Throughout history, humans have taken inspiration from nature to improve technology. Airplanes are modeled after birds, bullet trains are designed based on the shape of a kingfisher’s beak and neural networks — the foundation of modern artificial intelligence — are modeled after the human brain’s neurons and connections. In the same way that airplanes follow the aerodynamics of birds, AI and human cognition could follow an underlying “one learning algorithm” — a single process that explains how humans learn.
Computer scientist Leslie Valiant was one of the first to propose the idea that all learning follows a singular universal pattern — meaning that the same fundamental principles underlie how different systems, including both human brains and artificial intelligence, acquire knowledge. This idea, known as Probably Approximately Correct learning, argues that learning can be understood as a process of making predictions based on limited data while minimizing errors. In recent years, research has been done to test this hypothesis. For example, a study by Durham University found that when a person loses their sight, the area of their brain responsible for vision can adapt itself to be better suited for processing sound. This suggests that the human brain is highly adaptable and if AI can mimic this adaptability, it might be able to learn across multiple domains without needing specialized models for each task.
The one learning algorithm theory suggests that intelligence is inherently human, and an AI system, if designed to mimic the human brain’s flexible nature, could learn to solve any problem given to it.
A big step forward in this research is DeepSeek-V2, an AI model designed to reason and learn more like a human. As opposed to being trained for just one specific task, DeepSeek-V2 aims to tackle a wide range of problems by using a unified approach to learning. Similarly, research on Neural Algorithmic Reasoning suggests the possibility of traditional algorithms with neural networks, allowing AI to apply its knowledge across different areas rather than being limited to a single function. These innovations are pushing AI past task-specific learning and bringing it closer to a system that can think more flexibly — similar to humans.
If this idea is true, it could change the future of AI. Instead of separate models for different tasks — such as speech recognition, vision or problem-solving — one system could learn them all. AI would become more efficient, perpetually adapting to new challenges the way humans do.
However, some researchers argue that learning is more complex, and the only way to truly master a skill is with different processes for different tasks. Companies like OpenAI and Google DeepMind are actively researching aggregated AI systems, but challenges remain in achieving a single, universal learning algorithm. Some believe that AI will never fully replicate human intelligence.
Still, the search for a universal algorithm continues. If discovered, it could revolutionize AI and the current scientific understanding of learning, as well as push researchers to further explore the limits of both AI and the human brain.
