Over the last few months, I've been experimenting with AI to build bits of software. Recently I've been using it to refresh the Panda website and my coaching website. And I've noticed a pattern which I am sure will be familiar.
I ask for something and it comes back almost right. I then go back and forth trying to refine it. Doubts arise. Sometimes I revert to a sketch pad. We get there in the end.
AI gives us an approximation of quality. It rarely hits the nail on the head. It's easiest to spot in things like writing, or animation, or interaction design, where we 'feel' it when something is working.
When we read a great writer or watch a great scene in a film, it's satisfying because it's exactly right. It doesn't almost work. It really works. So well that it reminds us what real quality is.
I don't know for sure, but my assumption is that the models have been trained on quality stuff and on mediocre stuff. So when they choose what to present to you, there will always be an element of the mediocre in it. A great writer, filmmaker, software developer, or skilled worker in any field is defined by how they've lifted themselves above mediocrity by consciously filtering it out.
So where am I going with this?
Well, I think there is a danger that a lot of people and organisations will settle for approximate quality. The first, seemingly good response is very hard to resist. And if you don't really care about what you're doing, you won't.
It gets worse when the person using AI couldn't do the task without it. They can't judge the quality, so they accept what they're given.
The result is an approximate world, where things look about right on the surface but don't bear close inspection. There's no soul in it.
Does that matter? I think we like to support other humans who are doing things with inspiration and courage. It's most obvious in the arts. But I think it's true in commerce too. It's part of why branding exists. We like to feel that another human is talking to us. Even if they're selling us shampoo.
So we need people who know what they're doing using these tools. People who can tell almost right from actually right. Ideally, they'd have a role in shaping the tools and would be rewarded for their input.
So when we're working with AI, the questions are: Is approximate quality enough here? Or do I need the real thing? Would I recognise it if I saw it?