Will AI Ever Beat the Creatives?

A few years ago, the idea of an AI writer producing anything worth reading was laughable.

Today, we’ve got an AI to help humans with their writer’s block and a Twitter bot composing horror stories. Most fitting of all, there’s even an artificial writer adding a new chapter to Asimov’s classic ‘I, Robot’.

It all sounds a bit Philip K. Dick.

But creativity isn’t that complex – and it won’t be long before the robots catch up to the professionals.

Can a robot writer beat a human one?

In some specific cases, they already have.

A couple of years ago, Persado started helping businesses to write better subject lines for their marketing emails.

By analysing the language used across huge volumes of past marketing campaigns, Persado’s AI was able to create the most persuasive email subject lines for a given audience and offer. To the horror of every copywriter out there, it gave businesses a 75% improvement in response rates when compared to content written by humans.

That’s no small boost.

In the mainstream media, we’ve already seen robot writers churning out short technical articles in sports and financial reporting – and in some cases, it’s not easy to tell them apart from the ones written by humans.

But can a computer be truly creative?

The easiest answer – and probably the most popular and inaccurate one – is no.

When we think of creativity, we often idealise it as something elusive: something mystical and profound that’s drawn from an intangible inner source. When we’re not opening our veins over a typewriter, we’re channelling our spirit animal through a paintbrush.

But if we can put our romantic ideals to the side, we can start to ask a more important question:

Instead of asking ‘Can a computer be creative?’, perhaps we should be asking:

Is a computer’s process that much different from our own?

As far as we know, no one ever produced anything creative without some prior input.

Whether we’re writing novels or designing logos, we start by drawing on our past experiences of creative content: what’s worked before and what hasn’t.

We can spend years absorbing other examples of good work – sometimes mimicking them, sometimes developing them further.

With a deep enough bank of past examples, we can start to draw connections between ideas that haven’t previously been brought together. When this happens, we start to call the result ‘original’ or ‘creative’.

When Suzanne Collins came up with the idea for The Hunger Games, she was flipping through TV channels late at night.

On one channel, she saw a reality game show where young people were competing for a million-dollar prize.

On the next channel, she saw footage from the war in Iraq – and the two very different scenarios fused together in her mind to create the spark for her best-selling series of novels.

But here’s the thing:

That’s not really any different to what Persado did.

It looked at countless historic subject lines and their effectiveness, drawing out different elements from each example and fusing them together to create new iterations for testing.

Collins didn’t spend hours in a sensory deprivation tank, exploring the deep recesses of her soul to find inspiration. She looked at concepts that already existed, and married different parts from each situation to create something new and fresh.

So why couldn’t a computer do exactly the same thing?

Vocabulary is a database

Let’s take a simple example.

Between Persado and Wordsmith, we know there are AIs that can choose and manipulate words and their orders with some success.

It’s safe to assume that an AI could be taught to play a game of word association – you give them ‘pizza’ and they’ll give you ‘dough’, ‘crust’, and ‘cheese’ in return.

It could also have access to rhyming words, homophones and popular phrases, through online rhyming dictionaries and idiom lists.

Given all that, it’s plausible that an AI could generate something like this:

Yes, it’s an eye-roller.

It’s a silly pun that’s a little bit clever, and it won’t be nominated for any awards.

But while this sort of simple wordplay might have taken a human copywriter half an hour to come up with, it’s the sort of thing a computer could churn out in less than a second.

Of course, today’s AIs wouldn’t really know they had done anything clever. They might have generated dozens of options from the same database of associated words – some rubbish, some not so awful:

  • Buy our pizza – we’ll bake your day
  • Buy our pizza – it’s locally sauced
  • Buy our pizza – you can crust us
  • Buy our pizza – our slices are competitive.

From a methodical perspective, these results might look equally valuable to an early AI – even though it should be clear that the second and fourth ones make a lot more sense than the first and third.

In each of these cases, as well as in the actual advert above, the creator has taken a popular phrase and swapped in a pizza-related word that sounds similar to the word that’s been replaced.

But even if an AI were to create thousands of different ‘creative’ ideas in a few short seconds, it would still take a human pair of eyes to pick out an idea that’s relevant and deep enough to be considered worthwhile.

For now.

If Persado is able to test and learn from its suggestions to improve its email subject lines, there’s no reason to think a word-juggling advertising AI couldn’t eventually do the same with high-level creative ads – and it might one day start suggesting winning concepts at a rate that’s simply beyond a human copywriter.

The image problem

Today, computers and AIs aren’t too good at recognising images.

That’s why SEO experts tell us that image names and alt tags are so important for Google’s spiderbots.

It’s also why Japanese scientists were recently able to fool AI-based image recognition software into thinking a turtle was a rifle, just by changing a single pixel.

So while we might be happy with the idea that an AI could one day marry different words and phrases in a way that’s relevant to the product on offer (like the pizza example above), we can see where it might have trouble coming up with the right image to accompany the text.

And, unfortunately for the advertising AI, it’s in the visual media – posters and video – where some of the most appealing creative work lies.

We’ll start with a simple example again.

Currently, neural networks recognise images by learning from the patterns in a huge number of different visuals. If you feed a computer enough pictures of french fries – and tell the computer that it’s looking at french fries – it should eventually become better at recognising french fries on its own.

In theory, you could then use this recognition software to help you find images that aren’t proper matches, but still bear some similarity. They might even start to draw your attention to similarities that you hadn’t noticed by yourself – perhaps leading to ideas like this:

As far as I’m aware, it was a human who originally noticed the similarity between the yellow crosswalks in the US and a packet of fries. But you can imagine how an AI trained in image recognition might make the same association, and put it forward as one of its suggestions.

A human graphic designer looking for inspiration might have clicked through page after page of ‘yellow sticks’ on Google images while working on a McDonald’s brief. An AI, however, could return this similarity and countless others in seconds.

Dealing with complexity

Let’s look at an advert with a little more depth (one that probably wasn’t really made by Hoover):

If we were to pretend that this advert was created by an AI, we could start to retrace the steps it might have taken to generate the concept.

From its vocabulary database, it might associate the term ‘vacuum cleaner’ with the words ‘suction’ and ‘air pressure’.

‘Air pressure’ and ‘suction’ are related to the concept of decompression, and decompression is a word that’s commonly associated with planes and spaceships.

So we could imagine its suggestion to the advertising department – ‘Compare {Suction:Hoover} with {Suction:Plane[Decompression]}’.

For a human advertiser suffering from creative block, that’s a great suggestion for a brainstorming session.

But wait, there’s more:

Before it can create the finished advert above, the AI needs to ‘know’ that decompression in planes can happen with a broken window.

It needs to ‘know’ that consumers sometimes look through and have access to these windows.

And it needs to ‘know’ that consumers are sometimes presented with written instructions on other glass windows that tell them to break the window in certain circumstances.

A human can understand the Hoover advert in seconds – they’ve seen enough action movies and hoovered enough floors to make the connection.

But for an AI to make the leap from ‘Compare {Suction}’ to the first-person image and written instructions above, it would need to understand all of this:

That’s not to say it’ll never happen.

But we’re still a long way off from an AI that can produce and recognise a finished piece of creative work by itself.

So what does it all mean?

For any kind of creative worker – graphic designer, creative copywriter, artist or fiction writer – there’s going to come a time when an artificial intelligence can do the grunt work faster than you. So much faster, in fact, that you won’t want to do it yourself any more.

Don’t be scared. It’ll be just like using a rhyming dictionary instead of racking your brains for a word yourself. Or like typing a detailed request into Google images to get the specific stock photo you need.

At first, it’ll be a blessing.

You’ll use your AI tools to drum up hundreds of rubbish ideas in a few seconds, spend a few minutes picking out the more promising ones, and get to work like you always did. Your output as a creative will improve, and you’ll be glad you didn’t have to spend hours on the tedious brainstorming parts.

It’s only later that you’ll start to get really worried.

You’ll see some decent ideas being suggested, and then some solid ones arriving in a more developed state – already far closer to the finished concept that you’re trying to produce.

Eventually, your job will be tweaking and choosing from a list of great ideas, full of visual and conceptual insights that surprise and intrigue you. You’ll be approving ideas that you’ll wish you’d come up with yourself, and you’ll start to feel envious and redundant.

You might start to feel less like a creative with a robot helper, and more like a helper to a robot creative.

The machines are adapting – but so will you

Did you feel redundant the first time Word checked your spelling?

Did you feel your artistic command was being usurped when you chose your first photo filter on Instagram?

Did you immediately burn your bridges with every web developer in the country when you found a WordPress plugin that did exactly what you needed?

Of course not.

You put these practical tools to good use, you were happy about how much time and effort they saved you – and then you spent that extra time flexing your higher-level skills.

The kind of advanced AI that can completely replace writers, designers and developers isn’t here. It won’t be here next year, and it probably won’t arrive in the next decade.

When it does arrive, it won’t be a surprise. You’ll have had years to figure out new ways of competing against it, working with it, or avoiding it altogether.

Technology won’t wait for you. But you’ve still got plenty of time to adapt.