The narrative around generative AI often emphasises its data-driven nature and algorithmic foundations. We’re reminded that large language models are “trained on vast datasets” using “sophisticated mathematical expressions”—descriptions suggesting precision and objectivity. But this framing obscures a simple truth: AI development is profoundly, inescapably human and subjective.
Think about the journey of creating a language model. Long before any code is written, human activities generate the data that will be collected. Which books get published? Which websites become popular? Whose voices matter? These societal decisions create the raw material used to train AI models.
When collecting data, subjective choices continue. Someone decides which sources are valuable. Someone determines how much text from different domains to include. Someone decides which languages to focus on. These aren’t objective choices, but value judgements about what constitutes “good” data sources.
Data cleaning also sounds objective, but it’s riddled with subjective calls: Which content is “toxic”? What’s a “duplicate” that should be removed? Where’s the line between helpful context and noise? Does a concept fall into this category, or that one?
Model architecture? More subjectivity. Developers choose algorithms, structure neural networks, and set hyperparameters based on intuition and specific goals. Even evaluation metrics reflect human values—we prioritise certain capabilities over others based on what we subjectively think is important.
Language itself is inherently subjective. When is formal language appropriate versus casual? Which words carry which connotations in which contexts? These questions have endless possible answers, which is what makes language so rich and fascinating.
And finally, once deployed, the human touch continues through interface design and marketing, shaping how users perceive, and work with, AI systems.
This human, subjective element, that’s implicated at every stage of the process, doesn’t diminish AI systems—it’s what makes them so remarkable. These language models reflect our collective values in ways we’re only starting to appreciate. They’re not just processing data; they’re processing human expression, agency, and creativity in all its complexity.
This is why we need more direct involvement from people studying philosophy and the humanities; language models and AI are too important to be left to the data scientists.
The next time you interact with an AI system, remember you’re engaging with a technology that’s profoundly human at its core—a mirror reflecting our collective intelligence enhanced through brilliant technical innovation.
AI isn’t some alien intelligence that emerged independently of us. It’s us, amplified and transformed through our own creative technological vision.