GPT-5.6 Pro has solved a 30-year-old problem. What does this mean for science?

MarGib July 21, 2026

🌐 🇵🇱 Polski · 🇬🇧 EN

Reflections on the future of artificial intelligence often touch upon the moment when algorithms begin to solve problems previously considered unreachable – for example, complex issues in complexity theory that researchers have been working on for decades. Such a breakthrough would not only be a technological success but also a signal that AI is starting to play a key role in fundamental scientific discoveries.

Wizualizacja sztucznej inteligencji analizującej złożone obwody logiczne na holograficznym interfejsie w laboratorium badawczym — Conceptual illustration: AI solving a computational complexity theory problem

Let us imagine a moment that turns out to be a breakthrough for theoretical computer science: the announcement that an advanced artificial intelligence model has solved the circuit equivalence problem for the complexity classes NC¹ and P/poly. This issue, formulated at the end of the last century, has resisted attempts at a solution for three decades. Why would this step be so important and what does such a scenario mean for the future of science?

A problem that had no solution for 30 years

Computational complexity theory deals with the classification of problems based on their difficulty. One of the key questions is whether two logic circuits – representing Boolean functions – are equivalent, meaning whether they produce the same result for every input. Although this sounds abstract, the answer to this question is of fundamental importance for understanding the limits of computational power in computers.

The circuit equivalence problem for the classes NC¹ (problems solvable by circuits of logarithmic depth) and P/poly (problems solvable by polynomial-size circuits with additional "advice") was particularly difficult. Previous attempts to solve it relied on heuristics or were limited to special cases. None of these approaches yielded a general solution, and many experts believed that the problem might be unsolvable by classical methods.

Why was this problem so difficult?

The main obstacle was computational complexity. Checking the equivalence of two circuits requires analyzing all possible input combinations, which is practically unfeasible for large circuits. Furthermore, this problem is closely linked to fundamental questions about the hierarchy of complexity classes, such as the famous "Does P = NP?" question. Solving one of these problems could have a cascading effect on the entire field.

It is worth noting that this problem was not merely an academic curiosity. Its solution could have practical applications in integrated circuit design, algorithm optimization, or even cryptography. That is why it has attracted the attention of both theorists and engineers for years.

How can artificial intelligence approach the solution?

If we were to analyze the potential operation of such a model, we would not be talking about a simple text generator. Modern systems of this type combine advanced transformers with modules for symbolic mathematical reasoning. Equipping them with mechanisms for generating and verifying formal proofs may prove to be the key to success.

The problem-solving process

In a hypothetical research process, the work of such a system could proceed in several key stages:

Literature analysis: The algorithm analyzes thousands of scientific papers on complexity theory, looking for gaps and inconsistencies in existing approaches.
Hypothesis generation: Based on the collected data, the model formulates its own conjectures regarding circuit equivalence.
Circuit simulation: Using reinforcement learning techniques, the system tests its assumptions on logic circuit models for various input data.
Proof construction: Finally, the system constructs a mathematical proof using automated theorem proving.

The key to such a breakthrough could be the discovery of a new structure or class of circuits that would allow for the reduction of the problem to already known mathematical theorems, such as Barrington's theorem. This would show that a complex issue can be simplified in a way that human researchers had not previously considered.

Differences in approach: AI vs. traditional methods

Traditional methods for solving problems in complexity theory rely on human intuition, manual calculations, and limited simulations. Mathematicians and computer scientists often spend years analyzing one aspect of a problem, testing different approaches, and looking for analogies in other fields.

Artificial intelligence works quite differently. It processes vast amounts of data in a short time, testing millions of combinations and generating hypotheses based on statistical patterns. This approach has its advantages – it allows for the exploration of paths that might escape human researchers. However, it also has limitations that are worth keeping in mind.

One of the main challenges is interpretability. A proof generated by artificial intelligence can be hundreds of pages long and extremely difficult to analyze even for top specialists. It is not always clear how the model arrived at certain conclusions, which raises natural questions about the reliability and reproducibility of the results.

Proofs and verification

Reports of such breakthroughs always stir up great excitement and mixed reactions. On one hand, part of the scientific community approaches them with enthusiasm, pointing out that preliminary analyses do not show glaring errors. On the other hand, skeptical voices appear immediately. Many experts point out that machine-generated proofs often rely on novel, poorly understood techniques. Questions also arise as to whether the algorithm did not commit errors typical of heuristic approaches, such as oversimplification or skipping rare edge cases.

Verification process

The verification of such a complex proof must be a multi-stage process:

Community peer review: The work is made public and analyzed by experts from around the world. Discussions on specialized scientific forums help to quickly identify potential gaps.
Implementation: The source code and tools used to generate the proof are shared on platforms like GitHub. The community attempts to reproduce the results and verify their consistency.
Formal review: The work is submitted for review in prestigious scientific journals. This process can take months or even years before the community officially recognizes the proof as correct.

It is worth emphasizing that even if the proof turns out to be flawless, it does not automatically mean that all its aspects will immediately become clear. Researchers will have to spend a great deal of time analyzing and interpreting the results to fully understand and utilize this discovery.

Implications for science and technology

Solving the circuit equivalence problem for the NC¹ and P/poly classes could have far-reaching consequences. Here are some key areas that may be affected:

Complexity theory

If the proof is confirmed, it suggests that the NC¹ and P/poly classes are equivalent in terms of circuits. This, in turn, could lead to new questions and hypotheses regarding the hierarchy of complexity classes. For example, can a similar approach be applied to other classes, such as P and NP? Are there other, yet-undiscovered relationships between them?

This discovery could also influence our understanding of the limits of computability. If certain problems that were considered difficult turn out to be easier than previously thought, it could change current research priorities in theoretical computer science.

Practical applications

Although complexity theory often seems abstract, it has real-world implications for technology:

Cryptography: Some cryptographic schemes rely on the difficulty of problems in the P/poly class. Solving the circuit equivalence problem could lead to new methods of analyzing these schemes, which in turn would force the development of more resilient algorithms.
Integrated circuit design: New methods for analyzing logic circuits could accelerate the design and optimization of integrated circuits, which is of great importance for the electronics industry.
Algorithm optimization: Understanding circuit equivalence could lead to more efficient algorithms for difficult computational problems.

The future of research with AI

A potential success in this field would be further evidence that artificial intelligence is becoming a key tool in science. Previous breakthroughs, such as AlphaFold in biology or AlphaTensor in computer science, have already shown that algorithms can handle problems that have blocked human researchers for years.

What distinguishes the latest multi-task models is their versatility. Unlike narrowly specialized systems like AlphaFold, modern neural networks are not limited to one field. This opens up entirely new possibilities – from pure mathematics to physics and chemistry. In the coming years, we can expect increasingly frequent use of AI to investigate open scientific problems.

Of course, this does not mean that machines will replace humans. They will rather act as advanced tools that accelerate the research process, helping to generate new hypotheses and test them on a scale unattainable for humans. Collaboration between human intellect and the computational power of algorithms could lead to discoveries we never dreamed of before.

Reactions of the scientific community

Discussions about the role of artificial intelligence in science stir extreme emotions – from great enthusiasm to deep skepticism and concerns about the future of traditional research methods.

Enthusiasm and hopes

Proponents of this approach see AI as an opportunity for the beginning of an entirely new era in the exact sciences. In articles and academic discussions, artificial intelligence is presented as a catalyst that can drastically accelerate the pace of discovery and open up previously unexplored directions. More and more prestigious universities and research centers are deciding to launch special grant programs dedicated to the collaboration of humans and algorithms in complexity theory.

Skepticism and concerns

Skeptics, on the other hand, warn against excessive optimism. They point out that machine-generated proofs can be so complex and counterintuitive that it is extremely difficult to assess whether we are dealing with a real breakthrough or merely a cleverly masked heuristic that overlooks key details.

Concerns are also being raised about whether too much reliance on artificial intelligence will weaken our own analytical abilities. Will algorithms begin to dominate the research process, pushing traditional, human-intuition-based methods to the margins?

Ethical and formal issues are also significant. How should authorship be handled for scientific papers in which artificial intelligence played a key role? Should AI systems be included as co-authors of publications? These are questions to which the scientific community is only just beginning to seek answers.

What's next?

Potential breakthroughs in complexity theory are only the beginning of a long road. The scientific community now faces a multitude of challenges:

Proof verification: A thorough analysis of every step could take experts years.
Interpretation of results: Simply confirming correctness is not enough. It is also necessary to understand how these results translate to the entire complexity theory.
Practical applications: The next step will be translating theory into practice – from new algorithms to more secure cryptography.
Further research with AI: These experiences show that algorithms will be increasingly employed to solve the most difficult problems in various fields of science.

Creators of AI technology are constantly developing systems focused on automated theorem proving. Everything indicates that the pace of discovery involving algorithms will only increase. Are we witnessing the beginning of a new era in science? Time will tell, but one thing is certain – artificial intelligence is already redefining the way we approach solving the most difficult problems.

If you are interested in how AI affects education and the skills of the future, read our article on AI in higher education. And if you want to learn more about how artificial intelligence can change our thinking, check out the text on the impact of AI on the brain.