Organizations have long had governance for money, for risk and for information security. As artificial intelligence systems become part of everyday work, a new kind of governance becomes necessary: governance of knowledge itself. I call this, following a growing body of research, epistemic governance.
The term sounds abstract, but the questions behind it are very concrete. When an AI system gives an answer, where did that answer come from? How confident should we be in it? Who is responsible if it is wrong? And how does the organization's own knowledge change when people increasingly rely on machine generated responses?
Knowledge is not the same as output
A language model can produce fluent, well structured text on almost any topic. Fluency is not knowledge. Knowledge, in the organizational sense, involves justification: we know something when we can explain why it is true, where it came from and under what conditions it holds.
Many AI systems produce answers without exposing their justification. This is not a reason to avoid them, but it is a reason to design organizational processes that restore justification where it matters. That means linking outputs to sources, recording the context in which an answer was produced, and keeping a human able to examine the reasoning.
Three questions for every AI deployment
When advising organizations, I suggest three questions for every use of AI in knowledge work.
What is the source? Is the system drawing on the organization's own verified knowledge, on public information, or on patterns learned from unknown data? Answers grounded in identified sources are easier to trust and to correct.
Who validates? For each type of decision, there should be a clear understanding of when a human must review machine output before it is used, and who that human is.
What is learned? When AI output is accepted, corrected or rejected, does the organization capture that? Or does each interaction vanish without contributing to collective knowledge?
The risk of epistemic dependency
One subtle risk is that organizations gradually lose the ability to know things for themselves. If junior professionals always ask a system rather than reasoning through a problem, they may never develop the judgment that experience used to build. Over time, the organization becomes dependent on a tool it cannot fully evaluate.
This is not an argument against AI. It is an argument for designing its use deliberately, so that it amplifies human learning instead of replacing it. Some organizations, for example, ask professionals to form their own view before consulting the system, and then to compare. This keeps human judgment active.
Governance structures
Epistemic governance does not require an entirely new bureaucracy. It can be integrated into existing structures. Knowledge management teams can curate trusted sources. Risk and compliance functions can define where human review is mandatory. Leadership can set expectations about transparency: for example, that any document substantially produced with AI is identified as such, and that critical claims are traceable to sources.
Universities and research groups, including groups at UFSC, are actively studying these questions. The field is young, and organizations that engage with it now will shape better practices for everyone.
Ethics at the center
Ultimately, epistemic governance is an ethical matter. Decisions about hiring, credit, health and public services increasingly involve AI. The people affected deserve to know that someone, somewhere, can explain how a conclusion was reached and is accountable for it.
Organizations do not need perfect answers to begin. They need the willingness to ask, openly, who decides what their machines know, and how that knowledge is kept honest.
Organizações têm governança para dinheiro, risco e segurança da informação. Com a IA no trabalho cotidiano, torna-se necessária a governança do próprio conhecimento: a governança epistêmica.
Saída não é conhecimento
Fluência não é conhecimento. Conhecer envolve justificação: saber por que algo é verdadeiro e de onde veio. Processos organizacionais devem restaurar essa justificação, ligando respostas a fontes e mantendo humanos capazes de examinar o raciocínio.
Três perguntas
Qual é a fonte? Quem valida? O que é aprendido? E atenção à dependência epistêmica: profissionais que nunca raciocinam sozinhos não desenvolvem julgamento.
Ética no centro
Quando decisões que afetam pessoas envolvem IA, alguém precisa ser capaz de explicar como se chegou à conclusão e responder por ela.