An AI just won a math Olympiad but is it actually smart

Recent news of an OpenAI model achieving gold-medal level performance at the 2025 International Mathematical Olympiad has intensified discussions about the nature of machine intelligence. While such milestones demonstrate incredible proficiency, this capability should not be confused with genuine intelligence.

1/N I’m excited to share that our latest @OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI: gold medal-level performance on the world’s most prestigious math competition—the International Math Olympiad (IMO). pic.twitter.com/SG3k6EknaC

— Alexander Wei (@alexwei_) July 19, 2025

Modern AI models have become exceptionally proficient at a range of narrowly defined tasks. This includes mathematical reasoning, symbolic manipulation, code generation, computer vision, and complex language processing. These capabilities are the result of advancements in deep learning architectures like transformers, the availability of vast datasets, and immense computational power. These systems can perform sustained, multi-step reasoning and generate fluent, human-like responses within their specific domains.

However, the author asserts that this impressive performance is confined to the predefined scopes in which the models have been extensively trained. While it is tempting to view these achievements as a key step toward artificial general intelligence, Mann argues this would be a mistake. He makes a clear distinction between a machine’s ability to execute complex functions and the multifaceted, adaptable, and self-aware nature of what constitutes true intelligence.

This is how much time and money AI can save your business

The core of the argument rests on several key characteristics of human intelligence that machines do not possess. Mann provides a detailed breakdown of these missing traits, suggesting that until a machine can demonstrate most of them, it cannot be considered truly “intelligent.”

• Adaptability: Machines cannot seamlessly transfer knowledge to entirely novel or unforeseen problems without significant retraining. Their expertise is inherently specialized and does not generalize to different contexts in the way human intelligence does.
• Emotional intelligence: Current AI does not genuinely experience or comprehend emotions. Its responses, which may appear empathetic, are sophisticated statistical patterns learned from human interaction, not a reflection of true understanding of others’ mental states or feelings, often referred to as “theory of mind.”
• Self-awareness: Machines lack the ability for introspection. They do not reflect on their own internal processes, motivations, or the nature of their knowledge. They operate algorithmically and do not possess a subjective “self” that can ponder its own existence.
• Intentionality: AI systems operate based on programmed objectives and do not exhibit genuine curiosity or the capacity for autonomous goal-setting driven by internal desires or values. Their “goals” are externally imposed by their creators.
• Embodied experience: Lacking a physical body that directly interacts with the environment, machines cannot develop the common-sense understanding that comes from lived, felt experience. Their perception of the world is mediated by data and symbolic representations.
• Conceptual leaps: While generative models can produce novel combinations of existing data, they do not demonstrate the ability to invent entirely new paradigms or break fundamental rules in a truly original way that transcends their training.
• Causal reasoning: Machines excel at identifying correlations in data but often struggle with true cause-and-effect reasoning. They can predict what is likely to happen but have a limited understanding of “why” beyond statistical associations.
• Consciousness: Perhaps the most profound distinction, machines do not possess subjective experience, feelings, or awareness. They are not conscious entities.

Mann notes that the term “artificial general intelligence” (AGI) emerged in part to recover the meaning of “intelligence” after it had been diluted through overuse in describing machines that are not truly intelligent. He suggests that rather than racing to build AGI, the focus should be on considering the societal impact of current AI capabilities and limitations, while being careful not to confuse task-specific performance with genuine intelligence.

Featured image credit