Every few months somebody tells me AI is about to replace human thinking. I don't buy it, and the numbers are most of the reason.
What is actually in there
The human brain has roughly 86 billion neurons. Not the hundred billion you usually hear — that figure got repeated for decades before anyone counted properly, and when a team finally did the count in 2009 it came out closer to 86.
Each of those neurons connects to thousands of others. Total connections land somewhere on the order of 100 trillion synapses, and every one of them is doing something.
Estimates of how many operations a brain performs per second range from about 10¹³ to 10¹⁸, depending on who is counting and what they are willing to call an operation. That is a spread of five orders of magnitude — a hundred-thousand-fold disagreement about a thing sitting inside every person reading this. We cannot even agree on the number. That should tell you how far off "replicating it" really is.
And it does all of that on about 20 watts. Dimmer than the bulb in your room. Training a frontier model, by comparison, runs into megawatts.
The part that actually matters
Raw throughput isn't the interesting bit, though. The interesting bit is that a brain is never finished.
A model gets trained, then frozen, then shipped. After that the weights stop moving — it is the same model on the ten-thousandth question as it was on the first. Brains don't work like that. Every time you learn something the physical wiring changes: synapses strengthen, weaken, form, disappear. You are running and retraining at the same time, continuously, with no downtime and no deployment window.
That is the closest thing to a perfect algorithm I can think of. One that keeps rewriting itself while it runs, on twenty watts, without ever being taken offline.
So what does "replace" mean
None of this means AI isn't useful. I built a study app on top of a language model and it does things I could not have done alone.
But "replace" assumes the two are doing the same job, and they aren't. A model is extraordinary at the thing it was trained on and blank outside it. A brain is worse at almost every individual task and better at the only one that really counts: turning itself into something that can handle a problem it has never seen before.
Until something can rewrite itself while it runs, on the power budget of a light bulb, replacement is the wrong word for whatever this is.