When is good enough, good enough?

A question for the gen AI age

This is a re-posting (and slight re-edit, though the sentiment remains) of an article I wrote on LinkedIn.

It’s not new. I published this there in November 2024. But given some of the posts I’ve seen on LinkedIn recently bemoaning the use of generative AI, I went back to it.

I still stand by every word, to be honest.

Anything we produce has constraints. At some point, good enough is good enough.

What promoted the article was a research report with the attention-grabbing title:

AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably

Intrigued, I dug into the research. One thing I noticed immediately was that the subjects involved in the study were defined as “non-expert poetry readers”. That was something that the rather clickbaity headline didn’t make clear.

The report clarifies “non-expert”: 90.4% of participants reported that they read poetry a few times per year or less, 55.8% described themselves as “not very familiar with poetry”.

It turns out that one of the reasons why non experts rate AI generated poetry higher than that created by humans is that it’s too perfect. The AI generated poems were exactly what people expect poems to be: atmospheric, rhyming, full of emotion, containing beautiful imagery, lyrical...in every one of 14 attributes we expect from a poem, AI-generated poems were rated either higher or equal to human-created ones.

Another nail in the coffin for humanity, some might claim. Or, “So do our minutes hasten to their end”, as Shakespeare might have said (and did write).

But worry not, Will. The attribute that makes human poetry so compelling is its flaws. It’s full of them. Much like we are. Turns out that true perfection has to be imperfect (as Noel Gallagher crooned, though I think he nicked that from the Sanskrit scriptures).

To be honest, however, this post isn’t about the research itself, but rather the thought process it prompted. It relates to where I think many people currently are on the generative AI journey. We’ll come to that, so bear with me.

Some people might criticise the research for using non-experts. “Of course they’ll be fooled by AI,” they might triumphantly cry. “Ask a poetry expert and they’d have no problem picking the generative AI from the genuine angst.”

“Sure,” I mused, on a dog walk, “but isn’t most stuff largely consumed by non-experts?”

And when I say “stuff”, I mean everything. Like art, music, film, books, food and – bringing it slightly closer to home – corporate communications and marketing content.

I like art, but I’m not an expert. I can look at a painting and tell you if I like it. But I wouldn’t claim to know whether it’s a “good” painting, from an artistic perspective.

Of course, in many of these areas the “expert” voices are often the loudest. These are the critics (as in the professional role, not just the world’s natural naysayers).

When restaurant critic Jay Rayner says, “there was an astringent foaming pond of what tasted like a parmesan cream surrounding an underboiled egg”, I might say, “could’ve done with another minute in the pan, but cheese and egg, what’s not to like?”

Alex Petridis might say a song, “marries a scrabbly guitar sample to a lurching beat, but it feels like a simulacrum of early 00s oddness, rather than a fresh embodiment of its spirit”, while I’m simply shouting, “absolute banger!” as I leap around my kitchen.

Gladiator 2 might be referred to as “so derivative of its predecessor, it’s practically a remake”, but I absolutely loved the first one, so no problem there for me.

You get the picture.

But what am I suggesting here? That we should be happy with work that’s just “good enough”?

Well, yes. Good enough for the designed purpose, at least. And, critically, good enough within the defined timeframe and resources.

I used to work with a design agency whose tagline was, “Good enough isn’t”. I get how that conveyed a sense that they wouldn’t be happy with work that was anything but as good enough as it could possibly be. But “good enough as it could possibly be” will always be constrained by time and money.

Some might thing that settling for “good enough” will lead to work that’s sub-standard.

Well, no. Because “sub-standard” is inherently relative to the standard. And standards are different for different work (produced within the relevant constraints, etc).

Sure, if Ridley Scott is handed $200m to make a film and he turns out a 30 second TikTok, the studio isn’t going to be very happy. And fair enough.

But if you’re creating a flyer for you daughter’s football team’s fundraising event, it’s perfectly OK that the photo was snapped on an iPhone by one of the parents. It’s good enough for the purpose.

Bringing this back to generative AI, we’re all currently experimenting. In many of those experiments we’re finding that the output of generative AI isn’t always good enough for the standard traditionally expected for what’s being produced.

“We’ve created a TV ad with generative AI. Check it out.”

“It’s not very good.”

“But it only took us a few hours to create and hardly cost anything.”

“I don’t care if it was quick and cheap, relative to the standard expected of TV ads, it’s poor.”

All that said, the proof of the pudding, of course, isn’t someone’s subjective view. It’s whether the TV ad is effective. Return-on-Investment is still a critical measure. A TV ad using generative AI that’s only 50% as effective as a traditional ad, but which cost 75% less to produce would be creating a better RoI.

I’m not sure where close to working out those metrics yet, but I guess my point is that the output of generative AI will, in some cases (perhaps many) be good enough for the purpose.

It’s a bit like stock photography. It’s not the best photography in the world, but it’s fit for many purposes. You’re not going to commission Annie Leibovitz to take some generic shots of a cityscape for the new brochure, you’re going to buy some for a few dollars from Adobe.

“Good enough” is not about settling for mediocrity; it’s about delivering work that meets an acceptable standard within the given constraints. By acknowledging the relative nature of quality and the practical limitations we all face, meaningful work can be created more efficiently.

I do think this is accepted in certain areas. Agile software development and minimum viable products are both largely based on launching something that’s “good enough”, knowing that it will be improved iteratively over time, based on feedback and performance data.

We should embrace “good enough”. In doing so, we open the door to innovation and continuous improvement, grounded in reality rather than unattainable ideals of perfection.

Generative AI is already “good enough” for many purposes, and it’s quickly getting better. We just need to be clear on the standards we want to set for different outputs, and equally clear on the time and resources required to create them.

Mulling this the other day, I figured that the only people who can afford (and certainly not in the financial sense) to pursue perfection are genuine artists. Many of those lived (and some still do) in abject poverty. Those who want to make a living from their art immediately find themselves dealing with the constraints of resources, deadlines, and client demands.

As Leonardo da Vinci once said: “Art is never finished, only abandoned.”

Or, I might suggest, deemed as good enough. By someone, at least, if not the artist themself.

Anyway, I’ll abandon this article here. As always, I’m interested in the views of others.