It’s been a question that has been joked about and whispered with worry across agencies for some time. You can’t quite teach a computer to be creative or to manage clients can you? When AI takes over the world, the careers that will surely survive most certainly must be the ‘thinking careers’. An AI can’t surely think of an award winning campaign like the Volkswagen Beetle campaign ‘Lemon’ could it? Well apparently there are quite a few people who dare say it definitely can.
While it may be happening in bits, across the creative spectrum, for now this vision is for most a moon-shot. Yet, just as NASA and its visionaries put men on the moon, there are tireless technology experts aiming at making machines more creative. The first step for a recipe of success is getting the ingredients ready.
AI can already help in automating repetitive tasks like research, pulling out historic campaigns or mixing and matching multiple factors to create thought starters and even help in the creative end product – like taking over video edits. AI has also made baby steps in becoming a bit more creative. Like being able to compose music on its own accord. Or create abstract art. We are building AI into the tools of our creative trade, like cameras and design apps. So the question is only a matter of time when an AI is able to develop its own creative aesthetic, built on the shoulders of every advertising and communication giant that has gone before.
Creative strokes that compute
So where do we begin? Perhaps with the world’s most popular AI platform. IBM Watson has already been credited with editing its first movie trailer back in 2016 for the horror film Morgan. Mostly a marketing stunt – an AI editing a movie trailer for a movie whose core theme is an evil AI. Brilliant. Maybe this is how the world ends and not Skynet. The real scary question is how does a super intelligence achieve this level of creativity to do a job in a day that takes humans weeks to achieve? And how long will it take for this to become the norm? How long will it take to teach an AI to learn by itself without having to feed reams of data of prior work ?
The answer is what the field of computational creativity – programming a machine to be creative, focusses on. Sure there are naysayers in droves. Apart from every creative who fears to become redundant (being an advertising creative after is one of the few fun commercially viable avenues for your creative streak), there have been early computer science pioneers as well. In fact this pessimism harks back to Ada Lovelace’s argument that if a machine can only do what it is programmed to do, how can its output be truly unique and creative?
A friend of mine who is a film script writer and director once woefully quipped that there really isn’t much originality in cinema left. While this may have been a tad too pessimistic, there is some truth in that every tale told, every artwork churned out or copy crafted is often influenced by something created earlier. All creativity has a benchmark of what has gone before. It is using this existing framework and connecting it in new ways to current cultural nuances that makes things anew. It is this that we need to teach an AI to understand.
What makes an AI different from any other creative professional is the amount of prior work that it can sponge up. It never tires. It works overtime without fatigue and without additional pay. And will eventually cost less in more ways than one. Using techniques like Deep Learning, combined with the ability to tap into historic trends with Big Data, finally executed with cherry picking influences from small data it can build a library of creative assets and a formula for future success. In short it can not only make the ingredients of tomorrow’s ads, but whip them up too.
How would an AI do it?
So let’s take the example of a movie trailer. Hypothetically here’s how I think an AI would probably have done the edit. It would first dive into historic data. Trailers for horror movie from time immemorial. From the first Frankenstein movie to the latest Alien prequel or IT sequel. Let’s assume that the AI can trawl the internet. Could it then compare successful trailers with other marketing collaterals such as posters and first week box office openings? Could it analyse and estimate the ideal duration of each trailer and possible winning shot break downs. Maybe build associations with cultural context based on articles drawn from newspapers. It can then analyse the entire duration of the movie and slice and edit the footage and compose its own sound track score to go along with each key shot. It then renders the final output in formats that are ready to deliver to your local cinema on approval.
Sound far-fetched? The raw computing power provided by cloud computing, new graphic cards with better computing cores, and deep learning combined could provide everyone access to an AI platform from their browser. The cherry on the cake is when the AI can sell the creative to humans, packaged with projected business results (ROI), plotted against best in class case studies and historic data. Making it more palatable to the business decision makers and much cheaper.
Robotics have the concept of the operational envelope – where the surroundings of the robot are optimised for the task at hand. Think of robots that manufacture cars. If you have ever been to a car factory you will notice all the components are within the robot’s reach. While we may start out adapting our existing agency framework to work with AI’s eventually we may flip the table and have the AI’s grow smart enough to adopt humans into their operational envelope.
But why aren’t we there yet?
The first step to making a computer think creatively is Deep Learning. But Deep Learning is yet to make the creative leap to Deep Reasoning that forms the secret sauce for creative execution. What does this mean? A computer can teach itself within a framework. Which is why a computer can teach itself to beat a human at chess (where there are structured rules). Even if there are no structured rules or a determined end game, a computer can weigh out the best choice and beat humans at a game of Go. To truly understand this feat you need to know that Go is a game played on a 19 X 19 grid with black and white pieces and that has more possible arrangement of pieces on the board than there are known atoms in this universe. While you can teach a machine to follow a set of rules and power driverless cars with Deep Learning, you need more to make masterpieces.
The grail is Unsupervised Learning that can help Creative AI’s to analyse small amounts of available data to make creative leaps with the collective knowledge of a 100 year old creative director. So what we are saying is an AI that can intuitively understand understand an ad crafted by David Ogilvy vs one by Bill Bernbach without being force fed or served every ad ever written by these two doyens of Ad Land. There is no need to physically tagging or sorting the data before providing it to the AI as such. Making creative connections, and interpreting context after all is at the core of how agency creative folk work at the end of the day. The problem today is the amount of data you need to feed the AI. If it takes 15 million images to teach an AI to identify an object, imagine how many ads you are going to have to feed the AI to craft meaningful copy and art work. Hence the quest for an AI that gets the job done with so called small data. The AI pundits are confident that this is achievable.
Why we should be creatively optimistic for the future.
In the immediate future AI will be a tool that empowers Creative Directors, not replace them. There is no doubt of this. Put in a brief and an AI will give you a palette of campaigns and executions that have come before that have set standards in creativity and that are on budget. But merely as thought starters. Perhaps instead of campaigns it could pull relevant news articles to garner consumer insights and map this against internet usage data that is in the public domain. An AI may even be able to proof final artwork for grammar, errors and match it against media specs. I wouldn’t be surprised if Publicis’s Marcel can be trained to do this already. An AI that helps you build the creative decks that are sold in to clients.
The next step would potentially be to create possible execution iterations and having a creative team tweak these and fine tune them. An AI will earn its place as the fourth wheel in the creative team after the copywriter, art director and full stack developer in a few years down the line. A tireless indispensable helping hand.
Creative AI will in the future become an agency’s proprietary IP. The stuff that agencies are built around. You will have every holding company and independent crafting their own iteration. On the other end of the scale there will be open source AI platforms that are used or rented by the little shops. Just so that people can keep pace and the playing field is fair.
There will also be offshoots to this. I imagine this will lead to an unprecedented demand of historic advertising and cultural data along with the kind of assets that are going into ads. Think about it. Everyone from Cannes to the Effies and Spikes, could charge royalty for all of their archives of winning campaigns and shortlists, delivered in an easy to access API format that AI’s can interface with. Then there are the stock image, video and audio sites which could provide API’s to search and purchase relevant assets from their databases and integrate it seamlessly into the creative engine.
Fuelling the rocket ship.
There are already companies trying to sell AI powered tools for better creative work. A google search churns up things like Bidalgo. Want to think bigger? How about Pencil. Touted to be the world’s first Creative AI that can craft personalized ad campaigns that originates in culture and multiplies the art and effectiveness of online advertising. I kid you not. Most of that is verbatim from their website. An AI powered agency (that still has humans working on the creative nonetheless). While aiming for the moon, the company’s current focus is using AI to help in crafting personalised creatives. A simpler way for saying that it is going to marry personalisation and programmatic advertising techniques to benefit its clientele. I could be mistaken.
Time for a change. An agency powered by AI.
If for a minute we step back and say that a world with advertising AI is a given, what would it look like? Can AI be the bridge between agencies and their woes of today? A facilitator for potentially a leaner and effective creative team. The ability to carry some weight while letting the team get on with being more creative and have more of a creative life while at it. Another tool in their arsenal like social analytics and listening tools today. There will be growing demand for people who know how to first build such tools, then teach creatives to use these tools and finally creatives who get the best out of these tools.
Pitches we can hope will become radically streamlined. With different agency AI battling it out in micro-seconds in a market place to solve a brief and serve sample creative outlines for pitch consideration. Clients can then shortlist to see what humans can do in round two. Reducing pitch turnarounds and effectiveness while using historic data in optimising agency decisions whether they need to even pitch in the first place for an account based on business decisions and a client’s past track record with its previous agencies.
The timing to birth such an AI is probably right. Ad Land is seeing a decrease in the asking price of its creative output, with creative teams being asked to churn out twice the output with no real gains in remuneration. Frenemies are at the door just begging to get in on the action. Whether it is digital behemoths or the consulting companies that can bring more clout and resources to the table any day and have a unique brand identity advantage at a time of agency distrust. Ken Auletta’s book is a gospel to be preached amongst the naysayers who think all is well in Ad Land.
Perhaps it’s an open secret that the only way to get around Facebook and Google monopolies with their pretty walled gardens and data black boxes is to currently assail them using AIs to rapidly churn, iterate and hurl creative iterations at these two platforms until they are perfectly optimised. Until an AI can figure out what works best. A brute force attack. A parallel to draw is how stock trading is done these days using computer algorithms.
The final hoorah.
Whatever side of the fence you sit on whether AI is a creative boon or bane, the truth is there will be a time in the not so distant future where you get to not only answer the question, but live it. Whether the world sees you as an AI empowered creative or a traditionalist. Both will have takers.
The first group may tackle volumes of creative work like Ford production line, or have more time and tools to really come up with great ideas.. Clients will love them because they will be faster and collectively better than before, not to mention much, much cheaper. Even from large agencies.
On the other hand AI may drive the market up for human creative output. Making it more niche, expensive and hard to come by. Like a custom built sports car today. These will be fast fading creatives who have racked up hours of creative experience.