Friday, January 18, 2013

Skynet Hasn't Yet Become Self-Aware, But.....

In 2012, University of Waterloo neuroscientists and software engineers programmed an impressive open-source software simulation of the human brain to run on supercomputers. The "brain" includes a digital eye for visual input and a robotic arm to draw responses. It can even pass parts of an IQ test which require adaptive processing, such as recognizing the patterns of a number sequence.

The brain, referred to as SPAUN (Semantic Pointer Architecture Unified Network), conatins 2.5 million simulated neurons, which enable it to perform eight separate tasks, including drawing copies, counting, answering questions, and fluid reasoning. Videos online at Youtube like the one below demonstrate Spaun in action.

The 2.5 million neurons are organized into simulated subsystems which mimic those of a real human brain, including a prefrontal cortex, basal ganglia, and thalamus, connected with simulated neurons which accurately mimic a real human brain's wiring.

These subsystems behave in the way a real human brain would: the thalamus processes visual input, data is stored in the neurons, and the basal ganglia sends specific tasks to the cortex region designed to handle them. Even the human brain's limitations have been simulated, causing Spaun to struggling with storing more than a few numbers in short-term memory.

Although the number of neurons is about 1/400th the number found in a human brain (and spaun can't spontaneously generate and rewire its synapses, it's still surprisingly flexible, showing how simple tasks can be built upon and interwoven to build complex functionality. Lead researcher Chris Eliasmith wants to next provide Spaun with adaptive plasticity, the ability to rewire its neurons and learn new tasks performing them, rather than being pre-programmed.

Eliasmith explains in greater detail here.

No comments:

Post a Comment