It sounds like science fiction: a brain nurtured in a Petri dish learns to pilot a fighter plane as scientists develop a new breed of "living" computer. But in groundbreaking experiments in a Florida laboratory that is exactly what is happening.

The "brain", grown from 25,000 neural cells extracted from a single rat embryo, has been taught to fly an F-22 jet simulator by scientists at the University of Florida.

They hope their research into neural computation will help them develop sophisticated hybrid computers, with a thinking biological component.

One target is to install living computers in unmanned aircraft so they can be deployed on missions too dangerous for humans. It is also hoped that the research will provide the basis for developing new drugs to treat brain diseases such as epilepsy.

The brain-in-a-dish is the idea of Thomas DeMarse, 37, an assistant professor of biomedical engineering at the University of Florida. His work has been praised as a significant insight into the brain by leading US academics and scientific journals.

The 25,000 neurons were suspended in a specialised liquid to keep them alive and then laid across a grid of 60 electrodes in a small glass dish.

Under the microscope they looked at first like grains of sand, but soon the cells begin to connect to form what scientists are calling a "live computation device" (a brain). The electrodes measure and stimulate neural activity in the network, allowing researchers to study how the brain processes, transforms and stores information.

In the most striking experiment, the brain was linked to the jet simulator. Manipulated by the electrodes and a desktop computer, it was taught to control the flight path, even in mock hurricane-strength winds.

"When we first hooked them up, the plane 'crashed' all the time," Dr DeMarse said. "But over time, the neural network slowly adapts as the brain learns to control the pitch and roll of the aircraft. After a while, it produces a nice straight and level trajectory."

Previously, scientists have been able to monitor the activity of only a few neurons at a time, but Dr DeMarse and his team can study how thousands of cells conduct calculations together. But it is still a long way from a human brain.

"The goal is to study how cortical networks perform their neural computations. The implications are extremely important," Dr DeMarse said.

The first result could be to enable scientists to build living elements into traditional computers, enabling more flexible and varied means of solving problems. Although computers today are extremely powerful, they still lack the flexibility in working things out that humans take for granted.

Computers, for example, find it difficult to spot the difference between a table and a lamp if they are unfamiliar with them.

"The algorithms that living computers use are also extremely fault-tolerant," Dr DeMarse said. "A few neurons die off every day in humans without any noticeable drop in performance, and yet if the same were to happen in a traditional silicon-based computer the results would be catastrophic."

The work by Dr DeMarse and his team is attracting interest from scientists around the world.

The US National Science Foundation has awarded them a $US500,000 ($A640,000) grant to produce a mathematical model of how the neurons compute, and the US National Institute of Health is financing research into epilepsy.

-Telegraph