There's a big buzz at the moment about the new programme, developed at Berkeley, to reconstruct past languages. The media is treating it as if it's an entirely new thing - as if the software alone gives academics the ability to reconstruct proto-languages - but, of course, it isn't, and most of the methods of historical linguistic analysis have been around for about 150 years. The computer just makes it easier.
I hope this will enable more accurate and realistic simulations of ancient languages - I'm sure it will, actually. It will definitely take a lot of the labour out of the procedure, which is famously labour-intensive (this at least part of the reason why use of reconstructed languages is only slowly becoming the norm in studies of prehistory). To some extent, this was an inevitability. Importantly, though, the results shouldn't be any different to human reconstructions. They should merely be easier and rely on larger data sets.
The software was tested with proto-Austronesian, which isn't surprising, given that it shows relatively little variation and has a proven source. It had already been reconstructed long before this machine reconstruction came along, but it's certainly a good way of testing the accuracy of the thing. I'd really like to see it applied to Altaic and Niger-Congo, because the former is disputed (in my opinion inappropriately and without basis [IANA historical linguist, so take this with a pinch of salt]), and because the latter has so many holes issues (even the greatest experts on Niger-Congo languages have serious disputes and make mistakes that are, in hindsight, pretty silly). Maybe this machine will provide a short-cut to the level of accuracy and detail we find in reconstructed Indo-European languages.
Historical linguistics is fascinating on its own, and it shouldn't need a fancy new device to make it cool. It's already cool.