AI journey logs on experiments, learnings & insights

Twice this year the way I talk to an AI changed, and neither time had anything to do with the model getting smarter. The first was a dictation tool in the spring; the second was an evening in September, talking to a machine in real time with the wait taken out. Both times, what changed was the loop between us — and the next morning, at the keyboard, the same tool felt like filling in a form. This first half is about that loop: why we inherited it, why we named the thinking machine after a chat window, and why "friction" is hiding two different things, only one of which makes you think.

If the box is the wrong loop, why does the industry, when the box stalls, reach for points, streaks and praise? In this second half I go looking for the answer in the science of motivation and learning, and find that my own theory — that thinking and playing run on different chemistry — was wrong. What survived is better: one appetite, two ways of feeding it, and a borrowed drive that turns out to be agreement, not speed. Then I set out what a tool would have to do if its goal were not to keep you, but to make you need it less — and where my concept had to change to get there.

Every product has two halves: what you can do, and what the system says back. The first has been designed for forty years; the second was assembled by whoever held the code. Post 3 names that half — system response design — teaches its first rule from a red lamp in a control tower, and shows why AI tools just made it ten times harder.

Every AI tool built around a conversation now speaks from three places: a folded line, a panel beside the conversation, a row under the box you type in. The layout settled in two years by copying. What it did not settle is which of seven things in motion earns a light — and the confirm message shows what happens when nobody decides.

The AI assistant has moved from beside the document into it, in four stages nobody announced — and inside one session the tool changes kind several times an hour without saying so. Post 5 closes the series: what the switch removes from your day, the one line a tool owes you when it switches, and who is steering.

We’re taught that cognitive biases are glitches — little errors of reasoning we could catch if we just learned their names. But what if a bias isn’t a bug at all, but a defence: what your motivational system does when it’s protecting something it cares about, under pressure, as it meets the world? This is a deep dive into the machine beneath your mind — where your values actually come from (mostly not from you), why the person you show the world is never quite the person inside, and why the same architecture can produce a saint, a zealot, and a hollow high-achiever depending only on the state it’s in. And a harder question underneath: what happens to all of it when the machines we build start feeding the very needs that shape us? Part one of The machinery of bias — a five-part journey into the whole system.

A rumour starts with moving boxes outside an office — and months later it is a “fact” everyone somehow knows. We talk about cognitive bias as if it were a glitch in an individual head — a list of about two hundred named mistakes. But there is an entire production and distribution system for it: over a thousand catalogued parts, ten machines, three roles, one customer at the centre — and you live inside it. Part two of The machinery of bias, a five-part journey: the whole map at once — who creates, who reshapes, who routes, records and programs, and the one current of wanting that runs through it all. Before we take the engines apart one by one, come see the whole system.

Everyone can name three cognitive biases. Almost nobody can say where a single one is actually made. This is the factory-floor tour of the whole system: ten layers opened one by one — individual, social, public, synthetic, biology, media, economic, cultural-linguistic, institutional, knowledge. Two of them make biases. The rest make something the textbooks don’t have a word for — and at every machine, named products from its own shelf, so you can recognise its work in the wild. Part three of five. Come see where your biases actually come from.

A rumour can be stone dead where it started and still working, untouched, three layers away. This is the part where the machinery of bias starts moving: what a cascade is, the four pressure dimensions of a travelling claim, the six shapes, the alignment that makes one ignite — and the finding worth the whole series: damage is decided by the route, not the content. The rumour that opened this journey gets its ending, panel by panel, layer by layer. Part four of The machinery of bias. One question is left standing at the door: what could possibly stand in its way?

You cannot out-shout a system that eats shouting. The final part of The machinery of bias is the answer to the question the whole journey built toward: what actually stands in the way? The framework’s own records, read end to end, contain a map of the system’s brakes — real, documented, and unevenly placed: strong exactly where you live, absent exactly where your attention goes. Tactical levers for a cascade already running; strategic positions to hold before one arrives; the prediction discipline that turns reading into standing ready — and the one protection that installed itself while you were reading. The circle closes where it opened: I don’t know. We found out. Where will you stand?