This story was first published in The Aleph Report. If you want to read the latest reports, please subscribe to our newsletter and our Twitter.
How many of you believe that Artificial Intelligence (AI) will automate most jobs in 10 years?”
Around 90% of the room raised their hands.
“How many of you believe AI will replace YOUR job in 10 years?”
Around 15% of the room raised their hands.
This is a typical response. We all understand the theory, but it rarely applies to us. Artificial Intelligence keeps getting trashed-talked in many industries. Few appreciate how fast it’s developing and how rapid its iterations are.
One area that seems to feel immune to automation is creativity. The consensus is that AIs will substitute repetitive, highly specialized tasks. Those that require holistic thinking or high degrees of creativity will be spared. I beg to differ.
“Our model predicts that the second wave of computerization will mainly depend on overcoming the engineering bottlenecks related to creative and social intelligence.”
The Future Of Employment: How Susceptible Are Jobs To Computerisation?
AI systems are improving at an incredible pace. We already have Deep Learning systems that are capable of doing music improvisation, artistic image styling or creative new images.
So far though, such technologies are lacking internal coherence. Yes, an AI can compose an improvised musical piece. The piece though doesn’t have a purposeful intention or connection. The same happens for images. An extreme example of this would be the art of writing. Not only does it involves creativity, but it requires social intelligence, planning and plot consistency too.
Automating the creative process
Can creativity be automated? The creativity process usually follows three well-known steps. The learning & research stage, the development of structure or form and the creative freedom or breaking of the form.
The automation of creativity is following a similar path than in humans.
Before we can even start creating we need to learn and gather information. The help of a teacher or mentor is critical at this stage. They can help students focus on what and how to study. They create a template or roadmap for the students to follow.
That’s precisely what the first generation of automated tools is already doing. Human operators design written templates that the machine then uses to create written reports. From automated reporting of The Washington Post’s Heliograf to computerized financial statements written by Quill. These systems though, require human editors to create these templates. Their outputs are scalable and informative, but it would be a stretch to call them creative.
“Instead of targeting a big audience with a small number of labor-intensive human-written stories, Heliograf can target many small audiences with a huge number of automated stories about niche or local topics. There may not be a wide audience for stories about the race for the Iowa 4th, but there is some audience, and, with local news outlets floundering, the Post can tap it. “It’s the Bezos concept of the Everything Store,” says Shailesh Prakash, CIO, and VP of digital product development at the Post.
The danger though lies in discarding these early approaches and the potential to improve. They’re obviously pretty limiting but are the foundation for more comprehensive automated efforts. Ignoring them as failed creative attempts is not to understand how disruption works.
The second stage of the creative process is the personalized copy. You copy and repeat what others have done and add, progressively, your flavour to it. On this stage, you can already perceive a sense of purpose. There is a goal, an intention to the creation. It can be informative or mere emotional expression, but it’s driven by something.
A whole set of AI tools are trying to mimic this stage. Given a topic or a subtopic, AI agents research and write thousands of variations on the given theme. It’s impressive to see the machine produce all that content. Nonetheless, these algorithms rely on the existence of available written material on the topic. Writings that exist because a human wrote them. Again, simple automation, rewriting, summarizing, but no creating.
While most AI systems lack a sense of purpose or a reason for their creative traits, we should meditate carefully on our creative impulses.
Humans create, not for the sake of creating, but because of need. This need can be driven by our egos or for other reasons like self-expression, identity fixation or personal therapy. My point though is that it’s driven by a desire.
I wonder if some of the quirks we observe in AI systems aren’t the machine’s creative ways of dealing with their own needs. Google’s translation system produced its interlingua language to deal with translations. Why do we discard that as non-creative?
Not long ago, Facebook AI agents developed their language to communicate with one another. The goal was to achieve a negotiation between two agents. The researchers forgot to incentivize the use of the human word and what the AI came out with was its language. They decided to shut it down. Why is it that the only creativity that is valid is the one we humans can understand?
The third and final stage of the creative process is the abstract linking. Once you’ve mastered the form, you break it. Free of form, the artistic purpose links and connects distant objects, concepts, feelings. People measure creative prowess in terms of originality, intention, and coherence.
AI is still not at this level, but current advances are proving akin to magic. While these systems aren’t capable of plot coherence, they’re achieving remarkable creativity.
In 2015, Andrej Karpathy released a seminal article where he demonstrated the use of Deep Learning to generate texts, character by character. His paper and code created a before and after in generative text AIs.
A year later, researchers from the Google Brain project showcased an AI that was capable of building coherent threads between two distant phrases. Sentence gradients in a nutshell.
Building on these two, Robin Sloan, a writer and enthusiast programmer decided to use this technology to aid his writing. He made, among other things, an impressive text editor helper that inspires his writing.
Like Robin, other creatives are finding an increasing fascination with the use of AI systems to enhance their creativity. Ross Goodwin’s use of AI to create poetry around an image is mind-blowing.
Goodwin even took it as far as letting the AI generate a script that was then filmed under the name of Sunspring.
Another intriguing and powerful development is Johnson’s work in generating descriptive image paragraphs. A recent paper shows how can AIs can extract meaning from an image an write down what they see.
Future of AI creativity
While it might be true that creative AIs lack coherence, I have no doubts that we’ll see new systems that start building coherence into their models.
The combination of sentence gradients, deep learning networks and the rise of generational adversarial networks paints an exciting future.
I can see how an adversarial network can drive a plot idea while the convolutional network builds the different chapters. Even simpler than that, we could mix human-made plot and character templates with the automated text generation of current models.
I believe Artificial Intelligence can be creative and develop new innovative formats. Many artists lash back against the mere notion of having machines creating anything.
“But while algorithms can be useful tools in the artistic process, Wilson said he didn’t necessarily think robots will ever be able to create art more meaningful than humans because humans have one thing that no robot ever will: the experience of living a human life.”
For me, creativity, like many other human aspects, isn’t unique to us. Machines will, driven by their own needs, become creative in their own way.
The question though is, what does this mean for companies? Is this something that only writers care about?
The topic is an interesting one because many organizations are betting on a future where their jobs won’t be taken by the machines. Most of those “protected” positions are so because they entail a certain degree of creativity and holistic approach.
The more I observe how Deep Tech is evolving, the less convinced I am that creativity will be a safeguard for jobs.
On the other hand, I think that AIs can significantly enhance the creative process. We are already doing that. Machines are the heart of video games, CGI effects in movies, Frank Gehry’s Guggenheim, etc.
In many ways, companies using AI enhanced creativity processes will win the day. A good example is Reuters Tracer, their in-house tool to detect, validate and corroborate real-time news. Their capacity to write part of the story in real time is giving them a decisive edge.
Machines will become creative. They will be able to write, compose and paint as good as humans. You can debate it their art is soulless or not, but they’ll produce stunning pieces. Most of this creations will be consumed by humans and will give a serious scalable competitive edge to any company that employs such methods.