Mark White

What would it take for AI to replace creatives?

gretaWhat would it take for AI to replace creatives?

Could AI replace creatives? Exploring the nature of creativity, methods of training Artificial Intelligence, and revealing 7 ways that AI can make marketers more effective.


A recent Gallup poll found that over 73% of Americans believe that AI will be a net job killer. With a great deal of industry buzz, we ask does it spell the beginning of the end for creative roles? And how can marketers apply AI to their advantage?


Could AI replace creative roles?

Great creative always involves the intersection of creativity, culture and emotion.

  • Creativity is the creation of novel or improbable combinations of familiar ideas – a great example is the idea of tasting a rainbow that made up the famous ‘Taste the Rainbow’ Skittles campaign.

  • Culture is made up of shared ideas, customs and social behaviour. Examples are experiences of love and break-up and the entirety of language and slang.

  • Emotion in this context are the feelings which are represented in and most importantly evoked in the viewer by a particular piece of creative.


A great example of these three coming together is the ‘Like a Girl’ campaign from Always. It uses part of the cultural phrase ‘run like a girl’ in a novel context and is very emotionally powerful. 

 

Creativity, Culture, Emotion: how could AI approach these three facets of creativity?

AI is capable of combining elements in ways that are novel to the human eye - for example the AlphaGo AI performed moves that weren't obvious to skilled Go players, winning an award from the Korea Baduk Association for displaying ‘creative’ skills. There are interesting explorations on the theme of more complex creativity involving imagination in the book Pragmatic Imagination.

Culture and emotion are inextricable

However to elevate from mere novelty to creativity requires being able to discriminate good from bad from great. Whilst it may be reasonably straightforward to judge a particular move as being great in the context of a board game, it is much more difficult to evaluate ideas in the context of culture and emotion.

We understand emotion because we have emotions ourselves - when we see emotions in others, parts of our brain (primarily the limbic 'mammalian' brain and also the much hyped ‘mirror neurons’) fire in sympathy. Human brains are all structured alike due to our common DNA and as we all grew up in broadly similar environments, therefore we have similar connections that fire when we see content that evokes emotion. Emotions aren’t purely objective concepts in the way that gravity is, and can't be understood without reference to animal/human biology - i.e. a person who has never felt anxious wouldn't be able to really know what feeling anxious feels like, no matter how much description of the feeling they are given.

Likewise, culture is based on the many complex aspects of our environment – taking the example of the ‘Like a Girl’ campaign, the core part being the shared experience of hearing that phrase used on the school playground. Many of these aspects of culture also rely on a complex web of relationships with other parts of culture - just like how many of the references in an episode of Family Guy don’t have any meaning without the cultural knowledge of everything it is subverting. Whilst the meaning of any piece of culture is different to every individual because we all have unique life experience and exposure to culture, nevertheless there are shared intersections, which are at the heart of communication.

There is also a physical as well as cognitive element - seeing all the tricks in the iconic Nike Freestyle ad of 2001 is especially powerful because we all have the direct experience of how it feels to handle a basketball and therefore can relate on a physical level to how difficult the tricks are.



The hoover, the scanner, and the school: what would it take for AI to make truly great creative?

For AI to be able to create great content, we would therefore need to find a way to train it with a deep understanding of culture and emotion, including the way these are processed by the human brain – i.e. whilst it doesn’t need its own limbic system, it does at least need a reasonably accurate limbic system model, and a good enough knowledge of culture and how culture is perceived. There are three theoretical strategies available to build this training data.

  1. Hoover up as much online content as possible (Wikipedia, YouTube videos, etc) and assume that knowledge and a model of culture and emotion can be inferred from the content.

  2. Use some sort of brain scanner to take a dump of all knowledge and structures within human brains

  3. Create an approximate model of the newly-born human brain and take it through all the experiences a real human would go through on its way from being born to becoming an adult - ie literally parent it, put it through school, etc. This was called the ‘Child Machine’ by the father of Computer Science, Alan Turing.


1. The online content hoover solution

There are many challenges with attempting to build a model by hoovering up online content. It is an understatement to say that it’s extremely hard for algorithms to extract deep meaning from unstructured online content. Most information online is expressed using language, which is generally an extremely abbreviated way of communication compared to the richness of within the mind of the communicator. Why does it take Proust more than a million words to communicate his thoughts on loss of time? (And even then they’re very difficult for even a very literate person to understand)...

Most advances in the last couple of decades have been around deep learning which is good at pattern matching, but is very poor at actually understanding the underlying concepts and relationships, gaining a grounding in reality and understanding cause and effect. Progress in these critical areas is far, far slower than we’ve seen at tasks like pattern matching tasks like image recognition. It is also missing the key learning technique of interacting with the world (how does a baby learn gravity? By picking things up and dropping them. How do people learn courtship? By going on dates …). And even if all this were solved, we’d still need to assume that all the information required to build a cultural and emotional model is actually available on the internet or other accessible data sources.

2. The brain scanner solution

A brain dump from a human brain would do the trick. But our ability to ‘scan’ the brain is still very rudimentary. For example fMRI’s maximum level of detail can only capture groups of tens of thousands of neurons (never mind capturing the average 7,000 synapses per neuron), and can’t capture changes at much better than a one second accuracy.

Whilst there have been advances in more invasive methods such as Elon Musk’s Neuralink, these are also relatively primitive compared to the scale of the brain and far from being able to provide a complete picture. Adding to this, brain neurons are much, much more complex than those in artificial neural networks and we are still nowhere near a full understanding of how a single human neuron works. We’re therefore decades at the very least from being able to extract the contents of a brain.

3. The ‘Child Machine’ back-to-school solution

And naturally, bringing up an AI as a human with the full range of human relationships and experiences without it being treated differently is impossible (and not to mention unethical). As Alan Turing himself said in 1950 “One could not send the creature to school without the other children making excessive fun of it”. And this ignores the fact that even at birth, the brain is far from a blank slate.

So could AI unseat our creative minds?

Weighing up these possibilities in the short to medium term, it is highly unlikely that AI can develop emotionally resonant creative.

Without a reasonably accurate model of emotion and culture, it would be impossible for an AI to have an understanding of what would create a particular emotional response, and without that it is correspondingly impossible for it to create great creative content.


That being said, how can AI help marketers be more effective?

Here are a few of the many areas in marketing where AI can be a hugely powerful tool that marketers can leverage to make themselves more effective and efficient.

AI can analyse data to inform marketing strategy
Examples include textual analysis to help pull insights buried in survey data; looking at correlations between buying patterns and weather (particularly valuable for seasonal products); and identifying patterns in trending hashtags in social media.

AI can save time and drive efficiency
AI can be used to generate and assess content variations, for example, the Phrasee tool will generate many email subject lines and make an assessment of which would work best for a given audience.

AI can inspire creative
It can be used for creative inspiration – a good example was IBM’s use of AI to identify style elements around Gaudi’s work to provide inspiration for a sculpture at Mobile World Congress in 2017.


AI can enhance creative production

AI can help creatives bring to life concepts that previously wouldn’t have been possible. For example creating the deepfakes being used to bring dead actors and artists back to life, style transfer where the style of an artist is automatically applied to a piece of visual content, for example, making a photo look like a Van Gogh painting, or automation of labour-intensive tasks such as rotoscoping and compositing.


AI can refine ad targeting

It can be used to provide better ad targeting to ensure marketing messages will be presented to those who would find them most relevant – i.e. the algorithms used by Google and Facebook that model preferences of their users in order to inform which ads they are presented with.

AI can ensure consistency across marketing messages
AI tools can be used to analyse marketing content to make sure that it visually/textually matches brand guidelines and ensure consistency of look and messaging in a world where companies are having to make very frequent communications across many different channels.

AI can help us to understand audiences on a cultural level, identify areas of growth, develop culturally relevant content and create the basis for creative alignment across marketing teams.

At Codec we’ve built a tool that uses AI to perform real-time analysis of hundreds of millions of online content interactions, to enable brands to connect to their audiences with cultural relevance.

The AI identifies the audience networks that resonate with a brand, enriching existing demographic datasets, and highlighting potential growth audiences. We also use machine learning for natural language processing to analyse the personalities of their audiences, enabling brands to build content in a tone of voice and style that resonates with them.


Conclusion

The last few years have seen a lot of hype around AI and marketing. Some of this is based on solid foundations that reflect the great potential of emerging technologies, but the suggestion that AI will make marketers themselves redundant is definitely a step too far. 

Marketers should be aware of the many ways that AI technologies can be used as a tool to lever their skills, to understand their audience better, to produce more relevant, better targeted campaigns and open up new creative possibilities. But they don’t need to worry about their core creative and strategic competencies being automated or replaced any time soon.


Mark White, CTO, Codec
Mark was previously Principal Architect at Expedia and is now CTO at Codec. He is fascinated in how psychology, linguistics, neuroscience, philosophy, psychology come together in the field of AI, and how that can be used to empower creativity.


Discover your brand's unique voice in culture through our AI-driven audience intelligence. Get in touch to speak to one of our strategists.

Next post