Humans have been exploring the universe through experimentation. Experimental physics is the starting point of all understanding. But a hidden and crucial question is: is the experimental method we are using now really optimal?

Not necessarily.

Humans design experiments intuitively. We know how to assemble lasers, lenses, interferometers, and how to set up high-energy particle accelerators, but all of this is within the limits of human cognitive ability. Whether it is Newton or Einstein, even modern Nobel laureates live in the cage of "human intuition" in the final analysis.

That's the limitation.

A group of physicists saw this problem and simply stopped relying on intuition and simply turned to AI to help. A team at the Max Planck Institute in Germany simply calls itself an "artificial scientist laboratory", and the goal is clear: to create an AI scientist.

Not to write a dissertation, but to force the ultimate question: If we don't rely on human intuition, can we discover experiments that we never imagined?

The initial attempt was surprisingly simple: give the AI a bunch of virtual lab equipment – lasers, interferometers, beam splitters, light sources, detectors, put them in a "virtual experiment box" – and let it do the tricks. Messed up. AI doesn't feel like "asymmetry is wrong" and won't be deterred by warnings that "this has never been seen in textbooks." It only cares about the outcome.

As a result, the next day, the researchers turned on the computer and found that the AI had found an experimental solution that their team had not been able to work on for months. The experimental assemblage is asymmetrical, unintuitive, and even somewhat "ugly". But it,useful。

It's like opening Pandora's box.

The team later built a system called Pyto. It no longer deals with the actual device, but abstracts it into:Diagrams: The device is the node, and the connection is the edge. This is the field that mathematicians are most familiar with, but it is a new frontier of physical experiments. Complex experiments are "translated" into graph combinations and transformations. This step is very crucial. This is because the graph space can be directly connected to deep learning.

As a result, AI can evolve freely, trial and error, and combine freely in this abstract space. Once the solution is found, it is mapped back to the real device. In other words: humans are trying to experiment in three-dimensional space, and AI is searching for answers in high-dimensional space.

Further down the line is quantum experimentation. AI has begun to "reconstruct" a set of quantum system construction logic that humans have explored for decades.

One of these problems is called "entanglement swapping". According to the conventional idea, in order to entangle two particles, you must form an entangled pair and then interfere. The solution that the AI has found is stunning:It found a way to achieve entanglement exchange without using the original entangled state。 This is "impossible" in textbooks.

The results turned out to be correct.

This means that "the physical limitations that we always thought existed were not limitations at all". We are trapped by our own experiences, our intuition, our teaching materials. AI is not helping us "improve experiments", it is pushing the boundaries of human imagination.

What's even more outrageous: the mechanism given by AI is initially incomprehensible to physicists. I only see that the combination can run through, and the principle must be "read out" little by little. This is not training AI to do science, but forcing humans to "read the scientific language of machines".

Let's look at another case, the heavy player appears: the LIGO team, engaged in gravitational wave detection. A former Nobel Prize project. These people also came to the door to find AI, and the goal was very clear:Find out the gravitational wave detection devices that "humans can't think of".

As you can imagine, this is a "self-revolution" in the physical world: instead of designing experiments, we let the AI brute-force search and then we interpret its "crazy ideas".

The result? The solution given by AI is not "more optimized", but "completely unfamiliar". The way it is constructed, the sequence of the equipment, the measurement mechanism, all of which are out of the ordinary. Some scenarios have already beaten existing human designs in initial simulations.

It's hard to say that "AI is an auxiliary tool", it is already taking over the first step of physical experimentation.AI is not following humans, but leading the way.

There is also a more abstract but dangerous trend: to have AI read and understand the scientific literature of the entire human race, and thenPredict what experiments humans will do next, as well as proposing research paths that "humans have not yet thought of but have potential".

This is not about designing experiments, but about designing the future of the history of science.

The basis of this system is the "knowledge graph" – the compression of all papers into a vast network of who used what formulas, who measured what variables, and who made what conjectures. Then, a prediction algorithm is used to simulate the behavior of scientists.

The current version can already generate someIt seems to be very "scientific intuition" to select topics and experiment roadmaps。 The most important thing is that these suggestions come from a system that has no "curiosity".

That is, we are guided forward by a kind of wisdom that has no motives.

It sounds very sci-fi. But the physical circle has already moved on and is no longer waiting and seeing. At the frontier of experimental science, a "transfer of power" is taking place: from human scientists, to "non-human intelligence".

In other words:Experimental physics is no longer the sole business of mankind。

It's not AI that helps us draw, calculate, and simulate. It's AIDirectly propose experimental combinations, build configurations, and find counterintuitive paths, and then we chase after understanding.

The most paradoxical thing is that we still don't know whether AI is "smart" or "stupid". It has no ideas, no motives, no presuppositions, and will not be dragged down by historical authority, paper journals, and physical habits. It simply looks for effectiveness from a combination of pure structures.

But sometimes, that's enough.

Perhaps, what really hinders us from discovering new physics is never the limitations of experimental equipment, but the way we conceive experiments ourselves.

Instead of standing in front of the universe and asking questions, humans stand in front of AI and ask, "What exactly are you seeing?" ”