Why is the human brain more flexible than AI? Chinese scientists crack the underlying secret

Full release of hard technology丨New discovery of neurons may reshape artificial intelligence

Chinese scientists have cracked the underlying mystery of the brain's flexible decision-making. A single neuron is like a microcomputer, which may inspire the underlying algorithm of the next generation of artificial intelligence. Uncover why the human brain is more flexible than AI.

Recently, a scientific research team from the Center for Excellence in Brain Intelligence of the Chinese Academy of Sciences revealed for the first time the computational mechanism by which brain neurons participate in flexible decision-making, filling a long-missing gap. This achievement is likely to inspire the underlying algorithms of the next generation of artificial intelligence.

What is the difference between AI and the human brain? The language that large models are best at, AI's image recognition, requires at least tens of thousands of times more training than humans. It’s also learning, why can the human brain learn faster? The answer lies in the brain's basic structure - neurons.

There are many neurons in the brain, and we can think of them as small stations. Stations are connected by "synapses", just like the roads between stations. The essence of learning is to make the pathways between neurons smoother.

In order to solve the mystery of the brain's flexible decision-making, scientists designed a set of auditory decision-making experiments to train mice to play a guessing game. The mice will hear different sounds and need to find hidden patterns. After the mice figured out the current rules, the scientists quietly changed the set of rules and observed how the mice quickly adapted to the new logic.

The scientific research team used two-photon microscopy to observe neurons in the mouse auditory cortex. A neuron is not a bare dot. It grows many branch-like protrusions, which are dendrites. The long distal branch is called an apical dendrite. The top dendrite is responsible for carrying the inferred "abstract rules", and the cell body is also called the cell body, which is responsible for receiving sensory signals from the ear. Two sets of information are compared in the same cell - equivalent to making a conjecture while receiving reality, and judging on the spot whether the conjecture is true.

This experiment also proves that a single neuron itself is a small computer, and the interaction between the dendrites and cell body of the neuron can complete cognitive calculations related to decision-making. Only after the guess has been verified many times and confirmed to be reliable, the brain will slowly strengthen the synapses and preserve this rule permanently.

Current AI only simulates the synaptic connections between neurons, which is the "road between stations" we mentioned earlier. If the newly discovered neuron computing mechanism can be integrated into the algorithm, it is expected to create a new generation of artificial intelligence that can quickly propose hypotheses and deduce new rules with a small number of samples. Hundreds of millions of years of evolution have put exquisite computing power directly into a single neuron of the human brain. We may have just uncovered the tip of the iceberg of its mysteries.