Tuesday, July 26, 2011

The Chinese-Room Fallacy

I haven't been posting much lately because of work, but I came upon this paragraph and had to mention it:

This raises the question of whether Watson is really answering questions at all or is just noticing statistical correlations in vast amounts of data. But the mere act of building the machine has been a powerful exploration of just what we mean when we talk about knowing.

The above is a very good example of the typical problems one runs into when one pretends that an ill-defined term can be definitively compared to a well-defined term (or another ill-defined term, for that matter), and is representative of the problems Wittgenstein and Russel saw in philosophy.
Here is a breakdown of the problems with the paragraph:
  • Watson is clearly answering questions. He has been given questions, and provided their answers, so there is no question there. There is a question about whether or not he is thinking, which is by no means a necessary tactic for answering questions.

  • By the understanding of neurobiology of the past few decades, thinking and knowing appears to be statistical. While human brains clearly don't have the logical-reasoning capacity provided by the reasoning engine within Watson, the implied distinction between 'knowing' and 'just noticing statistical correlations' is quite probably imaginary.

  • Noticing statistical correlations is enormously difficult, especially on the scale upon which it occurs in this case. Saying that a machine is 'just noticing statistical correlations' is like saying that a turtle is 'just moving an eighteen wheeler around with his mind'.

As the title of the post implies, I consider this kind of confusion the source of other 'controversies' like the Chinese Room problem.

Saturday, July 2, 2011

Project xanadu document dump

This has only the vaguest relationship to the content of this blog, but I'm posting it here because it's the place I'm most likely to look later.

A little context: I am a huge fan of Project Xanadu, and a ridiculous percentage of my personal projects have to do with variants on the Xanadu designs (mostly xu88 and zigzag). Going through my scribd feed I discovered that I at some point gathered and repackaged a bunch of old Project Xanadu documents and stuck them up there. I have the tendency to download every pdf I see, and I have a tendency to on occasion upload every pdf I have hanging around that looks public domain to scribd so they don't prevent me from downloading pdfs from them. So, if I don't link it up again, I will probably lose track of it for another few years.


Note on copyright: IP belonging to Project Xanadu, Xanadu Operating Company, etc., is actually even more complicated than normal IP concerns. Early Xanadu material was made by an ever-refreshing group operating under 'Plan R doctrine': the idea that if one member lost contact with the rest of the group for a long period and no release seemed imminent, that one member would be expected to release the material he or she had in his or her possession to the public regardless of who wrote it. This conflicts with the fact that everyone involved kept copyright on their work, and various members of the group have different ideas about what qualifies as long enough for Plan R to be enacted, how little contact is necessary, and what constitutes a reasonable attempt to contact the rest of the group. Furthermore, XOC both operated on its own and was at one time a subsidiary of Autodesk, which may or may not affect copyright claims for material that was in XOC's possession during the Autodesk period or made during the Autodesk period by XOC. So, there is a fairly long history of one member (usually Ted Nelson) blessing an independent implementation or a particular release and then another member coming out of the woodwork unexpectedly with a suit or a complaint (this happened with gzz -- which is now known as FenFire and does not support zzstructures -- and may or may not be involved in the problems with the Udanax Gold release, which is source code in a SmallTalk-like language that compiles to C++ but for which no working compiler can be released).