[alicebot-archcomm] Coming up to speed.

[Jonathan Roewen]

> I am especially interested in creating A.L.I.C.E.- bot projects that can
> deliver "mission-critical", or "near-mission-critical" services to company
> employees and to their customers.  To meet this goal, I find it useful to
> see A.L.I.C.E. as the wrapper, the coordinator, the mediator, and the
> director of a host of services.

In part, I agree about this role of ALICE, as it allows the formulation of
the services in (mostly) natural language.

As developer of J-Alice, we've recognised this to some degree, and are
currently moving towards a server/client architecture, where the server's
only role is the processing of AIML, putting the remaining work on separate
clients which will (hopefully) be able to communicate with the server
software in a variety of methods, including TCP/IP.

> I mention all of this because I have a sense that A.L.I.C.E. bot projects
> could benefit enormously from the power of a clean and efficient way to
> supplement bot templates with returned results from calls to .cgi scripts,
> and even simple .html documents, using the standard "http:"
> protocol.   Now, I am aware of the <SYSTEM> tag and I believe I know what
> it can do - but I wonder if anyone has devised a <SYSTEM> to http:-world
> interface for unix, linux, windows and mac yet - and if so, I wonder how
> costly are shell calls to system resources and performance.

Shell calls in general would be somewhat costly, as it usually requires
instantiating an environment for the call to run in for each invocation ..
though the performance cost is obviously operating system dependent.

What may become the standard way of doing things is to write AIML
categories, using markup from other namespaces, such as XHTML, and allowing
further processing to be done elsewhere, breaking down the work into clear
and distinct blocks, which should eventually lead to better overall
performance (especially in multi-processor/distributed systems).

> I had some recent exposure to TELLME's voice recognition phone network
> (which has as its primary client AT&T 800 Directory Assistance -
automation
> that listens to requests for listings and speaks back the phone number),
> and it seems to me that what really leverages the power of their systems
is
> their interface to the http: world through their <SUBMIT> command ( see
> http://studio.tellme.com/vxml2/ref/elements/submit.html ) .  This empowers
> their developers to reach out and grab data from anywhere in the Internet
> (or optionally WAN) to supplement their responses.  WOW!   THIS IS
> POWERFUL.   I hope A.L.I.C.E. has a way, or will have a way, to accomplish
> the same.

Most of this ends up being interpreter specific, as certain implementations
may not provide the required facilities to achieve such tasks.  As developer
for J-Alice, I hope our interpreter will be capable of providing such
facilities in the near future, and am also looking forward to see the future
of ALICE & AIML.

Jon