What can't be automated?

I think the "not everything can be automated" is a bit of a mis-statement.  Given time, effort, and resource, EVERYTHING can be automated.  The more accurate statement is "given a limited amount of resource, not everything can be automated."  This would translate, as well, into something more on the lines of "not everything SHOULD be automated."

For example, I could probably automate, given time and effort and resource, all the UI of the application I have under test including all colors, fonts, backgrounds, component positions, etc.  This would take hours, days, maybe even weeks.  Meanwhile, a human being can sit down, pull up the application, and in just a few minutes with a quick look at the application, tell you if the general look-and-feel passes inspection.  Can it be automated?  Sure... but SHOULD it.

What about hardware interfaces?  Recently, I asked about the ability to automate the testing of scanning barcodes.  Given a barcode displayed on screen, can I call or import code to read the barcode and return the results?  Sure... lots of stuff out there. But what if I wanted to automate the actual interface between a hardware device that does the scanning and the software?  Can I do that?  Sure... but it would take a good bit of resource to write the code to mock up a simulation of the hardware.  I could even go as extreme as grabbing some Lego Mindstorms and build a robotic arm to do the scanning.  But, again, given the time, resource, and effort... SHOULD I.  

So... yes, I think everything CAN be automated... but not everything WILL be due to resource... and, for the same reason, because of ROI, not everything SHOULD be.