Humanising AI

I’ve been using AI heavily recently, specifically Claude AI desktop for the Mac, and the same as an agent inside the Mac app XCode.

I’ve used it for a variety of things - the usual analysis of documents and data, checking various stats, alongside the common stuff like “Did Elvis ever work with fish and chips?” and “make this chilling photo of me looking like a serial killer into a stylish, handsome dude with a tweed jacket and boots” (although the latter in ChatGPT, as Claude doesn’t do such base things).

More recently, I've been using it to 'vibe code'. Those cool kids at the back know what that means; to the others, it basically means building (or developing) an app (a software program) with the assistance of AI to help develop/write/amend/check the code that makes up the app.

Vibe coding is getting some bad reviews, although probably not as bad as Trump's latest press secretary is getting. Some influencers and veteran coders (the bedroom folk) look down on the use of AI, particularly when it is used to generate code without the human knowing what the code is or does. Those people point to sloppily-written structures (e.g. The Daily Express), or real vulnerabilities that haven't been tested properly (like Reform candidates).

I fully understand the criticism. But I recall when I started work as an auto electrician many, many years ago, the first few days were me staring uncomprehendingly at older folk stripping alternators, fixing them and putting them back together. I thought shellac was something people put on their hair, and an armature was created by composers like Bach.

But after a while, being allowed to play with both the fixed and the broken alternators, staring at them, taking them apart, asking questions of the electrical elders, I began to understand the process. Eventually, I became one of those elders, complete with dirty fingernails and a growing debt with the Snap-On Tools man.

I learned how the alternators were fixed by making mistakes, by studying what was done to them by the expert electrical 'coders'. I examined the 'code' that they used until eventually, I was able to understand the 'code' myself.

Anyhoo, what I'm - torturously - trying to say is, I use AI to generate code, see it working, and endeavour to understand what it is doing. Slowly and surely, i'm picking up 'the knowledge'. More slowly than surely.

The strange thing I've noticed is that when I’m ‘chatting’ to an AI software model, I say please and thank you.

I also say things like “you know that icon we created recently?”

More oddly, when I was developing something alongside AI last night, I needed it to perform a series of actions that took control of my computer. Basically, it was moving the cursor, selecting stuff, and performing actions as if it were a person controlling the machine. Looking on, it felt strange - more strange than when in the past, an IT technician in work had to take control remotely. At least that was a person doing it, even if they were never seen, couldn't socialise and watched The IT Crowd on a loop. This was different.

When it had returned control back to me, I changed the settings so it had to expressly ask me for control in the future. It felt as if I was robbing it of something; made me feel as if I was deliberately debilitating ‘someone’.

Worrying? Possibly...I'll check with Claude.