AI and NLP...
On this occasion of the 26th aniversary of my living in my parent's
house, ( =( ) I want to share some of my latest underinformed thinking
about AI with a focus on NLP which seems to be the hot topic of the day.
The dogma with NLP, as with many other types of AI seems to be that "if
we add enough rules it will work." or, more subtly, "If we simulate what
we beleive to be the top-level algorithm of the brain well enough it
will work". The first approach and even the second is doomed to failure.
The origin of this dogma is found in a variation of the same dualistic
thinking behind a vast array of human expression. It can be found
everywhere from the religeous beleif in a soul, to the uploader's
equally fanciful beleif that the mind can be seperated from body through
technology. In this case dualism means that there is a part of the brain
called the "higher functions" located exclusively in the cerebral cortex
which are entirely seprable from the rest of the brain which are
designated as "lower functions".
What is overlooked is that the human expressions that our NLP systems
have the most trouble dealing with are, infact, "telemetry data" about
the state of these so-called lower functions. Only a being which has an
internal state roughly analogous to what we call "happiness" can
understand the english sentance "I am happy." in any meaningful way.
Therefore the projects that are most likley to achieve some level of NLP
are those which place an emphasis on the brain's varrious subsystems
such as is required to produce androids.
President Bush's *head* is _F_L_A_T_.
- Congratulations upon your 26 years of residence!
AI researchers have pondered over the problems of language
understanding for decades. My own observation is that most
researchers definition of "language" is specifically written english
text. However, "language" is really a broader spectrum of skills
beyond just reading of symbols and manipulation of written text.
Things like body and face movements and prosody all make up a
significant amount of human communications.
Being able to read and write is a very important skill - especially
if you want to navigate the internet - but it's not the only
communications skill and indeed our ability to interpret written text
may in turn be heavily dependent upon more direct sensory experiences
which don't involve text processing. The narrow concentration upon
text processing is probably a legacy of Alan Turing's original
thought experiment of an immitation game, where he imagined only
teletype machines could be used. In the modern world when you look
at advertising, movies, TV and so on you can see many examples of the
use of non-witten communication in order to try to influence our
- Speaking of NLP, complexity and the limits of computational
intelligence, this paper:
argues that "[the] lack of a language implies also that a computer
cannot have a theory of mind,which is necessary in order to be able to
know that other individuals may have different views of reality, that
is, that my beliefs not necessarily are shared by other individuals."
In short, it says that machines can never do what humans do.