GENEVA, SWITZERLAND – APRIL 19: A detailed view of the blackboard with theoretical physics equations in chalk by Alberto Ramos, Theoretical Physics Fellow and visitor, Antonio Gonzalez-Arroyo from the Universidad Autonoma de Madrid (both not in frame) at The European Organization for Nuclear Research commonly know as CERN on April 19, 2016 in Geneva, Switzerland. (Photo by Dean Mouhtaropoulos/Getty Images)
Getty Images
Over the past few months, two notable developments occurred in AI that you may not be aware of. In July 2025, two AIs from Google and OpenAI won Gold status at the International Math Olympiad. Around the same time, a scandal erupted after an announcement that AI had generated new math was debunked shortly thereafter. Why is the relationship between AI and math so important? What does it mean for business?
What Happened? And What Did Not?
First, let’s explore what actually happened. The International Math Olympiad is one of the hardest mathematical contests in the world. This year, it appears that two AIs were able to solve extremely challenging math problems (that had been solved before). The problems were also solved simultaneously by human participants in the math olympiad.
What did not occur was the creation of “new math” or the solving of problems previously unsolved by humans. While claims to the latter were made around the same time, unrelated to the Olympiad event, these claims were later shown to be false.
The Significance Of AI Doing Math
The reason that both of these announcements generated interest is the complex and critical relationship between AI and Math.
AI has historically been adept at learning patterns (from past data) and using them to predict the future. To understand how this works, consider two examples. An AI can learn the relationships between house prices and factors like location, number of bedrooms, etc., and use that information to predict the price of a previously unseen house. Similarly, AI can learn the pixel patterns in images of eye scans and detect retinopathy.
Generative AI (such as Large Language Models) is capable of creating new content (text, images, etc.) by synthesizing high probability patterns from previously studied content. Examples of poetry written by ChatGPT fall into this category.
Reinforcement Learning AIs can experiment with their environment, trying different things, and learning from this experience what works to optimize for a task. Game-playing AIs fall into this category.
So, why is math interesting? Math is often considered an important frontier of AI since it implies reasoning. Solving math problems requires that a set of reasoning steps be taken, one after the other, to conclude. If AI can do reasoning, it can move to the next step beyond prediction, generation, and exploration. Note that reasoning in AI is not futuristic. Modern AIs are known to reason at various levels, with examples in <> and <>. While AI is steadily moving in this direction, math is still often considered the frontier as far as reasoning compared to human brains is considered.
This is also why the distinction between the two announcements (Olympiad and New Math) is so critical. While Olympiad problems are challenging, they are not new math. It is quite possible that the AI learned the solution from some data that it processed, recognized the pattern between what it had learned and the question asked, and assembled an answer. The debunked announcement about new math was important because the claim was that the AI created something through reasoning that it could not possibly have seen before. While this was proven to be false, this development is exactly what researchers wonder about. Will AI ever be able to do this?
What Does This Mean For Business?
The significance to business can be simply summed up as – the more AI can do on its own, the more it augments or even replaces humans in business workflows. Writing beautiful emails or marketing copy is one thing. Coming up with new solutions to complex problems is quite another. Coupled with existing advances such as AI agents that can interact independently and autonomously in workflows, reasoning provides powerful capabilities.
Without necessarily worrying about whether AI will ever be able to do new math, there are some takeaways from the confirmed developments that apply to business.
– The first is that AI has shown an ability to “mimic” reasoning for advanced problems. Even if this was done through the acquisition of vast information and pattern matching, the fact remains that complex problems were correctly solved. This development implies that AI can be harnessed not just for routine tasks but for increasingly complex tasks. For example, tasks that AIs can assist with include analyzing complex data and generating new solutions to business problems, reviewing technical designs and proposing optimizations, identifying supply chain bottlenecks, and suggesting new product features, among others.
– The second is that AI can now generate solutions that are difficult for humans to understand (and therefore to assess the correctness of). For a business, the implications are profound. Should employees pose challenging open issues to AI? Are they able to assess the answer for correctness? What level of expertise should the employee have? Should hard questions only be asked of AIs by expert employees?
What Should You, As A Business Leader, Do?
Creating a culture and practice of AI Fluency is the best way to protect your business and employees, and simultaneously get the best leverage from continual AI innovation. What this is likely to mean in practice:
– Ensure a healthy practice of AI use with skepticism and human assessment. Select tools that employees are encouraged to use, that have suitable usage policies that protect your business.
– Encourage employees to share internal learnings of tool efficiency, tips and tricks, and problem spots. Publish best practices for AI usage and revise regularly. Explain how AI works, to remove the mystery and clarify why AIs can be both valuable and also make mistakes. Show employees that their judgment is still critical and expected.
– Ensure human accountability. Even if an AI can solve ever more complex problems, the human is still responsible for the answer being correct, safe, and suitable for the business. The implied assumption is: ask the AI whatever you want, but make sure you can understand and evaluate the answer.
The Bottom Line
The recent news activity around AI doing math shows that we are moving into an era where AI solves complex reasoning problems, whether by mimicking reasoning or actually doing reasoning. For many business tasks, the difference is not critical as long as the outcome is of decent quality. Consider an experienced employee making a decision. Does it matter whether the decision came from logic or past experience, as long as it was correct? That said, a senior employee with vast background and experience and the ability to reason strongly, is likely to make better decisions. AI is heading in this direction.
This AI development (and those sure to follow) is beyond the control of most businesses. What a business can do is to be aware, enable experimentation in safe spaces, and guide employees.
