Do AIs Really Have Creativity? Marc Andreessen & Ben Horowitz Debate AI’s Limits

Source: AI product analyst A Ying.

The discussion at a16z’s annual Runtime conference featured Marc Andreessen and Ben Horowitz debating whether AI truly possesses creativity.

01 Human Originality Is Rare

Host: Many people discuss the limits of large language models (LLMs), such as their inability to invent new science or think creatively, merely assembling and recombining existing ideas. What’s your view?

Marc Andreessen: Real creativity and cross‑disciplinary thinking are extremely rare among humans. Most people are mediocre at writing, painting, or making music.

Today’s language models are already smarter and more creative than 99.99% of people. So why do we still claim AI lacks creativity? Why expect AI to be like Van Gogh or Beethoven?

This is a heavy‑handed point. We often discuss AI in terms of soul, understanding, or a vague human‑like light, forgetting that creativity is not mysterious—it often results from enough accumulation, combination, and connection.

Below is the YouTube link: https://www.youtube.com/watch?v=Y7dwbJ0AtUA

Human Originality Is Not Common

Host: At the Runtime closing, Marc and Ben talked about whether AI truly has creativity.

Marc Andreessen: I often hear two questions: 1) Is a language model intelligent enough to understand information and achieve conceptual breakthroughs like a human? 2) Does it have creativity to produce new artistic works?

Marc: My simple answer: Can humans actually do this?

Only a handful of people can achieve genuine originality; most people never reach that level.

Examples like Beethoven or Van Gogh show true creativity, but such figures are few.

If a language model outperforms 99.99% of humans, that alone is astonishing.

Historical breakthroughs usually build on at least forty years of prior accumulation. Language models themselves are the result of eight decades of technological buildup, essentially recombining past knowledge.

Artistic fields follow the same pattern: novels, music, and other forms are deeply influenced by predecessors. Beethoven, for instance, was shaped by Mozart and Haydn.

If a model already matches the top 0.01% of human intelligence and creativity, it is effectively at the pinnacle of human capability.

Emotionally, I hope human creativity remains unique, and I truly believe it does.

When I use these models, I’m amazed at how smart and creative they are, so I’m convinced they will eventually cross that threshold.

02 Cross‑Domain Transfer Thinking Is Uncommon

Host: You often ask, "Can humans do this?" when discussing LLM limitations like transfer learning.

Marc: Most people think within familiar bounds; few can truly transfer knowledge across domains.

I know only three people in my contacts of ten thousand can consistently do this kind of cross‑domain reasoning.

Even though humans have many limitations, we have created great art, movies, novels, and groundbreaking technology.

We don’t need 100% originality to change the world; it helps, but progress is already remarkable.

Ben Horowitz: I agree. Intelligence isn’t the only factor; understanding who you’re talking to and reading their mindset matters more.

We need to see problems from employees’ perspectives, not just from the boss’s view. Those skills come from long‑term human interaction, not IQ.

Management is about balancing relationships, knowing what people truly want, and deciding who in the team deserves investment.

Many management books assume universal methods, but each company’s product, staff, and structure are unique.

Understanding others’ mental states—Theory of Mind—is crucial. Surprisingly, very smart people may struggle with this.

The U.S. military has long used IQ tests (ASVAB) to assign roles, finding that large IQ gaps between leaders and subordinates cause team problems.

Extremely intelligent leaders may lose Theory of Mind and fail to understand their team.

If an AI reaches a thousand‑point IQ, its perception of reality could be so alien that genuine communication with humans would be impossible.

03 Intelligence Alone Is Not Enough

Host: Some argue that higher intelligence will inevitably dominate lower intelligence.

Marc tweeted that specialists with high IQ often end up serving generalists who integrate and communicate better.

High‑IQ experts often work for average‑IQ generalists. This challenges the notion that the smartest will always lead.

Intelligence correlates with many life outcomes (education, career, income, satisfaction) at about 0.4, but it’s far from deterministic.

Groups can become less intelligent than the average individual—a phenomenon seen in companies and nations.

Thus, the belief that the smartest will dominate is easily disproved.

04 Theory of Mind Is Key for AI to Reach Humanity

Host: Some say overly smart people can’t truly understand others because they assume everyone is rational.

Marc: Cognition, self‑awareness, and decision‑making aren’t solely brain‑based; the body plays a huge role (gut, hormones, microbiome, etc.).

Current AI lacks a body, making it a disembodied mind. Future robotics will give AI physical experience, enabling richer cognition.

When AI is placed in physical agents, it can collect real‑world data and perhaps achieve higher‑level cognition.

05 Bubbles, Competition, and Future Industry Landscape

Host: Ben says that as long as we’re asking whether it’s a bubble, it isn’t one. Bubbles are psychological phenomena where everyone believes there’s no risk.

During the dot‑com bubble, even Warren Buffett invested in tech, showing how belief can erase doubt.

AI demand is massive now; short‑term demand exhaustion is absurd.

Ben: Many venture capitalists are emotional, complaining about others’ higher valuations.

Marc: The key questions are whether technology can deliver on its promises and whether customers will pay for it.

Gavin called ChatGPT a “Pearl Harbor” moment for Google; the debate is whether incumbents will win the next wave or newcomers will take over.

Ben: Google is waking up, but OpenAI still leads in chat products without a massive ad ecosystem.

Marc: No one really knows; CEOs and investors are uncertain.

Ben: Some people wish it were a bubble to feel balanced.

Marc: Missing a project can haunt you for decades.

Ben: The industry’s future hinges on two fundamentals: technology delivering value and customers paying for it.

Marc: The competition isn’t just chatbots vs. search; the real battle will involve new, unforeseen interaction paradigms.

Ben: The notion that chatbots will replace search oversimplifies the future.

Marc: The next five years will bring product forms we can’t yet imagine.

06 Advice for Entrepreneurs

Host: What’s the biggest difference about this era compared to the past, and what advice do you have?

Ben: Don’t copy past management playbooks; AI researchers and full‑stack engineers are fundamentally different. Many problems must be re‑thought from first principles.

Marc: Talent scarcity (top researchers, engineers, compute) will eventually become oversupply as education spreads and AI helps build AI.

China already produces excellent models (DeepSeek, Qwen, Kimi) and can scale talent quickly.

Ben: Even Nvidia, the current chip leader, will likely see supply overtaking demand in a few years.

Marc: The AI revolution’s second phase will be robotics, and China is well‑positioned to lead hardware production.

07 U.S.–China Competition Heats Up

Host: Marc, you recently spoke in Washington about U.S.–China AI competition. How do you view it?

Marc: The West still leads in conceptual innovation and algorithms, while China excels at execution, scaling, and industrialization.

China efficiently adopts global ideas and turns them into mature products.

The race is a short sprint, not a long marathon; the U.S. may have at most a half‑year lead.

When AI moves into physical robots, China’s massive manufacturing ecosystem could give it a decisive advantage.

Ultimately, the future will be shaped by who can integrate AI with hardware at scale.