590
14 Neural Networks and Brains
In the 1940s, Alan Turing, and independently John von Neumann, pioneered
mechanistic descriptions of the computers and the brain, starting with
programmable logic elements. Mechanistic models of biological brains (‘wet
computers’) often start with simplified Hodgkin–Huxley neurons, and then build
computer programs to model how networks engage in specific tasks. Interpretive
models of biological brains can use the ideas behind degrees and levels of
complexity (see Sect. 15.4). Neural networks become agents, with their own rules of
engagement. 12 With agents and rules, the behavior of the brain becomes fathomable.
14.4 The Human Brain
A healthy human brain has about 10 11 neurons and about 10 14 synapses. (Single
neurons are stimulated or inhibited by about 1000–15,000 axon terminal bulbs forming synapses with the neuron.) By using a number of different neurotransmitters,
each synapse stores 10 to 100 bits of information. Thus the whole brain can store
about 10 15 –10 16 bits. (For more data about the brain, see Appendix J.) To support
the activity of the brain, the brain uses up to 20–25% of the body’s oxygen supply,
even though it weighs only 2.5% of the total body weight. The brain must dissipate
about 25 W in heat (5.5 g of glucose per hour). Up to 90% of the free energy used
by the brain is consumed by the work done in the nerve membrane sodium pumps.
In turn, the pumps are run by ATP.
The human auditory nerve has about 30,000 nerve fibers in the basilar membrane,
each capable of a few hundred action potentials per second, making an upper limit of
about 30 million pulses per second. But there is significant redundancy, with strong
correlations between nearby fibers, reducing the information to about 3 million bits
per second.
The human retina transmits data at the rate of about 10 million bits per second.
Some of this information, the kind considered critical, such as quick motions, is sent
through faster ganglions at about 875,000 bits per second.
As individual neurons can die, redundancy in the brain is an important strategy
for protection of important information and programming. One effective strategy
for memory redundancy would be holographic storage, i.e. delocalized wave
interference patterns stored over a sub-network of neurons. (See Appendix C for
a further description of holography and holographic memory.) If there were a
tenfold redundancy, the brain would still be able to hold about 10 14 to 10 15 bits
of information (about 10–100 TB).
Our brain has an information capacity far greater than the DNA which created it.
(Human DNA, with 3 billion base pairs, can hold no more than about 6 × 10 9 bits
in its coding.) We know from the development of a fetus that a significant amount
12 Magnasco et al., Self-Tuned Critical Anti-Hebbian Networks, Phys Rev Let 102:25, 258102
(2009).
14 Neural Networks and Brains
In the 1940s, Alan Turing, and independently John von Neumann, pioneered
mechanistic descriptions of the computers and the brain, starting with
programmable logic elements. Mechanistic models of biological brains (‘wet
computers’) often start with simplified Hodgkin–Huxley neurons, and then build
computer programs to model how networks engage in specific tasks. Interpretive
models of biological brains can use the ideas behind degrees and levels of
complexity (see Sect. 15.4). Neural networks become agents, with their own rules of
engagement. 12 With agents and rules, the behavior of the brain becomes fathomable.
14.4 The Human Brain
A healthy human brain has about 10 11 neurons and about 10 14 synapses. (Single
neurons are stimulated or inhibited by about 1000–15,000 axon terminal bulbs forming synapses with the neuron.) By using a number of different neurotransmitters,
each synapse stores 10 to 100 bits of information. Thus the whole brain can store
about 10 15 –10 16 bits. (For more data about the brain, see Appendix J.) To support
the activity of the brain, the brain uses up to 20–25% of the body’s oxygen supply,
even though it weighs only 2.5% of the total body weight. The brain must dissipate
about 25 W in heat (5.5 g of glucose per hour). Up to 90% of the free energy used
by the brain is consumed by the work done in the nerve membrane sodium pumps.
In turn, the pumps are run by ATP.
The human auditory nerve has about 30,000 nerve fibers in the basilar membrane,
each capable of a few hundred action potentials per second, making an upper limit of
about 30 million pulses per second. But there is significant redundancy, with strong
correlations between nearby fibers, reducing the information to about 3 million bits
per second.
The human retina transmits data at the rate of about 10 million bits per second.
Some of this information, the kind considered critical, such as quick motions, is sent
through faster ganglions at about 875,000 bits per second.
As individual neurons can die, redundancy in the brain is an important strategy
for protection of important information and programming. One effective strategy
for memory redundancy would be holographic storage, i.e. delocalized wave
interference patterns stored over a sub-network of neurons. (See Appendix C for
a further description of holography and holographic memory.) If there were a
tenfold redundancy, the brain would still be able to hold about 10 14 to 10 15 bits
of information (about 10–100 TB).
Our brain has an information capacity far greater than the DNA which created it.
(Human DNA, with 3 billion base pairs, can hold no more than about 6 × 10 9 bits
in its coding.) We know from the development of a fetus that a significant amount
12 Magnasco et al., Self-Tuned Critical Anti-Hebbian Networks, Phys Rev Let 102:25, 258102
(2009).