Faculdade de Filosofia e Ciências Sociais (Faculty of Philosophy and Social Sciences), Universidade Católica Portuguesa (Portuguese Catholic University), Braga, Portugal. (Photo credit: Diário do Minho at diariodominho.pt)
In my previous blog, When Assistance Becomes Dependence: The Risks of Over-Reliance on AI in Research and Education, I proposed reframing artificial intelligence (AI) not as a source of intelligence, but as a form of artificial assistance. This distinction is more than a matter of terminology. It reflects how we understand the relationship between human intellectual capabilities and increasingly sophisticated technological systems.
Artificial assistance can expand what researchers are capable of doing. It can accelerate literature searches, support data analysis, suggest alternative interpretations, and facilitate academic writing. Yet, these capabilities raise a fundamental question:
As AI becomes increasingly involved in the production of scientific knowledge, how do we ensure that researchers remain intellectually responsible for the knowledge they produce?
A recent visit to Portugal brought this question into sharper focus.
Reflections from Braga, Portugal: Research Integrity in an AI-Driven World
Recently, I had the privilege of visiting the Universidade Católica Portuguesa (Portuguese Catholic University) in Braga, Portugal, where I engaged in professional collaborations and stimulating conversations with faculty and researchers at the Faculdade de Filosofia e Ciências Sociais (Faculty of Philosophy and Social Sciences).
Our discussions centered on the growing presence of AI in research, particularly in the humanities and social sciences, where interpretation, contextual understanding, theoretical perspectives, and human judgment are fundamental to knowledge production.
The conversations were intense, intellectually engaging, and remarkably consistent in one important respect: AI is not a replacement for the researcher.
AI can support different phases of the investigative process, from identifying relevant studies to organizing information, analyzing data, and refining academic writing. However, the researcher remains responsible for methodological choices, research design, interpretation of findings, and the integrity of the investigative work.
This responsibility is not new. Scientific integrity has always depended on the honesty, competence, and ethical judgment of researchers. The history of science includes numerous examples of fabricated evidence, manipulated results, plagiarism, and other forms of scientific misconduct. These practices existed long before generative AI.
What AI changes is not the fundamental responsibility of the researcher, but the ease and scale with which questionable practices can enter scientific work.
A fabricated reference can appear in seconds. A plausible but inaccurate interpretation can be generated almost instantly. An entire manuscript can be produced without the intellectual engagement required for scholarly writing.
The resulting text may appear sophisticated, coherent, and academically credible. Yet, its apparent quality tells us little about the intellectual process that produced it.
This distinction between producing scholarly-looking content and engaging in scholarship deserves our attention.
Academic Integrity: What Can We Learn from Universidade Católica Portuguesa?
One important safeguard against the inappropriate use of AI is the development of institutional policies and guidelines that articulate expectations for responsible academic practice.
The approach adopted by Universidade Católica Portuguesa provides a compelling starting point for reflection.
In January 2025, the university developed its Política de integridade académica e científica: Carta de princípios no uso de IA (Academic and scientific integrity policy: Charter of principles for the use of AI).
The document provides guidance for the use of generative AI in academic and scientific work, emphasizing that the availability of AI tools does not diminish the intellectual responsibilities of students and researchers.
Four principles are particularly relevant to this reflection.
1. Authorship: Who Is the Intellectual Author?
Students and researchers must remain the authors of their work.
A text produced entirely by AI through a simple sequence of instructions cannot legitimately be presented as one’s own intellectual contribution.
According to the university’s guidance, legitimate authorship requires that individuals substantially contribute to the content and form of the work, understand and communicate its meaning, and explain the stages of its preparation and their involvement in them.
This principle recognizes that using AI does not automatically invalidate authorship. What matters is the substantive intellectual contribution of the human author.
2. Responsibility: AI Is a Tool, Not an Intellectual Source
AI tools do not possess personal responsibility and, therefore, should not be treated or cited directly as intellectual sources.
Instead, they should be understood as tools, comparable in this respect to statistical software, search engines, translators, and proofreading applications.
This distinction is especially important in research. A language model may generate an explanation of a theoretical concept, but that explanation does not replace the original scholarly literature or the responsibility to engage with it.
A researcher who relies on an AI-generated synthesis without consulting the underlying scholarship risks accepting interpretations that may be incomplete, inaccurate, or disconnected from their disciplinary context.
The convenience of obtaining an explanation should never be confused with the intellectual accomplishment of understanding it.
3. Accountability: Can Researchers Defend Their Work?
Academic work should reflect the intellect and voice of its author, regardless of whether generative AI was used during research and writing.
Students and researchers should be prepared to explain the content and logical structure of their work and to be transparent about where, how, and why AI was used.
The inability to provide such explanations may indicate excessive or inappropriate reliance on AI, potentially constituting a form of AI-assisted plagiarism.
Accountability, therefore, extends beyond acknowledging technological assistance. It requires researchers to demonstrate that they remain intellectually engaged with the ideas, evidence, and conclusions presented under their names.
4. Explainability: Understanding the Process Behind the Product
Authors should be capable of explaining the processes that produced the different components and arguments of their work.
In educational contexts, the policy recommends that students document AI-influenced elements or reasoning in an appendix. When written explanations are insufficient or impractical, oral discussion may serve as a complementary assessment strategy.
This principle goes beyond simply disclosing the use of AI. It asks individuals to demonstrate that they understand and can defend the intellectual work presented as their own.
These four principles are translated and summarized from Universidade Católica Portuguesa (2025).
Taken together, these orientations establish an important boundary: AI may participate in the research process, but it cannot assume the intellectual obligations of the researcher.
The example from Universidade Católica Portuguesa offers a thought-provoking perspective on how one institution is addressing the challenges of AI in academic and scientific work. However, this does not mean that its principles are universally applicable or should be adopted without critical reflection. Rather, they provide a starting point for other higher education institutions to examine their own practices, engage in meaningful dialogue, and develop safeguards that respond to their specific institutional contexts, communities, and needs. What works at Universidade Católica Portuguesa may not necessarily be appropriate for institutions operating within different educational, cultural, organizational, or regulatory environments.
These principles also point toward a deeper concern. Academic integrity is not simply about identifying whether AI was used, nor should institutional efforts focus exclusively on students. Responsible AI use must encompass the entire higher education ecosystem, including faculty, researchers, students, administrators, and professional and support staff whose academic and professional activities increasingly involve AI-assisted work.
The central question is whether all members of this ecosystem remain genuinely and intellectually engaged in the processes of producing, evaluating, and validating knowledge, as well as developing the work-related products and outcomes for which they are responsible. Ultimately, safeguarding academic and professional integrity requires a shared institutional commitment to human judgment, accountability, and meaningful intellectual engagement, regardless of the role AI plays in supporting these activities.
This brings us to what I consider one of the most consequential challenges of AI-assisted research: epistemic responsibility, meaning the intellectual obligation of researchers to understand, critically evaluate, justify, and take responsibility for the knowledge claims they produce and communicate.
The Epistemological Risk: Who Is Responsible for Knowing?
Research is not simply the production of information. It is a systematic process through which knowledge claims are developed, examined, challenged, and justified.
Researchers are expected to understand where evidence comes from, how it was produced, which methodological assumptions shaped it, and what limitations constrain its interpretation. They must distinguish observation from inference, recognize theoretical disagreements, and be prepared to defend the conclusions they draw.
AI complicates this relationship between the researcher and the knowledge being produced. Here are a few examples.
Literature searches: When AI identifies relevant articles, did the researcher verify the comprehensiveness of the search and consider potentially overlooked perspectives or contradictory findings?
Research synthesis: When AI summarizes scholarly publications, did the researcher engage sufficiently with the original sources to recognize theoretical disagreements or missing nuances?
Statistical analysis: When AI recommends a statistical procedure, does the researcher understand its underlying assumptions and appropriateness for the research questions and the data?
Qualitative interpretation: When AI interprets qualitative findings, does the interpretation reflect the researcher’s understanding of participants, context, and theoretical framework, rather than merely an AI-generated narrative?
These points underscore a fundamental distinction between having access to AI-generated outputs and possessing the knowledge and experience necessary to critically evaluate them.
The issue is not simply whether an AI-generated output is correct. The deeper issue is whether the researcher can explain why that output should be trusted.
Epistemic responsibility refers to the obligation to understand, evaluate, and justify the knowledge claims presented to others.
A researcher who accepts an AI-generated output without critically examining the underlying evidence and reasoning flaws becomes a passive consumer of AI-generated knowledge, rather than an active investigator engaged in the production and validation of knowledge.
When Research Tasks Become Automated, What Happens to Intellectual Engagement?
As AI systems evolve from conversational assistants into agents capable of executing complex, multistep research tasks, a fundamental concern emerges: Are we confusing the completion of research tasks with the intellectual engagement required to produce knowledge?
Imagine an AI agent conducting a literature review, synthesizing findings, recommending methodologies, generating analytical code, and drafting a manuscript. While these capabilities offer remarkable efficiency, what happens when researchers cannot explain why certain studies were selected, analytical decisions were made, or competing interpretations were evaluated?
Efficiency does not necessarily establish scientific validity, nor does a convincing explanation guarantee that its conclusions are grounded in evidence and a rigorous investigative process.
This raises an uncomfortable possibility: Researchers may increasingly be able to produce scholarly outputs without developing the depth of understanding associated with producing them.
The concern is not simply that AI may generate incorrect conclusions, but that researchers may gradually disengage from the critical inquiry necessary to evaluate them.
Ultimately, the challenge of AI-assisted research is not merely technological or ethical. It is epistemological, raising fundamental questions about the relationship between knowledge, justification, and human understanding.
Academic Integrity Must Protect the Process of Knowing
The principles articulated by Universidade Católica Portuguesa provide a foundation for navigating these challenges.
Authorship, responsibility, accountability, and explainability remind us that research integrity cannot be reduced to detecting AI-generated text or requiring disclosure statements.
A manuscript may transparently acknowledge AI assistance and still raise serious questions about the researcher’s intellectual engagement.
Conversely, a researcher may use AI extensively while maintaining strong methodological judgment, independently verifying evidence, and exercising genuine intellectual authorship.
The decisive question is not simply how much AI was used. It is whether the researcher maintained responsibility for the intellectual decisions that shaped the work.
This distinction should inform how we approach research supervision, doctoral education, scholarly publication, and institutional policies.
Rather than focusing exclusively on whether AI contributed to a final product, we should also examine whether researchers can explain the reasoning, evidence, and decisions behind that product.
Academic integrity must protect not only the credibility of what is published, but also the intellectual practices through which knowledge is created.
A Final Reflection: Artificial Assistance Cannot Replace the Responsibility to Know
Returning to my conversations in Braga, I am reminded that scientific inquiry is fundamentally a human endeavor.
Research requires more than the capacity to assemble information or generate plausible explanations. It requires curiosity, skepticism, disciplinary understanding, ethical judgment, and a willingness to confront uncertainty.
AI can expand the possibilities of investigation. It can reveal unfamiliar perspectives, support complex analyses, and make certain research activities more efficient. However, it cannot assume the researcher’s obligation to determine what is credible, what is justified, and what deserves to be communicated as knowledge.
Artificial assistance may help us produce answers. Epistemic responsibility requires us to understand why those answers matter and whether they should be trusted.
As AI becomes increasingly capable of performing research tasks on our behalf, this distinction becomes even more important.
The future of research should not be defined merely by how much intellectual work we can automate. It should also be defined by which intellectual responsibilities we choose to preserve.
Perhaps the question we should ask ourselves, as researchers and educators, is this:
If AI can increasingly perform the activities associated with research, how do we ensure as researchers to continue to develop, the intellectual capabilities that make those activities meaningful?
Answering this question requires more than statements of principle. It requires intentional practices that keep researchers intellectually engaged and in control of their investigative work.
A final note to my readers:
I write these essays as a form of independent public scholarship, developed outside my university responsibilities and grounded in my professional experience, research, and ongoing reflection.
If you find these essays useful or thought-provoking, you can support this work by subscribing to my LinkedIn newsletter, The Nautilus Navigator, sharing it with others, or Buying Me a Coffee.
Support is always optional. Reading, sharing, and engaging with these ideas are already meaningful ways of sustaining the conversation.
References
Correia, A.-P. (2026, September 18). When Assistance Becomes Dependence: The Risks of Over-Reliance on AI in Research and Education [Blog post]. Ana-Paula Correia’s Blog.
Universidade Católica Portuguesa. (2025). Política de integridade académica e científica: Carta de princípios no uso de IA.
Please cite this blog as:
Correia, A.-P. (2026, October 21). AI Can Assist Research, but Who Is Responsible for Knowing?[Blog post]. Ana-Paula Correia’s Blog. https://www.ana-paulacorreia.com/blog/ai-can-assist-research-but-who-is-responsible-for-knowing