What is being computerized?
The neuron, including its 'channel' (like the axon'), the membrane, ion gates etc. can be seen as a 'cable'. Membrane has a certain capacitance, as well as resistance.
Voltage declines over distance, in a 'natural cable' as well as in a neuron. The whole neuron can be thus 'broken down' and be represented by 'regular cable electronics'.
The origin of thinking this way goes back to 1956, Nobel prize winners Hodgkin and Huxley - read PDF here.
A neuron, however, has more 'localities' than a cable. Some parts/'departments' have more conductance, less resistance or more capacitation as other parts. Ion channels and the node of Ranier have to be treated separately. But once represented, this is no problem to a computer.
So: neural networks can be calculated by digital, metal computers, using mathematics of computational neuroscience, basing itself on the neuron as electric circuit with components.
To illustrate the brilliancy of these people: born 1930: roadmap to 'whole brains simulation' with computational neuroscience - brilliant (pdf 130p.)
The other way round, neurons can be used to be a computer
Thus: this is pretty much 'face value evidence' neurons are exact enough.
Computational neuroscience is fruther complicated when neurons cannot only be represented by electronic components, but as molecules become components themselves.
The 'molecule as transistor. See the impressive 30 minutes presentation on molecular electronics organised by Google, in 2010, 'Quantum of Solace' but then not about James Bond but molecular electronics: http://www.youtube.com/playlist?list=PL5D13EF2B0584716D