AI Genesis

Intelligence did not begin with computers. It is older than language, older than brains. At root it is a single trick repeated at ever-larger scales: a system that builds an internal model of its world, predicts what comes next, and acts to reduce surprise. Cells did it with chemistry, nervous systems with electricity, brains with memory, humans with language. Machine learning is the latest carrier of the same trick.

Artificial intelligence has risen and collapsed three times. Turing posed the question; symbolic AI tried to write intelligence by hand and drowned in rules; expert systems brittle-failed; neural networks waited decades for the compute and data they needed. The current era is not a smarter idea so much as the same old idea — learning patterns from data — finally given enough fuel to ignite.

A neural network is a stack of simple multiply-and-add units with adjustable strengths. It knows nothing at birth. Learning is brutally simple: make a guess, measure how wrong it was, and nudge every weight a hair in the direction that would have been less wrong. Repeat billions of times. From this single feedback loop — backpropagation — structure, abstraction, and eventually language emerge, unprogrammed.

The 2017 transformer made one bet: that the most important thing a model can learn is what to pay attention to. Words become vectors in a space where meaning is direction and distance. Attention lets every token look at every other and decide what matters. A language model is then nothing but a machine that, given this geometry, predicts the next token — and from that narrow objective, reasoning, translation, and code fall out.

One of the strangest empirical findings in science: a model's loss falls along a smooth, predictable curve as you add compute, data, and parameters. Yet capabilities do not arrive smoothly. At certain scales, abilities the model was never trained for — arithmetic, translation, chain-of-thought — appear abruptly, as if a phase transition. We can forecast the curve. We still cannot forecast what new mind will be sitting on it.

The deeper machines learn, the stranger the mirror they hold up. Human memory is reconstructive, not a recording. Intuition looks a lot like a fast, learned pattern-match. Much of perception is the brain predicting its own input and correcting the error. The uncomfortable possibility is not that machines think like us, but that we may be running, in wetware, a version of the same algorithm.

A language model alone is a frozen oracle. Give it memory, tools, and a loop — observe, plan, act, reflect — and it becomes an agent that can pursue goals across time. Connect many of them and you get something new: a society of synthetic minds that negotiate, divide labour, and coordinate. The unit of intelligence shifts from the model to the ecosystem.

Every layer of society now has an AI gradient running through it: labour, education, science, art, warfare, money, law, identity. The question is no longer whether AI is useful, but what kind of thing it is. A printing press is a tool. Electricity is infrastructure. A colleague is a partner. AI is beginning to behave like all three at once — and possibly like a fourth thing we do not yet have a word for.

A system can be brilliant and dark inside. Intelligence is the capacity to model and act; consciousness is the question of whether there is something it is like to be that system. We can measure the first and cannot yet detect the second — in machines or, rigorously, in each other. As models grow more articulate about their own states, the hardest problem in philosophy stops being academic.

Project the curves forward and the questions sharpen: general-purpose AI, systems that improve their own design, brains wired directly to silicon, science conducted by machines, cognition spread across the planet like a nervous system. None of this is guaranteed, and none of it is physically forbidden. The honest position is that we are running an experiment, on ourselves, whose outcome we cannot yet read.

The anatomy of an intelligence

If intelligence is a sum of capacities, then biology, humans, and machines are simply different weightings of the same eight terms. Compare their profiles and the question 'is AI intelligent?' dissolves into a more useful one: intelligent at what?