 |
|
| Computers Forum Index » Computer Artificial Intelligence - Genetic » The Robotic Scientist... |
|
Page 2 of 2 Goto page Previous 1, 2 |
|
| Author |
Message |
| Wolf K... |
 Posted: Mon May 25, 2009 4:10 am |
|
|
|
Guest
|
Curt Welch wrote:
Quote: Ian Parker <ianparker2 at (no spam) gmail.com> wrote:
If you are going to have a definition of "Intelligence" it must be a
definition that has RIGOUR.
Why? Language has millions of words, most of which are defined with no
rigor.
[...]
True of all human languages taken together, false when referring to any
one language. Besides, the vocabulary of any speaker of a language is
much smaller than the lexicon (== all the words ever used by
speakers/writers of that language.)
cheers,
wolf k. |
|
|
| Back to top |
|
|
|
| Ian Parker... |
| Posted: Mon May 25, 2009 10:43 am |
|
|
|
Guest
|
I think you have to try to reach a rigorous definition. When I talk
about "fighting Israel" I am, of course, making a subjective
assessment as well. I seem to recall a conversation on concrete.
http://www35.wolframalpha.com/input/?i=concrete
I must say that I had a drainpipe that came loose and I needed to mix
up a small quantity The packet told me 5-35C. In England temperatures
are rarely >35 although in Aswan they are frequently in the 50s. Also
a small quantity can set hotter than a large quantity that generates
internal heat.
Point is both Google and Wolfram Alpha have all this in their
information store. Wolfram is in some respects getting close to
understanding and what we would regard as AI.
Google translates without knowledge. Google and W in fact HAVE the
knowledge stored away. Wolfram BTW can give you the weather in Aswan
on any day in history!
We need rigour for a number of reasons. First of all we need to know
how well we are doing. How good is out translation? Bleu gives you an
assessment of this. The NIST hold an annual translation competition.
Arabic, Chinese, Urdu -> English. This needs to be judged fairly.
In Mathematics the search for rigour often gives us insights. You are
talking about learning and reinforcement. If we can quantify this we
will be able to develop algorithms. Certainly any idea of
reinforcement leads to a different distribution of "E" as in exp(-E/
kT) throughout phase space.
In reinforcement we have a number of convergence criteria. We can
examine how the phase space has changed after a number of attempts
resulting in convergence. WE need to know what the important
parameters of phase space are.
Let us take an example. A great many robots use stepping motors. These
in effect advance one step at a time. The robot is constrained to move
slowly as it has to stop in one click. Enter a robot with phase space.
This robot can move rapidly. It does however have a starting point and
a finishing point. The algorithm is to get to the finishing point with
minimum cost in terms of a balance of energy and time. There will be
regions of phase space where the robot is going to overshoot (whatever
inputs are put in). We define E to be a region we do not want to be
in.
In a similar way bipedal balance may be defined in terms of phase
space. An action like running or walking is effectively cyclic motion
in phase space. We can define this and iterate to it.
We are discussing primevally language here rather than robotics.
Language fits in well with Kolmogorov. How does robotics fit in? It
does, but not quite so obviously. It seems amazing to me that we have
not made better advances in robotics than we have. I think we need to
look at phase space.
- Ian Parker |
|
|
| Back to top |
|
|
|
| Ted Dunning... |
| Posted: Sun Jun 28, 2009 7:09 pm |
|
|
|
Guest
|
On Jun 27, 12:31 pm, Mok-Kong Shen <mok-kong.s... at (no spam) t-online.de> wrote:
Quote: Mok-Kong Shen wrote:
BTW, I find it difficult to imagine that a "pure" statistical
method could determine the parts of speech of an "entirely"
unknown language. [snip]
In fact Rao et. al. have with statistical techniques only shown
that it is very plausible that the Indus script is a natural language.
In fact Rao et. al. screwed up pretty seriously and didn't show
anything at all.
See here for a summary of the debunking by Mark Liberman, Richard
Sproat, Fernando Perreira and others. |
|
|
| Back to top |
|
|
|
|
|
All times are GMT
The time now is Mon Nov 30, 2009 1:45 pm
|
|