Those who find Stallman's view of automatic updates against common sense should note that they are just the logical consequence of GNU's demand for "freedom to study and change the program in source code form". There's nothing more to it.
From that point of view automatic updates which could be reviewed at the source code level are different from binary updates, although both serve the same practical purpose for the "non technical end user".
Maybe we need better tools to study and change the program in binary form. I'm getting really sick of companies using binary blobs to obfuscate and mystify what is happening inside devices they produce. There is nothing magical to binary code, just like source code it's simply clear-cut instructions for the CPU what to do.
Sure, you lose higher-level abstractions when compiling, which are sometimes hard to reconstruct, and sometimes impossible, such as comments and documentation... But there is a lot of information that can be derived from binary form. See for example the extensive literature on producing exploits from patches, even automatically, by using smart diffing that understands the assembly.
Currently this is pretty much limited due to nonavailability of advanced tools. To do anything beyond simple disassembly one needs expensive decompilation tools and disassemblers, and specialized proprietary tools that security companies are using.
But given that (in practice) it is neigh impossible to rely on only open source, I'd love it if binaries got some more scrutiny from the public instead of "just swallow them and pray everything is OK".
From that point of view automatic updates which could be reviewed at the source code level are different from binary updates, although both serve the same practical purpose for the "non technical end user".