Three Worlds and the Illusion of Creativity

How Popper’s Three Worlds Reveal the Limits of Artificial Intelligence

Photo by Alexander Grey.

Popper's Three Worlds

In the 1970s, British philosopher Karl Popper was developing a way of examining reality that he called the Three Worlds. These weren't worlds in the traditional sense but rather conceptual realms. I'll do my best to explain.

World 1 is the world of physical objects and states. It's the inorganic—matter and energy, the organic—beings alive or dead, and artifacts—the things humans create, their tools, technology, art, and music.

World 2 is the world of mental objects and states. It's our senses, thoughts, and consciousness itself. The world that exists in our mind, in World 2, is not that which exists outside it. It's the result of biological activity that emerged from World 1.

Imagine you're holding a shiny red apple. The apple exists in World 1, but our experience of the apple and our perception of it—the texture, color, and taste—that occurs in our mind as a result of our senses and thoughts. A blind man's thoughts on the apple are quite different. They know about "shiny" and "red" things. They may even be able to guess that certain objects that feel a certain way are likely to be "shiny." But their perception of the apple does not include sight.

World 3 is the world of "products of thought" or the objects that emerge from World 2 activity. It's objective knowledge, concepts, theories, language, philosophy, history, or culture. Those World 1 artifacts I mentioned? They are the physical manifestation of World 3 objects.

Let's examine the apple once again. The word apple, the concepts shiny and red, our knowledge of fruit, recipes involving apples, and art featuring them, are all World 3 objects.

The interaction between worlds is worth examining. World 1 (physical) provides the raw data for World 2 (mental) activity where we create World 3 (conceptual) objects. Finally, World 3 objects may inspire us to take action in World 2 and make changes to World 1.

For example, individual apple pies exist in World 1, but only as a result of activity in Worlds 2 and 3.

A Primer on Generative AI

Before we examine generative artificial intelligence (i.e. text, image, and music generators) under the Three Worlds ontology, I'd like to provide a short explanation of how they work.

These systems are "trained" on vast amounts of text, imagery, and music, by identifying patterns and relationships within the data. When prompted, they generate outputs by predicting the most statistically probable next word, pixel, or note based on what they’ve "learned." It's both easy and dangerous to anthropomorphize these systems but, to simplify, they are really, really, really good guessers.

When it comes to language, the results are remarkable. But for a good reason. Yann LeCun puts it best:

"Language, it turns out, is relatively simple because it has strong statistical properties. It's basically a serialized version of our thoughts."

To de-anthropomorphize, it's like if math can make really, really, really good guesses. Which, to be honest, isn't much of a surprise.

Where Things Get Weird

There is no doubt these systems exist in World 1 as a result of activity within World 2 and World 3. But what about their results? The text, images, and music these systems generate are World 1 artifacts, yet there was no World 2 or World 3 activity involved whatsoever.

A poem is the result of human experience, emotion, and intent. A similar poem, generated by AI, did not involve thoughts at all. It did not experience, emote, or intend. It constructs output by probabilistically selecting the most statistically likely next token, pixel, or note. It's no wonder people use the word *sludge* to describe the results of these systems. Their outputs are a facsimile of creation, void of all depth and originality.

Confusion arises because AI is itself an artifact, one that encodes vast portions of World 3 objects—without ever experiencing World 2. It reflects human knowledge like a shattered mirror. Only when humans engage with AI-generated content, by interpreting it and assigning it meaning, do their outputs get pulled into World 3.

Conclusion

I remain a skeptical pragmatist when it comes to generative artificial intelligence.

I'm skeptical because, while I see its utility, the business model is hitched to the development of artificial general intelligence (AGI). I don't see AGI as a likely result of the current flavor of these systems so I'm predicting a bubble and bubbles tend to pop.

I'm a pragmatist because, it's here and it's not going anywhere, so we should get used to it and leverage it like any other tool in our collective toolbox.

But as a creator in my own right, I challenge individuals to leverage AI for what it is—a catalyst for creativity, not a replacement for it.