The world seems to agree that AI can do (almost) anything. But is that really true? More and more people experience that AI just doesn’t make sense everywhere – so can (and should) we use AI for everything? Or are its use cases limited after all?
AI fans and AI companies never tire of telling us that there’s nothing AI can’t do. Looking at text, however, it quickly becomes obvious that that’s not entirely true – the problems with AI-generated copy range from hallucinations and factually wrong statements to complete boo hockey and texts that use many, many words to say nothing at all. Sometimes the negative effects of that are bigger, sometimes smaller.
Regardless of the question if using AI really adds value in the individual cases, AI threatens more and more jobs – which makes many people turn to what seems like one of the last AI-safe havens: creative work.
And while AI companies and AI enthusiasts are undeterred and keep screaming from the rooftops that AI can do it all, more and more professionals from all kinds of areas take a stance against this. Putting it nicely, they say that AI output is, in their professional opinion, poor.
And I agree. Particularly when it comes to creative content, which is my specialty, but also beyond that – and there’s 3 reasons for that:
- AI is probability, and probability isn’t creative
- AI output always looks good – to laypeople
- AI (used correctly) saves little time in creative tasks
Let’s look at these in a little more detail, shall we? 🙂
1. AI is probability, and probability isn’t creative
The first fundamental problem I see (simplified, but still true) is actually very straightforward: The underyling principle of all current AI models is probability. That means that in any given case, AI will give you the most likely output. And while that obviously is a simplification of complex processes, the fact remains: What AI gives you is what’s most likely.
Sure, considering the ginormous mass of data used to train AI, there are countless probabilities and your prompts influence what’s the most likely output in any given scenario – but still, AI only ever gives you what’s most likely.
And as long as that’s the case, AI is only really suitable for tasks in which the most likely outcome makes sense – such as processes that are always exactly the same. It’s great at those and using AI really adds value there: Individual workflows can be made much more efficient that way and AI can take care of repetitive tasks for you so you only have to check the output quickly before using it.
But as soon as something’s not following the same pattern every time, should be unique or distinct from the mass, the limits of AI become evident.
In my experience errors big and small start to pile up in those cases. In the example of text, these range from incorrect grammar and the combination of elements which just don’t work together to the loss of any logic whatsoever. Considering that AI in its current form can’t think but only calculate probabilities, this isn’t surprising at all; and yet, people ignore this fact.
And even if nobody really knows what creativity really is, it’s very clearly linked to novelty. This novelty can take different forms such as a different perspective on something common, an unusual element that refreshens something well-established or something that’s really and truly unprecedented.
And even if you ask AI to combine different concepts that’s usually not exactly successful. In the vast majority of cases, the concepts just don’t work with each other. The reason for this is simple: AI lacks the general knowledge of the world we humans have and constantly use to evaluate whether an idea makes sense or not.
2. AI output always looks good – to laypeople
Another big problem is that laypeople usually use AI for areas they don’t know well – for tasks they’d usually outsource to a pro. That means that on the one hand, the prompts aren’t that great, and on the other hand, these laypeople can’t properly evaluate the AI output.
That might not feel like a big deal. Until we start to think about why we use AI to take care of tasks. When it comes to content creation, the goal is always to reach a certain goal: a blog, for example, usually has the goal to build trust or demonstrate know-how.
The problem is, however, that soul-less AI slosh usually doesn’t achieve these goals.
And still, the AI output usually looks good to the layperson, because they don’t know which aspects are important. I see that all the time with text since AI emerged: AI-generated copy lacks logic, doesn’t have a common thread that ties the text together, either repeats itself endlessly or contradicts itself and says very little with so, so many words.
Let’s call this laypeople bias, and I’ll tell you how I myself experienced it: For years now, I’ve been working quite closely with a graphic designer and therefore get deeper insights into design topics. And while I’m definitely no pro in that area, I can now evaluate design aspects I didn’t even notice before. When I would’ve said “not too shabby” about a design in the past, I now frequently see issues and can even name various problems.
This just goes to show that we can only understand the challenges and common problems of an area when we’ve really dove in. That’s when we start to know how things should be done and see when something’s not good. But when I have AI do something for me that I don’t have a clue about, how can I evaluate the output? 🤔
3. AI (used correctly) saves little to no time
The third and final argument against using AI for anything and everything is the fact that – if used correctly – AI doesn’t actually save you a whole lot of time (if any at all).
All we ever hear is how much time using AI saves and how much it improves efficiency. But what I experience in my work as a translator and copywriter is something completely different. Because as soon as you do more than give the AI engine a three-liner to have it write an entire blog article, you’ll soon realize that the AI process is actually quite time-consuming. Particularly when it comes to the preparation before using AI and then also for finalising the AI output.
Workflow: using AI in text creation
Ideally, you provide the AI with high-quality reference material, give crystal-clear instructions regarding brand voice, tonality, form of address, preferred vocabulary and more. In addition, you tell the AI what the main message of the blog article should be, which arguments it should make and in which order the information should be presented, and much more.
Which is, when you think about it, quite a bit of work to do before you even get to the AI.
And then, once you get your first draft from the AI, that usually needs countless corrections. Whether you do these with the help of AI or yourself is totally up to you – but either way, you need to plan enough time for this: Facts need to be checked, you need to evaluate whether the arguments follow a logical order and whether the tone of voice and terminology have been implemented correctly and consistently.
All these tasks that need to be done before and after the AI step actually make up the bulk of the effort and time required in content creation – the writing part itself is usually not that much of an issue (at least for a pro 😉).
As you can see, it doesn’t really save you much time if you actually use AI properly when you write something.
(Here’s more on my take on AI in copywriting.)
Workflow: Using AI in translation
And the same is true for translation: In our industry, AI is really, really hyped – but make no mistake, it’s not the people actually doing the work who love it, it’s customers and translation agencies. They praise AI as THE efficiency tool that FINALLY makes translation fast and cheap.
They simply ignore that this couldn’t be farther from the truth. Let me explain that. Because yes, the translation process itself can actually be done pretty quickly with AI compared to human translation – but here, too, the preparation required to get decent results (and sometimes not even that) is massive. AI translations contain tons of errors that can’t be ignored, which means AI translation always, always has to be proofread.
And I can’t even say how often I have to correct terminology in AI translations even if the AI did have the glossary available. Somehow, terminology still isn’t used consistently in the output. Or how often AI switches between the informal “du” and the formal “Sie” when addressing the reader in German, or how often it switches between singular in one sentences and plural in the next.
And I won’t even start with how incredibly boring AI translations are – (proof)reading those in bigger quantities is one of the biggest challenges I ever had to face! Like a lullaby, the endlessly repeated phrasings rock you to sleep, turning sustained focus into a pipe dream.
Which is a huge problem because wrong translations are so frequent in AI translations: Statements are turned into the opposite because the AI ignores a “not”, incorrect translations of individual words render the localized copy incomprehensible, and word-for-word translations aren’t completely wrong, but also simply not idiomatic in the target language.
Fact is: Proofreading AI-generated translations is extremely draining and time-consuming – but it’s paid worse than the proofreading of human translation “because the AI did all the work.”
Sounds unfair? Because it is.
(Here’s more on my take on AI in translation.)
4. So where can we use AI sensibly?
As you can see, using AI sensibly isn’t as easy the world would have us believe.
I frequently talk about this with a client from the tech sector, Simon Jiménez, and love his approach: The software engineer owns chax, a company that develops apps and owns the AI-assisted app for requirements engineering storywise. And es, you read that right: AI-assisted app! It’s not an AI app, it’s an app that uses AI for individual workflows, where it takes care of repetitive tasks and the human in the loop always checks the output.
I’m more than twice as fast when using AI – but that’s only true for about 5% of my work. With that in mind, we always have to evaluate carefully where we use AI in our workflows.
Simon Jiménez about how we can use AI sensibly and in a way that adds value.
And in creative work? I’ll be honest, I, too, use AI when creating text – but only for brainstorming. When I run into a wall or can’t tweak a headline enough to be happy with the result, I turn to AI. The engine then gives me 10 results, all of which are unusable.
But they do kickstart my brain and give me new ideas for a good solution.
And THAT is sensible use of AI in creative work.
