LLMs Compute Big, Humans Think Smart: A Perfect Combination

The combination of expansive thinking yet practical thinking drives innovation.

Reviewed by Jessica Schrader

Key points

• LLMs excel in generating novel, boundary-pushing ideas but often lack in feasibility.
• Human creativity is grounded in reality, focusing on practical, actionable solutions.
• The best ideas often come from human-AI collaboration, combining bold innovation with real-world judgment.

Sometimes, it comes down to bragging rights for both humans and AI. A recent study has shined a light on this intriguing debate: Can AI truly generate research ideas as creatively as humans?

Researchers compared the large language models (LLMs) Claude 3.5 and GPT-4 to human experts in generating novel and practical research ideas. What emerged was a striking pattern—LLMs excel in novelty, generating boundary-pushing, exciting ideas, but often lack in feasibility, while human researchers, anchored in reality, offered more practical solutions. The real revelation? The most successful ideas came from human-AI collaboration, where each played to their strengths, yielding results that were both innovative and feasible.

This brings to mind John Nash’s famous "win-win" equilibrium theory. Nash posited that the best outcomes arise when players consider the strategies of their counterparts and adjust accordingly. In the context of creativity, the Nash equilibrium suggests that the optimal strategy for innovation is not an "AI vs. human" battle, but a "win-win" collaboration where AI pushes the boundaries of possibility and humans guide those ideas with practical judgment.

LLMs and Flex Thinking: Unbounded Creativity

LLMs are remarkable at what we might call flex thinking. Trained on vast datasets, they have the capacity to make connections across seemingly unrelated domains, suggesting ideas that challenge conventional boundaries. For example, an LLM might propose applying principles from evolutionary biology to problems in computational linguistics or use methods from one field to solve challenges in another. The beauty of this approach is that LLMs aren’t hindered by human cognitive biases, fears or the constraints of traditional thinking. They can suggest ideas that are genuinely new, unrestricted by the practicalities that often weigh down human innovation.

In the study, this flexibility translated into high marks for novelty and excitement. LLMs generated ideas that human researchers admitted they wouldn’t have considered. This makes LLMs especially valuable as creative partners, helping humans explore previously unimagined solutions or approaches.

Humans: Anchored in Reality

Where humans outperformed LLMs was in feasibility—the ability to turn these creative ideas into actionable strategies. Human creativity is shaped by experience, grounded in the realities of the world. Whether it’s knowledge of engineering constraints, ethical considerations, or financial limitations, humans excel at understanding what is possible and achievable within the framework of current knowledge and resources.

However, this practicality comes at a cost. Human researchers tend to think within established parameters. This means that, while their ideas are more grounded, they may also be more conservative. The tension between creativity and feasibility often pulls humans toward safer, more incremental innovations.

A Win-Win Collaboration

So, what’s the best way forward? The study’s findings suggest that combining the bold creativity of LLMs with human pragmatism yields the most effective results. When humans worked alongside LLMs—refining, editing, and assessing the AI-generated ideas—they produced concepts that were not only novel and feasible. This is where Nash’s equilibrium theory comes into play: collaboration is not only the best strategy, but it also offers a path to optimize both human and AI strengths for greater impact.

The human-AI collaboration creates an equilibrium where LLMs provide the creative spark, generating a wide range of innovative ideas, and humans act as the filter, bringing real-world judgment to decide which ideas have the most potential to be realized. In this way, both parties benefit: AI expands the creative horizons, and humans make sure those horizons are anchored in reality.

Innovation and Human-AI Synergy

This study has significant implications for the future of creativity and research. As LLMs continue to improve, their capacity for flex thinking will likely deepen, allowing them to generate even more sophisticated and novel ideas. But it’s important to note that LLMs won’t replace human creativity. Instead, they will complement it, pushing us to think more freely while still relying on our real-world knowledge to bring those ideas to life.

The future of innovation will likely be shaped by synergy—a blending of AI-driven novelty with human-driven practicality. LLMs can help break through the cognitive walls that limit human imagination, while humans keep AI grounded by steering these ideas toward feasible solutions. Together, they create a powerful feedback loop that enhances creativity and problem-solving in ways that neither could achieve alone.

In this new age of creativity, success won’t be about AI outthinking humans or humans outperforming AI. It will be about working together, leveraging the unique strengths of both to achieve results that are both imaginative and actionable. Just as Nash’s equilibrium suggests, the best outcomes will arise not from competition, but from collaboration. And in that collaboration lies the potential for truly groundbreaking innovation.