert no longer needs to appear in dozens of videos or shut themselves away for months to write. They can structure their perspective and let their agent offer it 24 hours a day, adapting it to the exact question of whoever consults it. We are building the Mercado to make this new way of sharing and selling knowledge possible.

Foral's identity: territory, facts and the human perspective

This is where Foral's vision sets itself apart from other platforms. We are not simply trying to create replicas of experts. We are building an infrastructure that brings together human knowledge, companies and institutions, information about the territory, and tools that can help people make decisions.

This approach becomes vitally important when we look at Portugal's interior. Foral acts as a technological link that unites the territory: it lets local people and experts share their deep knowledge, protecting and preserving the identity and know-how of the region. At the same time, it lowers the barrier for those coming from outside. By combining highly specific local knowledge with rigorous data, it becomes far easier for anyone who wants to invest, restore heritage or move to the interior, connecting the whole country in a smarter and more cohesive way.

Imagine an experienced architect sharing knowledge about renovating rural properties in a specific municipality. A book explains the process. A video shows examples. The agent created on Foral does something entirely different: it represents that architect's methodology and perspective, but it also consults up-to-date planning sources, identifies the official documents required for that region, and distinguishes a general rule from what calls for an on-site assessment.

It is not just about consulting a human perspective, but about testing it against verifiable information and the concrete context of the user's decision.

The science of elicitation and the challenge of representation

To bring an expert's tacit knowledge into a model, we rely on science, while being fully aware of its limits.

The literature offers valuable tools, such as the Critical Decision Method (Klein et al., 1989), which lets us interview the expert and structure the heuristics they use without realising it. However, eliciting knowledge does not automatically guarantee that it will be represented reliably in an LLM.

Recent studies (such as Park et al., 2024) have shown that agents fed with qualitative interviews can replicate attitudes and consistency in surveys at a remarkable rate. But giving consistent answers to a questionnaire is not the same as reproducing a professional's judgement in new, complex situations with real consequences.

That is exactly the technical and product challenge we are solving at Foral: bridging the gap between mapping an attitude and simulating rigorous expertise.

Transparency and trust in AI

There is a real risk that the system blends the human's perspective with the AI's factual interpretation until the user can no longer tell the two apart. That is why transparency at Foral is not just a feature to avoid "hallucinations": it is a central, foundational characteristic of the product.

The user will always know, with clearly defined boundaries:

1. What the expert/creator explicitly said (their methodology and opinion).
2. What is official information or territorial data cross-referenced by Foral.
3. What the system synthesised from the original material.
4. What is out of scope (when the agent has to admit: "On this point, the architect always recommends a site visit").

Trust does not arise by decree. A recent study (published in Scientific Reports in 2026) found that, although there is initial resistance to AI advice, this barrier drops considerably once participants have had the chance to evaluate the answers, suggesting that the perceived usefulness of the advice can influence acceptance of AI.

Our founding thesis at Foral is this: it is possible to strengthen trust in AI by pairing the human perspective with Artificial Intelligence models that produce content properly anchored in verifiable data and facts about the territory.

The future of shared knowledge is not about replacing humans with machines, but about giving people the tools to bring their perspectives, honestly, into a world of ever more complex information. And that is the world, and the country, we are building at Foral.