In my most recent paper (read it here), Tim Ralph and I demonstrate that a single-purpose optical quantum computer might be constructed much more easily than previously thought. In fact, building such a quantum computer is within reach of present-day technology. The paper extends recent results by Scott Aaronson & Alex Arkhipov, where they showed that a type of optical quantum computer, called a 'BosonSampling' computer, which is classically hard to simulate, can be constructed much more easily than 'universal' quantum computers (i.e. the most general type of quantum computer). Such a computer is 'easier' to build in the sense that no active switching, feedforward or memory is required - just single photon sources, passive linear optics and photo-detection (things that we can readily do in the lab).
In our work we demonstrate that even with very substantial loss rates (whether it be source inefficiency, loss within the circuit, or detector inefficiency), such a computer can likely perform calculations beyond the reach of any classical computer. This is great news for experimentalists, for whom photon loss is a killer in their experiments. Our results show that with enough single photon sources, even with a loss rate of 50% our optical quantum computer can likely do things that classical computers can't. This is unlike fault-tolerance results for universal quantum computers, whereby loss rates of even 0.1% are a killer. This all sounds very optimistic, but there is one drawback - no one knows what to do with this particular type of quantum computer. Aaronson & Arkhipov show that such a device almost certainly can do things that a classical computer can't, but no actual applications have been identified. So there's a catch - building such a quantum computer, which can outperform all the classical computers in the world put together, is within reach of present-day technology, but, no one has any idea what to do with it.
I hope these results will stimulate further research into Aaronson & Arkhipov-type devices, which will hopefully come up with some useful applications - a killer application for BosonSampling would represent a major step forward for the field. In the meantime, I hope some of the experimental groups in the optical quantum computation community will think about elementary demonstrations of such systems. Simple extensions of present-day multi-photon quantum walk experiments (e.g. this work by Peruzzo et al., and this work by Owens et al.) could act as an elementary test-bed for such systems.