Creativity Accelerator

Alexis André

There is a lot of romance that surrounds the notion of creativity, yet I argue that the process is simple. For me, being creative is the combination of two skills: being able to come up with something new then being able to evaluate if that new thing is worth something. The first part of the problem is easy, this is a natural process that happens all the time. By just existing and reacting to the world around us, we already approach a similar activity. Our way of living is new, as it is by definition unique to each individual. Coming up with new ideas is an extension of this situation. A combination of past experience, knowledge and intuition allows anyone to react to their environment. When faced with an unknown task, you are never truly clueless on how to proceed. You might not know if what you are thinking about is the right way to do things, but you surely have ideas. The difficult part is indeed how to evaluate your ideas, that is, not spending time on ideas that are meritless to focus on the ones that are promising.

This framework around creativity is interesting to me for a few different reasons. First it takes the fantasy aspect out of the creative process (you might be inspired, but that is another aspect I am not covering here) then it gives two sub-tasks to focus on in order to improve our creativity. How can we get better at being creative? If we are better with coming up with new ideas, then we are more creative. If we have a better understanding of how to evaluate our ideas, we are more creative. Suddenly the task of improving one's creativity is less daunting.

My work recently has been researching ways to help with this challenge: how can technology help people become more creative? Can technology provide solutions that targets those two specific tasks? Can we use technology to get better at being creative?

Coming up with new ideas

The first half of our challenge is now to come up with new ideas for a specific challenge. My answer to this is to design generative systems: custom designed pieces of software (mostly, but it can be hardware) that allow their users to explore the space of possible outputs in a meaningful way. In other words, where the latent space of the algorithm is explainable and explorable. Combined with randomness, we are close to a state where every new point in that space (that is, every output of the algorithm) is new. Designing such generative systems is not an easy task nor always obvious, but there is a difference in how we can evaluate the result: a great system that generates new outputs is different from just thinking about new outputs. The evaluation function for creating a generative system that creates new outputs is not looking at the same things as the evaluation function of creating new outputs. This added layer of freedom is key to this approach and provides another angle to tackle the problem at hand.

Once the generative system is in place, exploring the set of possible outputs becomes a constant flux of new ideas that are there to stimulate, at a very low cost in terms of labor (a posteriori). Imagine for example a pottery simulator that proposes various generative patterns mapped directly to shapes that can be turned on a potter's wheel. There is no need to physically bake the pieces to get a feel for how the patterns suggested by the algorithm look like yet at the same time it is straightforward to modify a few parameters to guide the idea towards someone's interest. Massive saving in time spent sketching or prototyping - the bulk of the time is spent in the design of the master algorithm.

The goal of the master algorithm is to offer options and to not select what might be "good" or avoid the "bad". The target is to show a wide range of possible outputs from a set of meaningful parameters, for the user to explore, not for the algorithm to decide. Focus on the breadth and variation rather on curation.

Evaluating new ideas

The big issue is then to choose which ideas are good, which ideas are bad and most importantly, which ideas are worth pursuing (a good idea might not be worth it for reasons unrelated to its artistic or creative appeal, like costs, practical concerns - context matters). This means being able to feel when things are right, wrong or interesting. As we are dealing with a broad sense of creative activities, there is no way to establish a common measure of greatness that would work fine for any domain and application. We need to rely on taste, that is, the personal decision of the creator to commit to an idea.

I will even go further and argue that taste is the most valuable skill that one person can possess. The ability to decide and to feel confident in your own choices. Not following influencers, not relying on external validation. Just trusting your instinct that this specific idea, this specific outcome is a good one, that it feels right.

Sadly, this is not something you can easily learn from other people. You can learn the tropes of the field, the shared cultural language of the domain at hand - knowing the rules to break them is a common saying. But you cannot really learn *taste* from other people. What you can do though, is *train* your qualitative evaluation process. There is no supervision, no perfect answer to compare your instincts to, but you can train your taste by just exercising it many times.

There again, generative systems provide a very practical platform to train your taste. As the generation of new ideas is becoming automatic or extremely efficient in terms of time spent, there is a abundance of ideas that need to be evaluated, to be categorized into good/bad, right/wrong, with potential/useless. A way to train your evaluation muscle.

Human Engineered Intelligence

This process is actually very close to how machine learning techniques got extremely good at creating convincing pictures and more: splitting the process into two sides. One that generates new outputs and one that evaluates those outputs. To train a machine to generate pictures, run the process until the generator side is able to come up with pictures that the evaluator cannot distinguish from real pictures. Come up with new ideas; evaluate those ideas.

Similarly, training your taste is close to unsupervised learning (maybe reinforcement learning if we consider the reward to be this innate feeling of getting it right), with finding meaning or structure in a set of data.

Generative Systems for Humans

One of the key differences here is that I argue that the most valuable skill is the personal taste. This translates to creative direction when you can guide the process towards your preferred outcomes. As such, the systems should be able to stay explainable, controllable and explorable. If you cannot work the tool to get to your desired output, you are not in creative control: you are not working with the tool, you are collaborating with it - and it has its own direction as well.

The generative systems I am offering are created within this bottom-up approach: start from something that is familiar to the user, where they understand the logic, the process behind the generation before expanding to some more advanced state that might overcome the capabilities of the human brain while still being explainable (a posteriori). Allowing the user to control exactly what is happening and as such, to stay connected with all that matters: taking the right (according to their own taste) creative choices.

In other words, I'm creating generative systems that widen the creative horizon of the creators and artists who trust their taste. Ideas beyond what they could imagine on their own, but in a familiar context and scale: following their technique, style and sense of aesthetics.

Traditions were once new

This idea of working with a generative system to create might sound off in the context of traditional art and craft, yet I argue that it complements the practice in a way that does not compete with its artistic nature. Technology - from the evolution of tools, the introduction of new chemicals, production techniques, materials - has always been a driving force for the evolution of any artistic medium. Any traditional way of doing things was once new, we only look at them as traditions once they become standards and time passed. Generative systems might be oblivious in the future, and the usual way of thinking about creation. Creating the traditions of tomorrow with the technology of today.

We are not there yet, but we might as well start.

Mass production of unique assets

The other use of generative systems

There is another way to work with generative systems that presents a different approach on how to add value to content.

Let us consider the generative system described above. It is a collection of ideas, algorithms and parameters that does not attempt to evaluate their output in any subjective way. It tries to keep all meaningful parameter exposed so that the user is free to explore the space of the algorithm, looking for combinations that trigger the imagination or provoke any feeling. It does not however try to decide for the user what is good or not.

Yet this is obviously another challenge to master. A generative system that is tuned to also perform curation of its outputs, only allowing the good outputs to shine through. Taking random input and shaping it into something that is guaranteed to be at least not bad (I would argue that this is where to draw the line: making sure everything is ok and maybe good or great is more important that getting some result that might be exceptional at the cost of many bad outputs). This system is now able to create a collection of unique assets by its use of a generative process, furthermore in a fully automated process.

This brings the concept of mass production of unique items to the front of the scene. When the production of custom artefacts becomes affordable, the bottleneck is not on the production side, but on the creative side of the process. Producing one thousand unique items requires one thousand unique designs, patterns, ideas. Generative processes can provide those one thousand unique ideas. A collection of pieces that share a common aesthetic, style yet each individual result is different and unique.

Generative Creative Direction

Generative art by Alexis André

This begs immediately the question of where the creative direction is when creating such generative systems and what it all means in the first place. I would argue that creative direction is even more important than before as it needs to work at the meta-level. This is not a matter of preference between two different outputs, but over a whole subspace of possible outputs. Comparing an infinity of possible results that could happen in a specific branch of the algorithm with another subbranch. Thinking about the ramifications of one change over the course of the whole process.

The task is different yes, and like all creative problems, there is no clear yes/no answer. It comes down to taste and how this is expressed in the design process of the generative system. Another layer of creative expression but fundamentally the goal is the same: making sure the choices matter in the end.

Massive customization

Let's look at some very specific examples where unique content is already offered and where a generative approach could take it to another level: jewelry and more specifically wedding rings. The process is already streamlined to offer to couples that are engaged unique rings. Choosing the stone, the material, the design (from a catalog of existing pieces), engraving... allows the customers to feel that their rings are unique and are a physical testimony of the nature of their relationship (if we believe the message put forward by those brands). The pieces are already made to order, so adding an extra layer of generative process to add some truly unique elements to the design of the rings should prove to be the low-hanging fruit of applying those ideas. Making customization part of the process, where design is not just a collection of choices, but a whole space of possible outcomes.

Individual expression

In the modern world where AI systems have been the target of a lot of complaints about what they use for training data and what is the actual value of their outputs, what does it mean to create if most of the labor of creating is taken care of by a complex system? We are back to focusing on one aspect of creation: having something to say and less on the "how". If generative AI systems are rehashing past techniques and ideas to explore the space in-between existing data, it becomes harder to extend the space in new directions. It becomes harder to find your own voice: this is the "how" of creating.

On the other side, it also means anyone without the skills needed to actually realize the creative idea (drawing, painting, sculpting) can now access a system to find what they might want to say, if they want to express themselves through this abstract creation process. Being able to execute an idea is the first step to become a creator, and the tool might not matter for everyone.

Generative systems help by training the evaluation part of the creative process, eventually leading to the development of personal taste, to make sure that the outputs chosen by anyone are meaningful decisions based on individual preference and are less dependent on influencers and trends.

Creativity as a commodity

When being creative is a skill that anyone can learn and explore, what will people make? Will people focus on their own taste, their own unique style? What will people use to express themselves?

We might end up in a situation where everything (from physical goods to virtual items) is created for a unique person in a way that allows for everyone to find themselves in those items. Where everything reflects the uniqueness of every being. But thanks to the power of generative systems, we also can share some common traits (at the meta-level, in the design of the algorithm). Similar, but unique.

It's the world I'm building. A world where the joy of creation is shared by many and where many are sharing their creations.