Laboratory automation has been growing more and more elaborate - multiwell tools for "parallel processing", and benchtop robot arms, for example, that will handle repetitive tasks. But flow cytometry offers some startling possibilities.
Flow cytometry, essentially, runs cells through a machine one at a time so that the machine can count them. Usually they're counted based on fluorescent tags (up to 18 at a time, apparently) but some machines can distinguish untagged cells by their refractive indices. The means by which they're fed through one at a time is one of those implausible-science-fiction-sounding techniques, hydrodynamic focusing, in which a thin channel is made effectively much thinner by guiding a layer of rapidly-moving immiscible fluid along the walls. Cells then zip through the thin stream of fluid down the center of the channel and pass through a scanner. Some scanners are able to image each cell as they zip past, others just look for tags.
In any case, run some soup of single cells through one of these machines and you see a distribution of cells in some high-dimensional space, based on their fluorescence properties. A few machines allow you to partition this space up based on whatever criteria you like, and then sort the cells: the thin one-cell-at-a-time stream is made to break into droplets, and the machine deposits a charge on each droplet based on what category the cell falls into. Then electrostatics let you separate the cells into different containers. This bizarre scheme is in actual use in sex-selection clinics, where sperm bearing X chromosomes are distinguished from sperm bearing Y chromosomes.
Where this gets even weirder, I think, is when you think about using these machines with machine-learning algorithms. You pour a cellular soup in one end, the machine runs through it, cogitates a bit, then identifies distinct populations of cells based on their tagging properties. It then runs them through again and hands you a vial containing a clean sample of each population, along with a summary of its tagging properties. All that's left is for you to name the new species...
Okay, all right, simply finding new kinds of cells isn't necessarily that interesting. So hook one of these machines up to one of the artificial-evolution experiments trying to build new biochemistry. You can pull out any cells that start going down a path you don't like. Is this not science-fictional?