AI has been understood as a technology that assists computers and machines in simulating human intelligence, reasoning, and problem solving. This comes with inherent assumptions of what human intelligence, reasoning, and problem-solving constitute, and which human’s standard of intelligence, reasoning, and problem-solving count. These assumptions affect the data that is collected and used to train an AI tool, and alter the path of the AI tool’s impact beyond the training data, when it is applied to the real world. Sentipensar, a principle that comes from Colombian popular culture and Latin American liberation thought, helps us reflect on the impact of these assumptions and design decisions, and offers an approach to govern this dynamic.
Before proceeding further, we want to note that Sentipensar is a concept that comes local, community-based traditions among Indigenous, Afro-descendant, farmer, and riverine/swamp fishing communities in Colombia. It was popularized by Orlando Fals Borda, and elaborated upon by Arturo Escobar and others who work in the tradition of decolonial thought and sumak kawsay. We engage with it as a challenge to the epistemology embedded in AI governance, that is, the way AI governance frameworks determine what counts as knowledge, rather than as a technique to be replicated. We also acknowledge the risk that sentipensar may become a branding exercise for empathy in AI design and caution against it. We welcome correction and deeper engagement.
Understanding Sentipensar
The term “Sentipensar” is a combination of two words – sentir, meaning to feel, and Pensar, meaning to think. The term effectively means “feel-think.” Instead of prioritizing one over the other, the term calls for a single, integrated act of feel-think where the knowledge that emotion and knowledge each carry arrive together as a unity. It is an epistemological practice where the way of knowing is grounded in the reality that emotion is not separated from reason. Rather, they are integrated, experienced, and exercised simultaneously. Most academic social science frameworks tend to systematically devalue emotional knowledge, or the knowledge that grief, joy, solidarity, and indignation carry, because they have inherited a Cartesian epistemology that treats affect as noise that contaminates the logical clarity of reason. Sentipensar revives the tradition of knowing that has been suppressed by Cartesian epistemology.
Sentipensar is also political epistemology, in that it refuses the separation between the knowing subject and the felt world, by calling for an integrated inclusion of objective analysis and situated experience. Any knowledge adequate to responding to problems faced by whole communities must be rooted in their relational and affective life, rather than divorced from it. Knowledge that is produced about communities from the outside in lacks this rootedness, no matter the technical sophistication it may be couched in.
In Sentipensar, a community is not a subject of research or object of study, but a participant in the production of data on their lives. They bring their felt knowledge of their lived realities alongside the analytical tools that academic researchers offer. The question of what matters, what problems deserve attention, and what solutions are worth pursuing, are not determined by external researchers. Instead, this knowledge emerges from the community itself.
Sentipensar is not an invitation to make AI governance more empathetic in the instrumental sense, of including more testimonials in impact assessments, or adding emotional language to consultation reports. It is an invitation to question whether the epistemic framework that underlies AI governance, its faith in the separation of reason from feeling as a condition of reliable knowledge produces the kind of knowledge adequate to governing technologies that shape how people live.
What Sentipensar unsettles in AI governance
AI governance frameworks are built on a rigorous separation between technical knowledge (what AI systems do) and values (what AI systems should do). Technical claims are assessed through empirical methods, such as benchmarks, audits, red-teaming, and statistical analysis. Value claims are addressed through consultation processes, ethical frameworks, and impact assessments that translate concerns into measurable criteria. This separation is considered a virtue, as it is seen to protect technical assessment from contamination by advocacy, and ensures that governance decisions are based on evidence rather than emotion. Sentipensar challenges this separation at its very root. It proposes that the severance of feeling from thinking does not produce more reliable knowledge but rather produces impoverished knowledge that systematically misses what communities most need governance to address.
AI systems that cause harm produce responses in affected communities, such as grief, rage, fear, and humiliation. These experiences also carry genuine epistemic content. The rage of a Black man wrongfully arrested because of facial recognition error is not noise contaminating an otherwise clear account of what the system did, but is actual knowledge about what the system did, what it means in a social context of racialized policing, and what kind of accountability would be adequate. AI governance frameworks that systematically route this knowledge through processes designed to translate it into quantifiable harm metrics are not making it more rigorous but are moving away from a comprehensive understanding of reality.
Sentipensar also challenges the methodological design of AI governance consultation processes. The standard model of soliciting inputs from affected communities through surveys, comment periods, and stakeholder workshops, analyzing their responses, and synthesizing them into policy-relevant categories reproduces separation and distinction. The community's felt knowledge is extracted, processed, and returned as rationalized policy preference, stripped of the emotional and relational context that gave it meaning in the first place. Sentipensar tells us to move away from this, to centering the people most affected as participants in defining the problem, framing the inquiry, and determining what ultimately counts as a solution. This is, in effect, co-governance.
Finally, Sentipensar also invites a question about the epistemic status of AI systems. Current debates on AI emotion and sentience are primarily philosophical and technical, particularly relating to whether AI systems can have genuine emotional states. However, Sentipensar asks a different question, namely: What kind of knowledge does a system that operates without integrated feeling-thinking produce? If separated thinking is epistemologically impoverished, then AI systems that model reasoning in the absence of anything like feeling will systematically produce the kinds of knowledge that impoverished thinking produces, that is, technically precise and contextually thin information that is inadequately suited to the human situations they are deployed to address.
References
Fals Borda, O., & Rahman, M. A. (1991). Action and knowledge: Breaking the monopoly with participatory action-research. https://www.sidalc.net/search/Record/KOHA-OAI-TEST:7365/Description
Escobar, A. (2020). Pluriversal politics: The real and the possible (Vol. 7). Durham, NC: Duke University Press.
Fals-Borda, O. (1979). El problema de cómo investigar la realidad para transformarla por la praxis. Ediciones Tercer Mundo.
Walsh, C. (2010). Development as Buen Vivir: Institutional arrangements and (de) colonial entanglements. Development, 53(1), 15-21.
This article draws from the wisdom and work of Indigenous people. While educating ourselves on Indigenous worldviews is important, we understand that our actions can also contribute to and enable appropriation. As part of our ongoing attempts at practicing accountability, we invite readers to consider donating to Indigenous groups, collectives, organizations, or initiatives to support their lives and work.
