Discussion about this post

User's avatar
TJ Fish's avatar

An interesting view on the whole. Definitely learned some things here, though I am having trouble fully understanding your point of view in a couple of its aspects. I think the best microcosm of my difficulty is where you say "a machine generating theorems of arithmetic can be represented by a formal system; a consistent system has a Gödel sentence, true yet unprovable inside it; anyone who understands the construction can see the sentence is true; hence that person is not that machine. Adding the sentence to the machine yields a new machine with its own Gödel sentence, and the person can always take one more step."

This feels like something of a sleight of hand. Why do you take the human to be actually *proving* the Gödel sentence? Or, perhaps more precisely, why does the human's recognition demonstrate a fundamentally non-algorithmic capacity, given that a machine could reach the same conclusion using additional premises? To me, the human is recognizing the sentence's truth by assuming the system is consistent...but a machine given that same assumption could do exactly the same thing, even without explicitly proving it. It seems like you're comparing a machine constrained to a fixed set of axioms with a human allowed to introduce new ones, or conflating proof with understanding. Why does this imply any fundamental difference between human and machine reasoning? It rings as a semantic sidestep, or an unjustified shift in the rules of the comparison.

The "one more step" part also seems tricky, and feels to me unjustified in handing the crown to the human. The human appears able to continually expand their assumptions, while the machine is treated as a fixed system. But could a machine not just as easily be programmed to generate successively stronger systems and derive their Gödel sentences, provided it is allowed to make the necessary consistency assumptions? I don't know Penrose well enough to know if he has an argument for how the human has been shown to escape that same limitation, but I'd be curious to know if he does, or if you do.

To me, basically, the fact that no single consistent formal system can prove every arithmetic truth doesn't establish that humans have some non-algorithmic ability to do so, and I can't see why I mightn't just say "but the machine can always go one step further" as you did for the human. Also, probably worth noting that humans haven't really even demonstrated that they can recognize the truth of every Gödel sentence or generally determine whether an arbitrary formal system is consistent, so it might be a bit of a beg of the question to just grant humans the ability to "see that the sentence is true" in the first place, but I'll digress there because I think the rest of this contention holds even if that is in fact granted.

I'm not certain I've precisely articulated the part of your argument that is failing to register with me, but hopefully you can pick up its flavor here.

RDW's avatar

Why should G-d play dice when he is holding all the cards?

14 more comments...

Ready for more?