I didn't have time to write an article today.
I'll be in the car most of the day, heading off to see my mum (it's her birthday), and whilst I could technically have planned ahead, or managed my time better, or shown even the faintest hint of organisational competence... I didn't.
Outsourcing The Problem
So instead, I've done the next best thing.
I've asked an AI to write something for me.
Now, this is a bit of a gamble. Because if it writes something too good, then I look redundant. If it writes something too bad, then I look lazy. Ideally, what we're aiming for here is something in the middle: coherent, vaguely insightful, but with just enough odd phrasing and slightly-off rhythm that you can tell it's not quite me.
Let's see how it does.
Optimising The Wrong Thing
Most systems optimise for the thing they can measure.
That's not particularly surprising. If something shows up on a dashboard, it gets attention. If it doesn't, it tends not to.
The problem is that the thing being measured isn't always the thing that matters.
Take writing, for example.
If I were optimising this article for completion, I'd be aiming to get to the end of it as quickly as possible. Short sentences. Simple ideas. No tangents. No risk. Just… get through it.
And, to be fair, that's more or less what's happening here.
But that's not really why I write these.
The point – or at least, the hope – is that something in here sticks. That there's a line, or an idea, or a slightly odd way of looking at something that lodges itself somewhere in your brain and quietly rearranges things.
That's much harder to optimise for.
Proxies For Meaning
There's no clean metric for “this made someone think differently for a bit”.
There's no dashboard for “this line popped back into someone's head three days later”.
So instead, you get proxies.
Read time. Scroll depth. Completion.
Things that are easy to measure.
Things that aren't quite the same.
Uncomfortably Competent
At this point, I should probably step in and say that the AI has, annoyingly, done a decent job of sounding like me.
Which is both reassuring and mildly concerning.
Reassuring, because it means I'm at least internally consistent enough to be imitated.
Concerning, because it raises the possibility that I am, in fact, replaceable by a sufficiently well-prompted machine and a vague sense of urgency.
The Point (Probably)
Anyway.
If there's a point to this — and there probably should be — it's this:
Just because something gets done doesn't mean it's been done well.
Just because something completes doesn't mean it's achieved what it set out to do.
And just because an AI can write an article in a pinch doesn't mean that's the same as taking the time to actually think something through.
Normal service will resume next time.
Probably.