<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Sun, Mar 1, 2015 at 4:36 PM, Anders Sandberg <span dir="ltr"><<a href="mailto:anders@aleph.se" target="_blank">anders@aleph.se</a>></span> wrote:<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><br></div><div>But this is cell failure. Synapses fail at proper transmission *nearly all the time*! </div><div><a href="http://www.pnas.org/content/91/22/10380.full.pdf" target="_blank">http://www.pnas.org/content/91/22/10380.full.pdf</a></div><div><a href="http://zadorlab.cshl.edu/PDF/zador-jn-mi.pdf" target="_blank">http://zadorlab.cshl.edu/PDF/zador-jn-mi.pdf</a></div><div>Basically, there is a great deal of noise and variability introduced in synaptic transmission. The system is reliable since it uses many synapses and neurons, which are individually misbehaving a lot of the time. </div></blockquote><div><br></div><div>### Exactly - while neurons may be somewhat reliable survivors, the closer analogue of the transistor is a synapse (give or take on order of magnitude in complexity), and synapses are much less reliable than whole neurons. </div><div><br></div><div>Of course, this implies that the network structure of the human brain has to have evolved while taking into account this unreliability - so there is a humongous amount of structure devoted to error-correction. The challenge for designers of simulations who want to port the brain into silicon by mimicking the behavior of scanned neural circuits will be to separate the unneeded error-correction features and code only the necessary information-processing elements.</div><div><br></div><div>Rafal</div></div>
</div></div>