AI: A 21st century labour-saving device

“It’s more of a labour-saving device than anything, though there’s still a fair bit of labour required.”

In the late 1950s, my great uncle bought a washing machine. It was a spectacle, probably the first in the town where he lived. Neighbours came to see it work. People from the next town came to see it. Some brought things to wash in it. It heralded the beginning of the end for wash days, a weekly day-long affair during which at least one member of the family was tasked with cleaning every stitch of clothing in the household. This involved boiling large vats of water, scrubbing, rinsing, running through a mangle and drying. Now, we were free to use that time as we pleased, no more washing. Or so we thought.

That was the promise, not the reality. I kept thinking of those halcyon days during a project earlier this year here at Big Motive. Somebody, not I, had the brave idea of using artificial intelligence (AI) to help improve reams of written content which a client had built up over the years. 

Today, it feels like the integration of AI into various sectors has become prevalent, normal even. What is also starting to feel normal are the enormously awful foul-ups being committed by AI. We didn’t check, but we assumed our client didn’t want us to weave swearing into their content. 

We proceeded with caution.

First, we chose our weapon. This was easy. There are currently two companies, OpenAI and Google, who offer users the chance to experiment on their AI tool for free, so we went with one of them. 

A reminder of our task, we wanted to create something which would make written content a little easier to read and understand. So we identified a few pages to experiment on, and off we went. 

Or off we didn’t, actually. It quickly became apparent that the free versions of AI chatbots were limited at best. They forgot things, they made mistakes, and they misinterpreted requests. It felt like a waste of time.

It seemed like we had two problems. The content samples we were using needed more than a few changes. It seemed like the chatbot was understanding some of our requests, not all. 

Those requests are called prompts. We learned that prompt engineering is a thing. Take this example:

“Make this shorter”; that’s an easy prompt, one which was easy to stumble upon. Imagine seventeen iterations along though, you’ll find yourself saying, “No you useless machine, that’s too short and you’ve lost the main point of the content. Go back, keep the meaning, just write it in UNDER 700 WORDS PLEASE”.

A change was needed. We switched from the free version of the AI chatbot and invested in a paid model. This allowed us to teach the chatbot, adding things to the ‘intelligence’ it already had. We built in a collection of rules which it must always follow. These rules applied to all content which the client had, mainly focused on keeping things short and simple. 

This, eventually, worked. I think we settled on the fourth iteration of the tool before letting it loose in the world. It still misbehaves from time to time, but correcting it is easier. 

So our first AI project is done and dusted, out in the wild fending for itself. What did we learn from it? Much. 

Firstly, AI is a labour-saving device and I cannot see how it will be anything else in the next century. 

It is not smart, it is well read. It has ingested the internet and bases its knowledge on that. Is that a wise thing to base your knowledge on? We suggest not. 

It is a blunt tool. Given its lack of common sense, it will follow your prompts come hell or high water. It is the robot that walks into the sea on your command. Knowing that makes it hard to rely on but easier to occasionally use. 

It is not about to replace any jobs, far from it. It feels like something that people will have to learn to use to make their life easier, like the internet. 

In summary, we don’t miss the washboard and mangle. The work will still need done, but now it can be done a little faster, and – hopefully – a little better.