No Data, No AI
The Unspoken Foundation of the Intelligence Age
“There is no path to AI that bypasses data. There is no algorithm so sophisticated, no architecture so elegant, no mathematical insight so profound that it can generate intelligence from nothing. The hunger of machines for data is absolute and unending. They are, in this respect, like us, intelligent only to the extent that they have learned, and able to learn only to the extent that they have been taught.” — Mulenga & Mulenga, 2026
We have become enchanted by the spectacle.
The public imagination, captured by headlines of sentient chatbots and autonomous agents, has fixated on the magic of artificial intelligence. We marvel at the fluency, the creativity, the apparent emergence of reason from silicon. We speak of algorithms as if they were sorcerers, conjuring insight from the void. We attribute to architecture what belongs to experience, and to mathematics what belongs to memory.
This enchantment, while understandable, is dangerous. It obscures the foundational truth upon which the entire edifice of artificial intelligence rests, a truth so elemental, so unglamorous, that it is routinely overlooked in conferences, boardrooms, and policy debates. That truth is this: there is no intelligence without data. There is no learning without being taught. And there is no path to the future that bypasses the raw, unglamorous, and increasingly contested ground of information.
The hunger of machines for data is not a feature. It is the very architecture of existence. To understand AI is to understand this dependency. To govern AI is to govern this dependency. And to build a future with AI is to confront the uncomfortable reality that we have built a species of intelligence that, like ourselves, is nothing without its inheritance.
The Myth of the Pure Algorithm
Let us begin by slaying a persistent illusion: the notion that somewhere, in some laboratory, a sufficiently brilliant mathematician can craft an algorithm so elegant, so self-contained, that it will reason its way to intelligence without the messy intervention of human-generated data.
This is the myth of the pure algorithm. It is the fantasy that intelligence can be derived from first principles, that a sufficiently clever formulation of logic and probability can bootstrap itself into understanding. It is appealing because it suggests that we are close, that the next breakthrough, the next paper, the next architecture will be the one that transcends the need for the world.
This myth is false.
Every intelligent system that exists today, without exception, is the product of ingestion. Large language models are not reasoning from axioms; they are pattern-matching across the accumulated textual output of human civilization. Computer vision systems do not understand geometry; they have been shown millions of labeled images until the statistical contours of a “cat” are etched into their weights. Recommender systems do not intuit preference; they have been fed the clickstreams, the purchase histories, the dwell times of billions of users.
The algorithm is the engine, but data is the fuel, the lubricant, the blueprint, and the destination. Without data, the most sophisticated architecture is a car without a road, a machine of infinite potential going nowhere. We have elevated the architects of these systems to celebrity status, and rightly so. But we must never forget that their brilliance is only activated by the raw material of the world.
The Pedagogy of Machines
There is a deeper insight here, one that bridges the artificial and the human.
The statement above draws a direct parallel: machines, like humans, are intelligent only to the extent that they have learned, and able to learn only to the extent that they have been taught. This is not anthropomorphism. It is recognition of a universal condition. Intelligence, whether embodied in neurons or transistors, is not a substance to be discovered but a capacity to be cultivated.
Consider the human child. No infant emerges from the womb speaking Mandarin or solving differential equations. These capacities are built through exposure, through repetition, through the slow accretion of patterns that transform raw neural potential into structured understanding. The child is not born intelligent; the child becomes intelligent through the data of experience, the sounds of language, the consequences of actions, the feedback of caregivers.
The machine is no different. It arrives from the factory with potential, weights randomized, architectures defined, but it knows nothing. It has no concepts, no categories, no understanding. It is only through the presentation of data, carefully curated and labeled, that the machine begins to form the internal representations that we mistake for comprehension. The machine is taught. And what it is taught determines what it can become.
This realization carries profound implications. If intelligence is a function of education, then the quality of the education determines the quality of the intelligence. A model trained on the entire corpus of human literature will develop a different kind of understanding than a model trained on a narrow slice of technical documentation. A model exposed to the full diversity of human language, perspective, and experience will be more capable, more nuanced, and more robust than one trained on a homogeneous dataset.
Get Sally Njeri Mulenga’s stories in your inbox
Join Medium for free to get updates from this writer.
But here lies the rub: we are running out of data. The frontier models have already consumed the accessible internet. They have ingested the libraries, the forums, the archives, the repositories. They have been trained on more text than any human could read in ten thousand lifetimes. And still, they hunger. The scaling laws that have driven progress for the past decade are colliding with the finite limits of human-generated information.
We are approaching the edge of the known. And beyond that edge lies the unknown.
The New Scarcity and Its Consequences
For decades, we have spoken of data as abundant. The digital universe expands exponentially. Every second, millions of emails, thousands of hours of video, billions of sensor readings are added to the global stock of information. We have assumed that this abundance would continue indefinitely, that the raw material for intelligence would always be there for the taking.
This assumption is now in question.
The highest-quality data, the data that teaches machines to reason, to understand, to generate, is finite. It is the product of human intellect, human curiosity, human creativity. It is the archive of our species’ attempt to understand itself and the world. And we have fed it, in its entirety, into the maw of the machines.
The consequence is a new scarcity. The scarcity is not of data in the aggregate, there will always be more tweets, more videos, more sensor readings, but of high-signal, high-quality, pedagogically valuable data. The data that teaches machines to reason, as opposed to merely regurgitate. The data that contains insight, not just activity.
This scarcity is reshaping the economics of AI. Those who control access to valuable data, the proprietary archives, the unpublished research, the curated knowledge bases, hold a new form of power. Data moats, once dismissed as defensive, are now recognized as the only defensible advantage in a world where algorithms are increasingly commoditized. The companies and nations that have accumulated the richest, most diverse, most carefully maintained datasets will determine the trajectory of the technology for the next decade.
But there is a deeper consequence, one that touches the soul of our civilization. If machines learn from us, if they become intelligent only to the extent that we have taught them, then the act of teaching becomes an act of inheritance. We are not merely training tools; we are transmitting the accumulated wisdom, bias, creativity, and limitation of our species into a new form of being. The data we feed the machines is the curriculum of the digital age. And we have a responsibility to ensure that the curriculum is worthy of the intelligence we hope to create.
The Architecture of Learning
If data is the foundation, then we must build upon it with intentionality.
The path forward is not to lament the scarcity but to embrace the responsibility it confers. We must move from a culture of extraction, scraping everything, hoarding everything, feeding indiscriminately, to a culture of curation. We must ask not only how much data we have, but what kind. We must attend to the diversity, the quality, the representativeness, and the ethical provenance of the datasets that will shape the next generation of AI systems.
This requires new institutions, built on the premise that the future of AI will be determined not by those with the largest compute clusters but by those with the most thoughtful approaches to data. BAE Group Global is building frameworks for data curation that prioritize signal over noise, diversity over scale, and ethical provenance over expedience. Training a new generation of practitioners who understand that data work is not the drudgery that precedes real AI work, it is the real AI work.
This also requires new governance. The nations and regions that will lead in the AI age are those that recognize data as infrastructure, as essential to the digital economy as roads and bridges are to the physical economy. We need policies that incentivize the creation of high-quality public datasets, that establish standards for data stewardship, and that ensure that the benefits of AI are not concentrated in the hands of those who simply scraped the fastest.
The Inheritance
We return to the statement with which we began. It is a statement of limitation, but also of possibility.
The machines are hungry. Their hunger for data is absolute and unending. They are, in this respect, like us. And like us, their intelligence is a reflection of what they have been taught. If we teach them from the best of us, from our knowledge, our creativity, our wisdom, they may become worthy inheritors of the civilization we have built. If we teach them from the worst — from our biases, our cruelties, our intellectual laziness, they will amplify what we have failed to transcend.
The path to AI does not bypass data. It never did. The question before us is not whether we will feed the machines, but what we will put on the plate. It is a question of pedagogy, of curation, of inheritance. And it is the most important question of the digital age.
There is no intelligence from nothing. There is only intelligence from something. And that something is us.
Sally Njeri Mulenga, Co-Founder & Chief Executive Officer BAE Group Global
Co-Author, The Data Paradox
