Current LLM and GAI Models Are Not Creative: By Design
Economic innovation is based on novel, unexpected, and original ideas.
True disruptive innovation depends on the generation of novel ideas, originality, and groundbreaking epiphanies. Current Transformer and GAI models have yet to create an invention capable of being considered original and useful by any national patent office, or generate a product or service that has transformed a market or industry. Why? Certainly, we are still very early in the development phase of AI models that can contribute to innovation processes, but for current AI models like GPT-4 and Claude, the challenge runs much deeper and is systematic to current AI solutions.
Transformer and LLM “creativity” is pattern-based simulation.
Current LLMs like GPT-4 do not “think” in a traditional sense—they predict the next most probable token based on pre-learned text and semantic patterns and relationships. Though incredibly powerful, current LLMs simulate a type of artificial creativity by:
- Generating variations on known ideas (like joke structures or existing patents).
- Applying reinforcement learning (RLHF) to prioritize stronger creative responses as evaluated by human users.
Current AI models face an unresolvable limitation.
LLMs, by design, cannot generate truly novel concepts—they have been built to work from existing patterns and have been designed to rearrange patterns to generate a type of difference and mimic the generation of new ideas. By design, current transformer-based models like LLMs are not capable of generating breakthrough ideas that go beyond past human ideas–the basis of novelty, originality, and disruptive innovation.