39 Essays on Infinite Lifespans   Aubrey de Grey risk. Sometimes only a small risk, but even that may, at that time, be greater than our risk of dying of old age; and there- fore become something we take much more seriously than we do today. Further, some such risks – notably, exposure to new infectious diseases – seem likely to yield non-uniformly and unpredictably, if at all, to the increased expenditure that will surely be marshaled against them. [20] Thus, our con- tinuing and ever-improving success in avoiding physical or mental frailty will be at the increasingly unsatisfactory price of eschewing fulfilling activities that just might kill us. For this reason, I expect that in centuries to come we will work to develop what might be termed “non-invasive static upload- ing” or, more prosaically, backing-up of our cognitive state. At this point it is plausible, though obviously not known, that all the moderately persistent components of that state (that is, everything except short-term memory) are encapsulated in the network of synaptic connections between our neurons. The strength of those connections probably matters too, but maybe not to all that much precision. It is also plausible that, a century or three from now, an extremely high-resolution version  of  magnetic  resonance  imaging  may  exist  that  can scan a living person’s brain and detect all such information. This amount of data could easily be stored electronically by then, as it exceeds by only 3–5 orders of magnitude the stor- age capacity of today’s personal computers. Further, cells in culture will by then be amenable not only to differentiation along  desired  lineages  but  to  stimulation  to  form  synaptic connections with particular other cells to which their axons are  juxtaposed.  This  means  that,  in  principle,  a  copy  of  a living person’s brain – all trillion cells of it – could be con- structed from scratch, purely by in vitro micromanipulation of neurons into a synaptic network previously scanned from that brain. (Doing this in less than geological time would be possible because it could be highly parallelized. Since most