From an article by Allan Snyder in the Philosophical Transactions of the Royal Society B (Biological Sciences)
Savants cannot normally give insights into how they perform their skill and are uncontaminated by learnt algorithms. It just comes to them. They just see it. With maturity, the occasionally offered insights are suspect, possibly being contaminated by the acquisition of concepts concerning the particular skill. Yet, I have labelled one savant, Daniel Tammet, a Rosetta stone (Johnson 2005).There are more publications here on Snyder's web site.
By far, the most compelling argument for savant skills residing equally within everyone is that they can emerge ‘suddenly and spontaneously’ (Miller et al. 2000, p. 86) in individuals who had no prior history for them, either in interest, ability or talent (Treffert 2006; Sacks 2007, pp. 157 and 313). Striking examples include skills in art, music (Sacks 2007), mathematics (Treffert 2006, p. 85), calendar calculating (LaFay 1987; Osborne 2003) and possibly AP (Zatorre 1989, see p. 573). The same appears to hold for synaesthesia (Sacks 2007, p. 180), as theory suggested (Snyder & Mitchell 1999), which is reported frequently by autistic savants (Heaton et al. 1998; Sacks 2007; Tammet 2007, 2009). Furthermore, these acquired savant skills have been known to diminish with recovery from illness (Sacks 2007, p. 315).
Acquired savants arise from a variety of causes (Treffert 2006) including left frontotemporal dementia (Miller et al. 1998), physical injury to the left temporal lobe (LaFay 1987; J. Hirsch & A. Snyder 2005, personal communication), left hemispheric strokes (Sacks 2007, p. 315), severe illness to the central nervous system (Treffert 2006) and even when under the influence of hallucinogens (Humphrey 2002; Sacks 2007, p. 181).
Here's a previous post on this topic.