The Flashlight and the Lantern

You've watched a leadership team stare at a dashboard that clearly showed the shift — and not one person changed their read. The data was lit perfectly. The room was dark.

There's a deep reason for that, and it's older than software. Every animal that has ever survived had to solve one problem: how to eat without being eaten. That requires two opposite kinds of attention at once — a narrow, committed beam to seize the target, and a broad, open vigilance for everything else: the predator, the change in weather, the thing that has never been seen before. Nature's answer was to divide the brain. The neuroscientist Iain McGilchrist has spent two decades documenting what each half delivers. The narrow-beam mode, in his words, "sees little, but what it does see seems clear. It is confident, tends to be black and white in its judgments, and jumps to conclusions." The broad mode is the reality check — he calls it, crudely, our bullshit detector.

Now name the two lights. AI is the most powerful flashlight ever made: a brilliant, narrow, controllable beam. Point it at a defined question and it illuminates the answer faster than any human could. But a flashlight only shows what it is aimed at. It cannot tell you which room you are standing in, what is moving at the edge of the dark, or whether you are pointed the right way at all.

The deeper risk is not machines becoming too intelligent. It is organizations becoming narrowly intelligent — brilliant inside the beam, blind to the room.

The wider, ambient awareness — the whole situation, not just the target — is the lantern. The lantern is contact with reality. And the era's problem is exactly this: the beam gets cheaper and brighter every month, while the lantern stays scarce.

McGilchrist's clinical findings read like a spec sheet for unmanaged AI. The narrow mode confabulates rather than admit it doesn't know. It "creates explanations and fills in gaps of information in order to build a cohesive story and extinguish doubt." Its job, as one researcher put it, is to create a model and maintain it at all costs. Swap "model" for "output" and you have every hallucinated citation, every confident wrong answer, every dashboard that explained away the customer who was leaving.

And if AI is the flashlight, automation is the conveyor belt beneath it — tireless, consistent, indifferent to whether what moves past belongs there. A bright beam over a blind belt is precisely how a broken process gets industrialized: seen clearly, moved quickly, never questioned.