The initial impact of generative AI is powerful, but learning to use it as a viable tool requires a more pragmatic mindset.
Becoming familiar with the rapidly-growing number of generative AI systems generally sees them share notable stages in user understanding and perception.
The first one we're going to call the "Awe Stage". In Awe Stage, the user experiences a feeling of elation as a few keystrokes, containing often somewhat random and unstructured requirements, yields seemingly impressive results.
"It did all this with only that? Wow!" is a rough approximation of what occurs, the computer equivalent of a magician pulling rabbits out of hats.
This is direct observation, resulting from acquaintances from a professional publishing background getting to grips with their first Large Language Model systems to experiment with textual content. It's exactly the same with generative AI image systems too - possibly even more so as those AIs can overwhelm with instant visual splendour.
Awe Stage continues for a while, perhaps some weeks of use, before the next stage: "Tainted Love".
You know the system is still good and very clever... but you start to see it has flaws and limitations. You only have to say "hands" to users of generative AI image systems to cause a facial twitch, and that doesn't even touch upon the broader general human limb issue that many seem to have.
Only this morning my colleague tested a random AI for a basic image of a very well-known footballer, and the outputted image had a fairly decent stab at likeness - decent, unless you count the addition of a random third leg. Perhaps that's why he scores so many goals.
The "hallucinations" of generative AI when producing textual content are another example, too often seen in the supremely plausible inaccuracies that have seen some generative AI labelled as "excellent liars". It won't take you many uses of some to start to question if that person really did say that quote about something.
From a user point of view, such incidents in this stage do leave you disappointed, felt even harder because it comes so soon after the Awe Stage. Like discovering your wonderful new partner doesn't get on with your cat or dog, you start to question the little things.
Yet the fact is, even given learning or limitations by the user, if the AI system is performing to requirements around 50% of the time, then users seem to stick with it and just adjust their usage and prompts to suit.
Because guess what? Humans, being the supremely adaptable creatures they are, find workarounds, add precision, learn tricks, and tell the system what not do, all to make sure the AI yields the results desired. So who's doing the learning with these Large Language Models then, really?
There's a reason Prompt Engineers exist as a profession in this digital age. If a person has the aptitude and dedication to immerse themselves in a generative AI system, their endeavours will genuinely yield better results. So which would you rather have: a good AI, or a clever engineer? Only one is an actual asset to you.
You could ask a similar question about your writers. Do you want to train your writers how not to write a libellous piece, breach a court reporting restriction, or get better exclusives? Or do you want to train someone else's AI?
But back to the AI - they are really then just the same as with any tool we use in publishing. Or any tool or instrument used anywhere else. The more you use it, the better you'll get with it.
That leads us neatly on the next stage, "The Production Stage". This is where the generative AI system is seen truly for what it is: a tool, to be used in the circumstances where it will save time, labour, and money.
It's not to be taken off the shelf for each and every content task, certainly not as such systems currently stand and particularly given legal battles over the use of copyrighted material in training the AIs, although initiatives such as that announced by Getty and Alamy will remedy that situation as they become more widespread.
Generative AI systems can and do have a place in publishing, but for now AI should not be allowed anywhere near the Publish button.