One thing has become clearer to me as I've started using AI in a different way these past few weeks: AI lacks common sense.
Here's an example:
When The Angry Optimist switched publishing platforms, I used AI the whole time. (As I described last week.) At one point, a file needed to be imported from the old publishing platform to my web host, where it would be bundled into a package for the new platform.
But the web host blocked the upload, for reasons that were a bit unclear.
I asked ChatGPT to troubleshoot and find a solution.
After a while, it suggested I get a new web host.
That sounded like a drastic and expensive move. I don't know much about web hosting, but this seemed strange. So I asked if we couldn't just upload the file to Dropbox instead?
“Sure thing!” ChatGPT replied cheerfully.
So that's what we did, and it worked great.
That made the original suggestion almost inconceivably dumb.
I've run into several similar situations over the past few weeks.
Now that AI is doing everyday, practical things, these misses become more obvious. Of course they were there before too, but they don't show up the same way in a piece of text or an image.
The centaur's edge
This isn't a warning against using AI, but I want to once again make the case for being a centaur.
Half human, half machine.
It's in the mix of human and artificial abilities that the best results are found.
AI is fast, knows an enormous amount, and never gets tired, but it can get it into its head to pour a ton of work into a cumbersome and expensive detour.
Realizing that it actually is a detour takes common sense. That elusive concept. Because it's hard to put your finger on exactly what common sense is: you just know it, or feel it. That's probably why AI has a hard time with it.
Not always dumb as a rock
That's not to say AI is incapable of smart, simple solutions. Quite the opposite. There are plenty of those too.
One problem that came up during the move was that the topic tags disappeared from the articles when they were imported into Substack. So to fix it, you'd have to go through all 7,000 articles by hand and tag them. For a human, that would take about two months of full-time work, and even the AI would need several days.
Then ChatGPT figured out that it could build a little program on my computer that went through the articles, checked the tags on the old platform, and added them on the new one.
The whole job took several hours, but it ran in the background on my computer, and suddenly it was done.
A simple yet brilliant little solution.
AI keeps getting better, but the centaur will stay strong
AI has a lot more common sense today than it did four years ago, when ChatGPT was first released. That common sense, along with its other abilities, will get even better.
But human abilities will keep being important. Common sense, but also ideas, creativity, and especially drive.
AI can be inventive and creative, but only you have the things you've lived through, your knowledge, and your experience. It's when these get mixed with the machine's abilities that the really exciting stuff can happen.
Mathias Sundin
Angry Optimist