AI is changing everything. But too often, the conversation happens in a language most people were never invited to speak. This lexicon aims to bring AI for everyone. No prior knowledge required. Just curiosity in a reality that is rapidly changing.
AI ethics is the study of what is right and wrong when it comes to building and using AI. It asks questions like: is this fair? Who does this harm? Who benefits? Who decides? Think of it as the conscience of the technology industry, asking the questions that pure engineering does not.
AI governance is about who makes the rules for AI, and how those rules are enforced. Just like a city has laws, traffic signals, and courts to keep things running fairly, AI governance is the system of rules and institutions that is supposed to keep AI development fair, safe, and accountable. Right now, that system is still being built.
Algorithmic bias happens when an AI system treats some people unfairly, usually without anyone intending it. For example, a hiring tool might favour men because it was trained on historical data from when most hires were men. The system learned a pattern, but that pattern was a reflection of old inequality, not a neutral truth.
Ayni is a word from the Quechua people of the Andes. It means sacred reciprocity: what you give to the world, the world gives back. It is not just a nice idea. For Andean communities, it is the principle that holds relationships, nature, and community together. It is also the principle that holds AyniAI together.
Epistemic humility means knowing that you do not know everything, and that where you come from shapes what you see. In AI, it means recognising that every system is built from someone's perspective. The data it learns from, the problems it is designed to solve, the people who built it: all of these leave a mark. Humility is acknowledging that mark exists.
Epistemic justice is about who gets to be taken seriously as a source of knowledge. Historically, many communities have been treated as subjects to study rather than experts in their own lives. Epistemic justice says: everyone has knowledge worth listening to. In AI, this means the people most affected by a system should have a real voice in shaping it.
Human-centred AI simply means designing technology around people, not the other way around. Instead of asking "what can this system do?" it starts by asking "what do people actually need?" and "will this make their lives better?" It sounds obvious, but it is rarer than you would think.
Indigenous epistemology refers to the ways of knowing that indigenous communities have developed over thousands of years. This includes how they observe nature, make decisions, pass on knowledge, and understand relationships between humans and the world. These are not myths or folklore. They are rigorous systems of knowledge that have kept communities and ecosystems alive for millennia.
Mutual aid is the practice of people coming together to meet each other's needs, reciprocally. Everyone gives and receives based on what they can offer and what they need. Peter Kropotkin argued that cooperation is the collaboration gene that allowed humans not merely to survive but to build culture, knowledge, and collective flourishing. Where competition selects for dominance, mutual aid selects for thriving.
Imagine a company says "we will treat all users fairly." Operationalisation is the work of figuring out what that actually means in practice what rules the system follows, how principles are measured, and who checks that it is really happening. Without this step, good intentions stay on paper.
According to Thilo Hagendorf, the operationalisation gap is the gap between ethical principles and their real operationalisation. Organisations articulate values, publish frameworks, and commit to Responsible AI but the principles rarely translate into concrete technical decisions, design choices, or accountable processes. The gap is not a failure of intention. It is a failure of implementation.
Most AI systems are built on one idea of what is smart, fair, or useful. Pluriversal AI challenges that. It says: there is not one world, there are many. There is not one way of knowing, there are many. And AI should be built to work across all of them, not just the ones the designers knew about.
Pluriversality is the idea that there is not one single correct way to live, think, or organise society. Many different worlds, cultures, and knowledge systems coexist, and each is valid on its own terms. It comes from a famous phrase: "a world where many worlds fit." Applied to AI, it asks us to stop building technology as if everyone thinks, lives, and needs the same things.
Governments and companies write rules about how AI should behave. But writing a rule and following it are two very different things. Policy operationalisation is the work of making sure those rules actually show up in how the technology works day to day. It is where the real challenge begins.
Responsible AI means building technology that actually considers the people who will be affected by it, not just the people who build it. It asks: does this system cause harm? Is it fair? Can people understand and challenge it? It is the difference between building AI fast and building AI well.
Technological imagination is the ability to picture a future that does not yet exist. Not just a faster version of today, but something genuinely different. Right now, most conversations about AI are driven by fear of what could go wrong. Technological imagination asks a different question: what could go beautifully right, if we were brave enough to build it?