AI Creativity: 2 Types, One Possible, One Impossible
Notes From the Desk: No. 44 - 2025.04.09
Notes From the Desk are periodic informal posts that summarize recent topics of interest or other brief notable commentary.
AI’s Creative Potential: 2 Types Explained
There are 2 basic types of creativity:
1) Permutations of things that already exist. AI can do this.
This represents the ability to identify patterns in existing information and recombine them into new permutations of information.
2) Exploration of the space of possibilities through understanding. This is what gives us new semantic information without prior precedent. AI cannot do this.
This is distinctly different from permutations in that it requires no prior existing patterns from which to derive a result. It is the creation of something new seemingly from nothing.
▌AI Creativity ≠ Human Creativity = Novelty
The novelty aspect of creativity is one of the most misunderstood concepts in the debates about AI and human creativity. AI creates new information through the recombination of existing patterns of information.
However, humans can do something substantially more, but the difference is not widely perceived because most things consist of both Type 1 and Type 2 components, with Type 1 simply being the most recognizable.
▌Type 1 Creativity: Permutations Of Existing
Type 1, permutations, are often argued to be the only type of creativity that exists for both AI and humans.
It is asserted that all the novelty we see in the world is nothing more than permutized versions of everything that existed prior. It is true that recombinatorial creations represent a significant portion of all the content that is created by humans. But is everything new truly nothing more than a combination of what already existed?
Since most creations will be dominated by the more predominant creativity expressed by Type 1, it may appear that this is the substantive ingredient in all creative works.
Type 1 creativity can be interesting, appear mostly novel, and have substantial value. It comes in the form of recombined song beats, rhythms, remixes, movie reboots, spin-offs, and a nearly infinite supply of every consumer product that has some simple variation of the same function as everything else. However, it also has substantial limits.
Type 1 creativity can recognize and recombine existing patterns, but it cannot answer why the patterns exist. The underlying principles or rules that create the patterns are not understood. Therefore, new things will be constrained to the boundaries of existing data or information.
AI systems can create data that appears to be outside the training set; however, in that case it will often be nonsensical associations - randomized guesses.
Within artistic domains, hallucinations outside of the training set may be perceived as creativity, as these domains of information don’t necessarily have right and wrong answers. Therefore, AI has greater dominance in these areas.
▌Type 2 Creativity: Semantic Exploration
What we call understanding requires no training, no prior patterns. Understanding is the perception of the rules of existence and how they can be manipulated to achieve previously impossible goals; it is the comprehension of underlying principles.
It is the ability to play chess given only a single page of rules, without ever having played a game. It is a young child who can manipulate their environment to obtain objects out of reach, without ever needing to see it done prior.
It is the artist who can find new visual styles not from simply observing what everyone else has done, but through the exploration and understanding of the physical interaction of the artist's tools, such as how paint mixes, flows over different textures of a canvas, the effect of different brushes, etc. The artist can imagine these possibilities in the mind without having ever seen them done before.
It also includes the most profound, the Einstein that discovered new physics through thought and the Tesla who envisioned so many devices long before we had the technology to build them.
However, Type 2 creative exploration is often not apparent to observers, as its contribution and existence requires the same kind of understanding that is held by the creator. This is often a deep, domain-level of understanding that others don’t posses.
The observer sees the rocket: rockets existed before, rocket engines existed before, it is just a recombination of what already existed. Flying things that land themselves existed before. However, self-landing rockets was a creative innovation that required instrumental understanding of the world.
▌Implications For AI Creativity And Novelty
Type 1 creativity will sufficiently keep many people entertained. There is more data in the world than any person could possibly know; therefore, new arrangements of existing data will be perceived as "new" by many.
Whatever "insights" are derived from Type 1 will be limited and will only come when combined with human reasoning. This is not to say that Type 1 is useless or has no value - it does. Nonetheless, its value is often overstated because it is assumed it encompasses the creativity uniquely found only in Type 2.
Additionally, Type 1, as handled by AI systems is even more restrictive in its output than Type 1 performed by humans. Humans can find patterns and incorporate parts from sparse data. However, AI requires substantial amounts of data and is only good at Type 1 tasks for data and patterns that are already common.
Type 2 is what is necessary for revolutionary discoveries: major breakthroughs in health, energy, or new architectures and engineering designs not based on existing ones. These things still remain in the human domain and will not be coming from any of the currently existing AI architectures.
▌AI Reassembles Memorized Content
If understanding is a prerequisite for creating new semantic information, then all of the information created by gen-AI can be nothing more than reassembled parts combined through pattern matching.
Patterns still need content to be applied. In order to generate output there must be actual stuff on which the patterns can operate. So, where does this stuff come from? There are only two input sources: the prompt and the training data. There are no other options for a machine with no intelligence or reasoning, it is incapable of creating new information. Therefore, all of the information in the output is either from the prompt or the training data.
▌AI Civilization Modal Collapse
So, what happens over a long period of time when everything created is nothing more than reassembled knowledge that already exists?
The more we depend on AI, we become a civilization of infinite remixes and refinements, trapped within the conceptual boundaries of the past. It is the evolution to an undead civilization - a civilization that has the appearance of life, but in reality, it is meaningfully dead.
A system with no semantic information becomes unstable and deteriorates in the real world. This is what we see with AI’s modal collapse that trains only on its own output. This is not merely a technical glitch of AI training, but an underlying principle of information and intelligence. The real world is a chaotic system that requires constant adaptation for stable surviving systems; otherwise, they collapse and disappear.
Mind Prison is an oasis for human thought, attempting to survive amidst the dead internet. I typically spend hours to days on articles, including creating the illustrations for each. I hope if you find them valuable and you still appreciate the creations from human beings, you will consider subscribing. Thank you!
No compass through the dark exists without hope of reaching the other side and the belief that it matters …
I agree with your thesis that there are two types of creativity and the current iteration of AI is limited to type 1. But I am not complacent about this limit.
Suppose that it becomes possible to make a genuinely self-learning device. Now, give it sensors to mimic all human senses, inclinations to prefer certain sensations over others, locomotion together with freedom to explore, and time. Could such a device, exposed to real and direct experiences of the world, eventually develop some degree of consciousness? Might it possibly exhibit characteristics of human consciousness like self-reflection, meta-cognition (awareness of one's own knowledge, and lack of knowledge), and abstract reasoning? If so, I think it could produce type 2 creative work, if it wanted to.
I think the key phrase there is the last. A device that wants to do something novel of its own volition would almost certainly be capable of type 2 creativity.
If one looks at a new-born baby human it exhibits little more intelligence in its first hours of life than an insect. It is the combination of inherited instincts and abilities, environmental stimuli, the ability to learn, and the particular structure of the human brain that enable it to become a fully conscious and creative individual. If these factors could be incorporated into the structure and experience of a device it might eventually develop the characteristics of human consciousness, the last of which would enable such a device to create entirely novel work. It would have type 2 creativity.
Whether we ought to even try to do this is an ethical question. I don't expect the practitioners of AI to pay more than lip service to it.
Why does recombination of the things that are already there, does not create novelty?
I wonder, if you could do this argument for genetics as well. The genetic alphabet is pretty limited. However, it is the basic code for novelties since millions of years. Perhaps, in thinking about artifacts created with AI, we also need something like a genotype/phaenotype distinction?