This illustration of artificial intelligence has in fact been generated by AI: view source

In the age of AI chat bots and content generators, we are witnessing a revolution in how we create and consume information. But as we, humans, increasingly rely on AI to produce our content, are we unintentionally setting up a future where AI’s growth stagnates? This question stems from a simple observation I made a several months ago: “AI is trained on human-generated data, but what happens when that data is increasingly AI-generated?”

The Current AI Boom

Today’s AI models, including large language models like ChatGPT from OpenAI, Claude from Anthropic, and more, are trained on vast amounts of human-created content. This content is original since it is made by humans and contains creative, natural(real), and motive-driven details that present an enormous number of perspectives from individuals. It includes everything from books and articles to social media posts, photos, and videos. For instance, many photographers, researchers, authors and more are publishing new and original work every day. The richness and diversity of this data has allowed AI to make impressive strides in understanding and generating human-like content.

The Potential Paradox

However, as AI becomes more integrated into our daily lives and workflows, we’re facing a potential paradox: AI is increasingly used to generate content(text, images, code, etc.). This AI-generated content may be fed back into future AI training datasets. Over time, a larger proportion of the available data could become AI-generated rather than human-generated. This could lead to a feedback loop where AI is learning from it own outputs rather than from original human creativity. One assumption is that the creativity-content generation rate of individuals decreases with AI overdependence.

The Risk of Homogenization

If this trend continues unchecked, we might face several consequences as time passes. AI-generated content might become more homogeneous over time, lacking quirks, innovations, and unexpected connections that human creativity brings. For example; the only reason OpenAI’s DALL-E can generate an image of humans playing football on Planet Mars is that - it has been trained on original image data created by many photographers, fiction scene designers, and satellites. I doubt if it can generate a realistic instance that it was never trained with. Therefore, without exposure to truly novel ideas and expressions, AI’s ability to innovate and generate original content could plateau.

Preserving the Human Touch

We are currently in the early phases where students generate essays using AI at the same time lecturers use it check whether thier content is AI-generated or not. It’s very astonishing how very few score less than 15%. One notable lesson from Jordan Peterson’s lectures states that, “thinking begins when you start writing”. This suggests that if individuals take on writing, traditionally with pen and paper or electronically with keyboard and writing app, they think, grow and nurture new ideas and perspectives.

To avoid this potential stagnation, we need to consciously preserve and value human-generated content by encouraging individuals to create more original work. Another approach that is being used by MetaAI from Meta suggests clearly labelling AI-generated content. An image generated using MetaAI contains a label at the bottom-right to indicate that it is not original but AI-generated. According to OpenAI’s Sam Altman, human creativity combined with AI assistance accelerates our workflows. In this way, we get to accomplish more faster in a short period of time. In my opinion, individuals should seek for fundamental/conceptual understanding of their fields and only use AI as a tool and go-to reference for explanations and insights in pursuit to solving a real problem.

The Path Forward

Before this era of AI chat bots, many users were creative, original and natural in their creations/data and more. Of course, both AI and human generated content are rising high and humans are still evolving, creating and seeking the understanding of the world around us. Therefore, the goal should be to use AI as a tool that enhances human capability, rather than as a replacement for human thought and expression. Personally, I suggest creating specialised AI models that are focused on one and only one field is the path forward.