Superpositional thinking and Chinese rooms

Much of my thinking on certain topics is beginning to coalesce in interesting ways.  I’ve been having an ongoing discussion with Ken Wharton about quantifying entanglement and time symmetry, helped along in my thinking by the very patient Rob Spekkens and Matt Leifer.  Then, along comes Christopher Altman to tell me he’s actually been working on something along the lines of what I had floating around in my head (my thinking was partially spurred on by some of Kelly Neill’s posts). What have I got so far? Well, let’s take a look at the Chinese room argument again from a new perspective.

I think one of the main problems with the Chinese room argument is that it does not quantify understanding. What does it mean to understand something? Sprevak does not discuss this in his recent article. In a previous post, I argued that perhaps true AI would require quantum computers since real thinking likely involves superpositional thinking. But is that enough to provide understanding?

It isn’t if we use the Chinese room argument again. Superpositional thinking on the part of the English speaker in our Chinese room would mean he/she could simultaneously prepare several pieces of paper, C, that could be passed out of the room again. In essence, it means he/she could anticipate the input. But does this necessarily constitute understanding? Nope. Our English speaker still could remain blissfully ignorant to the meaning behind the Chinese symbols. So, in itself, superpositional thinking does not provide true understanding. All I can see by this argument is that it could provide a way to speed up the process carried out by our English speaker. But we already know that quantum computing is faster than classical computing for many types of problems.

For our English speaker to truly understand the ideas behind the Chinese symbols they need to be translated into English (or someone needs to explain them to him, which is essentially the same thing). So, does translation then produce understanding? Again, no. It is still possible for the English speaker to execute the program and he/she at least understands more of the symbols, but it is akin to recoding the program into the native language of the operating system (the English speaker). But there are plenty of things written in English that I don’t understand. I may understand the individual words, but I don’t understand the way in which they are combined. For a rudimentary example of this, imagine I’m a spy and intercepted the code “The jackal screams at midnight.” I know what all of those words mean individually, but I have absolutely no idea of their context. It seems to me, then, that true understanding requires context. As it turns out, Aerts, Broekaert, and Gabora considered this about eight years ago. More recently, Svozil has made in-roads into this exact process by proposing a context translation principle.

Thus, I suspect true artificial intelligence will, perhaps, not only need superpositional thinking but also some sort of context translation ability.

Now where does time symmetry come into all of this? Well, I’ll hold off on that conjecture for the moment but I will say that I think, at the microscopic level, the universe is perfectly reversible and time-symmetric but that this doesn’t scale well. More on these thoughts later…

~ by quantummoxie on February 23, 2008.