AI artworks
In the world of art, the topic of AI or generative AI art is divisive and a common thread of discussion. Generally, any discussion breaks down into a two-sided argument of those who support and use the tools and those who are vehemently against it. But what actually is generative AI when it comes to art and why is it a hot topic.
At its core, Generative AI employs advanced algorithms and deep learning techniques to generate images. The process begins with users providing input, such as selecting themes, colours, styles, or other preferences, and then a text description. This input serves as the foundation for the generative AI algorithms to work. AI algorithms then analyse this input and generate images that reflect the user's preferences while incorporating elements of randomness and creativity.
Users can then interact with the generated artwork, exploring different variations or requesting modifications to better suit their preferences. This interactive feedback loop helps refine the AI models further, ensuring that the generated artwork continues to evolve and improve over time.
One of the major points of contention that the art world has with generative AI is that it needs samples to be trained. Samples are essential during the training phase to learn the distribution of the data. By exposing the model to a diverse set of real-world samples, it can learn the intricate patterns, structures, and relationships present in the data.
The main issue with this is that the models were trained on the artwork of artists without consultation or request. The developers of these models used a method know as ‘scrubbing’ to pull all examples of art from the internet to train their diffusion models. This artwork was then fed directly into the models without any input or knowledge of this happening from the artists.
In my opinion it’s hard to argue that this approach is unethical and unfair on the artist community. This is essentially a tool that will be used to replace and displace art based jobs, that has been trained against their copyrighted images. Taking skills that they have learned over decades in order to break this down into data to train a service that will make others millions, if not billions, in revenue without the artists getting anything. For example, Midjourney is pulling in revenues of $200million annually, with a potential market valuation of $10billion, even though they wouldn't be able to do what they do without the samples taken from artists.
Through training these AI models on artists work there is a belief that generative tools build images through a photobashing type technique, similar to a digital collage. Taking pieces of different images and piecing together a picture using algorithms to create the final piece. However, this is not how they work.
Diffusion models are used to build the images. These are a class of generative AI models used for tasks like image generation, text synthesis, and more. At the heart of diffusion models is the concept of iteratively applying diffusion (think static) steps to generate new samples. During each diffusion step, noise is added to the data, gradually blurring it. The model learns to reverse this process, effectively "un-diffusing" the noisy data to produce realistic samples.
Sounds confusing, because it is. However this hopefully illustrates that the process is very different from photobashing images together. What this also means, is that it would be very unlikely that someone would accidentally create an image like any of the ones it had sampled, which is another fear people have with gen AI. However, this doesn’t mean that generative AI cannot be used to plagiarise other artists work. In fact these tools make it very easy to make copies or replicate the style of an artist. With a simple text command of “in the style of…” any artist who has a decent size portfolio online can be copied.
The defence against this is that this is down to the user and not the tool itself, and this is true, but this has removed any barriers of skill needed before. Considering how the creators of these tools got their sample data they could easily remove that command and stop it from being possible to replicate any artist. After all, an artist’s style is often the most important thing to them and what differentiates them from everyone else, and the platform owners owe it to artists to stop it from being so easy to rip them off.
Considering the nefarious methods that sample data was gotten, and the fact that these organisations are making no effort to protect artists, it’s hard as an artist to support the tools. Regardless of this, these tools are available, and no matter how much animosity I have towards the vendors, I don't hold it against anyone using these tools to create images as long as they are not actively ripping off others. at the end of the day, like any digital art tool, generative AI is just another.
Within the world of AI artists there is a wealth of difference between how people are using the tools. You have the most basic, who are just entering text commands and are accepting what ever pretty image it spits out (with a few minor changes here and there). This is your everyday user and it’s hard to call these people artists or even creatives. They have a basic concept, but no real vision. They appear to just be spitting out images at a fast rate to try and bolster some kind of following on social media.
At the other end of the spectrum, you have creative individuals who are not only using the advanced features of the AI software but multiple tools in order to use generative AI to help them achieve a vision. From a digital perspective this could include tools like photoshop, blender, and other similar applications. Some artists are using AI to help them with ideas, or they are taking the images it creates into the real world and then applying paint or other mediums to bring the final piece to life. This is where the tools truly intersect with the art world and are used in the workflow of true creatives pushing the limits of what is possible.
Unfortunately, in amongst all the noise and the generic images that are being pumped out of AI art tools, it is hard to find the true creatives using the tools and creating unique pieces of art. In my opinion, most of the AI generated art work has a similar feel and look to it that isn't all that appealing. I will also add that with most AI creations, I loose interest pretty quickly once I find out how its been made. For me, I am fascinated by the skill and the effort that people put into a piece of work, not just the final outcome. I see the skill that people build up similar to those who do sports. For example, in athletics the 100 metres sprint is the pinnacle of speed, and when you see people like Usain Bolt achieve the times he did, it's just mind blowing. but if someone told you they could do it quicker in a car, you wouldn't care, because there is no skill in it, no training, its just a powerful tool.
When we talk about AI art and its impact on society, after the ethics of the samples, we have to question what impact will AI art have on artists, and will it have an impact on the available jobs for creatives?
Regardless of the level of effort that some may put into the creation of their AI works, there has been a relatively unanimous push back against AI in the fine art world. No competitions or exhibitions that I have seen accept AI artworks, and there are very few galleries out there who will hang it on their walls. From this perspective it could be seen that Generative AI, and its creations, will have limited impact on the fine art world. Here people want to buy handmade work and invest in artists who have honed and fine-tuned their skill over time and have differentiated themselves from others.
Other aspects of the art world may not fare as well. For low price art, or art that is aimed at the average person, the market has just got harder. A lot of people have no idea about AI art, or could identify what has been created by it, they just want cool images to go on their wall at a reasonable price. In this space AI art has made a competitive market even more so, just based on the sheer amount of content they can flood into it.
There is also has been some push back at the affordable level against AI created artwork. those in the know, do want art that has been created by a person. unfortunately to get round this I have even seen AI artists advertising their work as handmade, implying they have been painted or drawn, to appeal to those who still want handmade work at an affordable price.
When we move into the space of jobs for artists the picture could get even bleaker, especially as the tools improve. It’s possible that over the next few years we could see creative jobs eaten up by Generative AI. Replacing high-cost artists with anyone who can type a command in and generate an original image. Now this is going to depend a lot on the quality expected, but as the tools get better, this will become easier.
There have already been some companies who have used AI tools to create adverts, posters and even album and book covers. In a lot of cases there has been backlash from the creative industry at these uses, but that is only because of clear tells that can be spotted in AI art at this time. As it gets better, and it’s harder to tell, more and more organisations will look to save money by using these tools. Unfortunately, I wouldn't be surprised to see this sector shrink and lose many of the jobs currently available over the next 10 years or so.
Does this mean we will see AI take all art-based jobs? I don’t believe so. Where generative AI will excel is in the areas where fine controls are not required. When it comes to more creative type roles, where artist-based skills such as colour theory and perspective are required. There will remain to be roles for artists. However, we may find that generative AI becomes a tool that is used to support these roles rather than replace. Think of a fashion designer who hand draws some new designs but uses generative AI to produce some photo realist images of the garments before they make a physical mock ups.
Essentially generative or AI art is not going anywhere now. It’s been let out of the bag and is now making some organisations good money from monthly subscriptions. It will also continue to improve over time, getting better and more accurate, but even with all this it will not mean the end of the human artist. It may reduce the number or jobs available, but anywhere where something of quality needs to be produced there will be a human in the driving seat, be this with or without the use of generative tools. Remember these tools do not know what they are creating, its all just algorithms, so until AI gets smart enough to know and truly learn what it’s doing, we will always be in control. And when it does get that smart, we may have more to worry about that where the art comes from.
In my next blog I will continue the AI theme with a look at what scares me about how we are using AI and the worrying trends that are already starting to appear.
Photo by Google DeepMind on Unsplash
