With an increasing number of generative AI solutions popping up for video and image generation, a new debate has emerged. On one side, you have the enthusiasts of generative AI, delighted by the prospect of machines conjuring up images and videos at the click of a button. On the other, there are the critics, raising a collective eyebrow and asking, "But at what cost?"
The crux of the opposition lies in ethics. Generative AIs for video and image creation are, after all, trained on a vast reservoir of existing human creations—artworks, photographs, illustrations, and more. The concern is that the outputs of these AI systems might eerily echo the works of past artists. The risk? Accusations of plagiarism, copyright infringement, and a lingering question: Are these AI-generated pieces too close to the source material to be deemed truly innovative?
Yet, amid the noise, there’s an undeniable allure. The democratization of creativity means that more people than ever can now produce content—be it for social media or personal projects—that would have once been beyond their reach. Creatives, too, stand to benefit, potentially speeding up their workflows and unlocking new realms of possibility.
But the question that looms over this brave new world is whether generative AI will elevate the collective creative output or simply flood it with derivative works. As we stand on the cusp of this new era, it’s clear that the answer will shape the future of creativity itself.
The Volume of Creative Assets Has Surged
With generative AI, the barriers to becoming a “creative” have all but vanished. Now, anyone with a keyboard and imagination can dive into the world of AI art. The internet has witnessed a surge in communities dedicated to this emerging craft. From Reddit to Twitter to Discord, thousands of aspiring AI artists are honing their skills, crafting ever-more precise prompts, and eagerly sharing the results with the world.
To quantify this explosion of creativity, Everypixel took on the task of estimating the number of AI-generated images produced since text-to-image models first appeared in 2022. As of August 2023, the staggering total stands at 15.5 billion images.
For context, consider this: it took humanity's photographers a full 149 years to amass 15 billion photographs. Generative AI, on the other hand, accomplished the same feat in a mere 1.5 years. The implications are profound. The sheer volume of content being produced suggests a tidal wave of creativity—one that could very well lead to unprecedented levels of innovation.
Potential Degradation in AI Quality Without Creatives
In our previous piece, we discuss a world where data is the new oil, and tech giants have been gleefully drilling the internet dry, siphoning off every scrap of information they can find. But as recent research suggests, this relentless data mining could have unintended consequences, particularly for the LLMs that power much of today’s generative AI. The researchers found that when AI systems are trained on AI-generated content, they fall into a troubling phenomenon known as "model collapse." In simple terms, the AI forgets the rare and nuanced, becoming increasingly distorted by its own regurgitated output.
Now, imagine a future where every creative mind decides to rely solely on generative AI for their work. The result? These AIs could very well implode under the weight of their recycled content, leading to a significant decline in the quality and originality of AI-generated creative works. Fortunately, not everyone is ready to abandon traditional methods.
Take, for example, James Cuda, the CEO of Procreate—the go-to graphics editing software for iOS users. When asked whether his company would follow industry giants like Adobe in integrating generative AI features, Cuda’s response was as bold as it was refreshingly candid: Procreate will not be adding generative AI, and he emphatically stated that he “really f****** hates generative AI.”
This stance by companies like Procreate, and the creatives who continue to produce non-AI-influenced work, is crucial. These original creations provide the fresh material that generative AI models need to evolve and improve. Without a steady stream of novel content, there’s a very real risk that generative AI could collapse into a feedback loop of mediocrity, dragging global creativity down with it.
The Path Forward: Generative AI’s Coexistence with Creatives
Generative AI is, without a doubt, a remarkable tool. It has the potential to elevate creativity to unprecedented heights, allowing anyone with a spark of inspiration to step into the roles of artist, designer, or whatever creative title they fancy. It offers the promise of novel and innovative content at a scale never before imagined. But there’s a catch: for these tools to remain effective and truly innovative, they must be fed with the genuine, unfiltered work of human creatives who operate outside the influence of AI.
Yet, the path to a harmonious coexistence between human creativity and generative AI is riddled with two thorny issues. The first is the ever-persistent challenge of copyright, and the second is the need for a steady supply of fresh, original content to train AI systems. Both are, frustratingly, difficult to untangle for the same reason: How do we determine where human creativity ends and AI influence begins?
It's not as simple as labeling all AI-assisted content as unoriginal. The crafting of a prompt, and the fine-tuning of inputs—are creative acts in their own right, unique to the individual wielding the AI tool. Until we can devise a method to clearly distinguish between the AI’s contribution and the human touch, the specter of model collapse will continue to hover over the future of generative AI.
Comments