James Rogers wrote:
> In short, we mostly do what you are talking about, but you've actually
> over-estimated the difficulty of integration of domain-specific modules
> (using our architecture, at least).
...
> The
> integration is a freebie that comes with the underlying architecture,
> not something that I spent much effort designing.

Interesting!

That is not the case in Novamente, which is a multimodular architecture in a
slightly different sense than your system...

In the Novamente design, we have two different types of modularity:

-- multiple MindAgents carrying out different sorts of cognitive functions

-- multiple Mind-Units (each potentially containing more than one machine),
each carrying a mix of MindAgents and oriented toward a particular function

There are then two kinds of integration: integration of Mind-Units into an
overall Novamente configuration, and integration of Mind-Agents into an
overall cognitive process.

(In Novamente, we actually are working only with a single simple Mind-Unit
right now, but the design is general in this sense; and in the Webmind AI
Engine, a somewhat similar design, we had multiple Lobes, the analogue of
multiple Novamente mind-units).

Actually, the *software* aspects of integration come for free with the
underlying architecture.  this holds for both kinds of integration.  this is
tricky in some technical points, but not fundamentally difficult.

And, integrating Mind-Units to form an overall configuration is really not a
big deal either.

The substantial thing is integrating the MindAgents with each other.  This
is not hard theoretically because they have all been designed to work with
each other.  But the parameter tuning issues can be tricky, as the optimal
parameter values for one mindagent may depend on the context it's working in
(including the other mindagents involved).  Automatic adaptation --
"cognitive homeostasis" as we call it -- plays a major role, and this is
another thing we did a lot of work on in Webmind, but have barely started
with in Novamente....


-- Ben G


-------
To unsubscribe, change your address, or temporarily deactivate your subscription, 
please go to http://v2.listbox.com/member/

Reply via email to