It also assumes that your project isn't close enough to some existing one that you really need to define a completely new interface instead of just using or extending that of an existing device.
If you do out the math it turns out there's a 50% chance of collision after only 77,163 hashes.
[1] https://en.wikipedia.org/wiki/Birthday_problem
[2] 256 days=2^8, 16=2^4
[3] 365 days=2^8.51, 19=2^4.255
No, but there are thousands of preset IDs in the hundreds of drivers already present on the computer.
For this reason, isn't it safe to pick a random 32-bit identifier?
Devices don't just collide with others on the same computer, but with other drivers on the same computer.
Edit: Unfortunately sillysaurus3 deleted the parent comment. It had an interesting summary of a simulation of the collision problem.
But randomly selecting an id doesn't make much sense. For example here are all the ids Linux knows about: http://www.linux-usb.org/usb.ids. Obviously don't select Logitech, Microsoft, Canon, Samsung, etc. manufacture id. Either select an unused manufacture and then randomly select the device id. Or select a manufacturer id that is no longer used like US Robotics (I doubt they are making any new modems) and then randomly select a device id.
As soon as make your device, connect it to a Windows machine which will try and find a driver -- if it doesn't, probably no conflict. If there is a conflict, it's with a product that distributes it's driver manually - so you don't have to worry. Or a product that doesn't exist yet, which could be an issue.
If you're a small-volume project, then presumably people will need to have custom kernel patches anyway and will be unlikely to actually have a device with a conflicting ID.