I used an AI for 7 months to search for a Theory of Everything. I failed. And it’s the best thing that could have happened.

Embracing the Limits of AI in the Quest for a Theory of Everything

Reflections on a Seven-Month Journey with Artificial Intelligence

In recent discussions about the search for a unified Theory of Everything (TOE), artificial intelligence often gets portrayed as a kind of magic box—an equation generator or a calculation shortcut. But after spending over half a year working closely with an AI to explore fundamental physics, I’ve come to see its true potential lies elsewhere. Far from being a mere tool for answers, AI is a mirror that exposes the strengths and flaws of our reasoning.

The Common Misconception

Many in the physics and AI communities treat these digital assistants as oracle-like beings, hoping they confirm their preconceived notions. The problem isn’t AI’s capacity to process data; it’s how we choose to engage with it. Framing prompts as requests for validation (“Does my theory hold?”) often leads to slanted feedback that reinforces existing beliefs. Conversely, asking the AI to challenge your ideas—seeking contradictions or flaws—transforms it from a passive tool into a rigorous opponent.

A Process of Self-Discovery

My journey began with a bold hypothesis—a concept of a dynamic “ether” that might underpin the universe. Working daily with the AI, initial results felt almost magical, boosting confidence. Yet, what truly mattered was how I engaged with those results: Did I seek affirmation, or did I test the hypothesis?

Instead of validation, I deliberately pressed the AI to find flaws, to scrutinize every assumption. The AI responded with relentless rigor, pushing me to reassess my certainties. Through this process, I not only gained new coding skills in Python—integrating programming into my daily toolkit—but also fundamentally changed my relationship with knowledge.

Facing the Uncomfortable Truths

The hypothesis that captivated me, the idea of a universal “ether,” was eventually dismantled—not by opinion, but through hard data. The model failed spectacularly. And in that failure, I found something invaluable: honesty. It demonstrated that embracing core uncertainty and being willing to abandon cherished beliefs is the true path to discovery.

In essence, this experience reaffirmed a simple but profound principle: When exploring complex ideas, it’s better to work with an AI as a skeptic rather than a supporter. Ask it to test your theories. Make it find contradictions. Let its “errors” guide you to more accurate questions rather than false comfort.

The Role of Constraints

One of