The Common Sense Gap
One of the biggest hurdles for AI is replicating human common sense. This is the vast, unspoken library of knowledge we use to navigate the world. For instance, a person knows not to put a pizza in the oven for three hours, but an AI might not grasp the consequence
unless explicitly trained on that specific outcome. This is because AI learns from patterns in data, not from lived experience. It lacks a genuine understanding of cause and effect, social norms, or the basic physics of everyday life. This deficit can lead to illogical or nonsensical answers to simple questions, a major barrier for robots or assistants needing to operate in the real, unpredictable world.
The Empathy Deficit
AI can be trained to recognise keywords related to emotions, but it cannot actually feel or understand them. True emotional intelligence—the ability to show empathy, build relationships, and read social cues—remains a uniquely human trait. An AI chatbot can generate a grammatically perfect message of condolence, but the words feel hollow because they aren't backed by shared experience or genuine feeling. This is a critical limitation in fields that depend on human connection, such as therapy, leadership, and customer service, where a stock empathetic response can often make a situation worse. Without real empathy, AI cannot build trust or be held truly accountable for its decisions on a moral level.
Creativity vs. Recombination
Generative AI can produce stunning art, music, and text that appears creative. However, what it's actually doing is recombining and remixing patterns from the massive datasets it was trained on. It doesn't have personal experiences, emotions, or the intention to create something genuinely new; it calculates, it doesn't feel. While AI is a fantastic tool for augmenting human creativity by generating ideas, it struggles with true originality. Human creativity stems from lived experience, imagination, and sometimes, intentional rule-breaking—things an algorithm cannot replicate. The deepest, most resonant creative works come from a place of purpose and emotion that machines simply do not possess.
The Physical World Problem
A concept known as Moravec's paradox highlights a strange truth in AI and robotics: things that are hard for humans (like complex calculations or playing chess) are easy for computers, while things that are easy for humans (like walking or picking up an object) are incredibly difficult for machines. The skills of a one-year-old, like perception and mobility, require enormous computational resources to replicate. This is why, despite impressive demonstrations, today's most advanced humanoid robots still struggle with mundane tasks like folding laundry or loading a dishwasher. These actions require a mastery of sensorimotor skills honed over millions of years of evolution, something that can't be easily programmed.
The Hallucination Hurdle
A persistent and well-documented issue with large language models is their tendency to "hallucinate"—that is, to state falsehoods with complete confidence. Because these models are designed to generate plausible-sounding text, they can invent facts, sources, and details that are entirely incorrect. This makes them unreliable for tasks that require absolute factual accuracy, such as news reporting, medical diagnosis, or financial analysis. While techniques are being developed to reduce these occurrences, the hallucination problem remains a core technical challenge, reminding us that fluency should not be confused with intelligence or truthfulness.
