My assumption: your system has this agent/guy moving around a home/office
environment. [Adjust question accordingly if not]

He has a simple task: "Move from A to B or D". But the normal answer "Walk
it" is for whatever reason no good, blocked.

The simple test is this: how many alternative ways of moving from A to B,
will the system (his brain) be able to search for and find?

[Sub-questions: how will alternatives be laid out in memory, and how will
the system search for them?]

That's the basic test of adaptivity : HOW MANY ALTERNATIVE WAYS OF
ACHIEVING ANY GOAL CAN A SYSTEM FIND?



Oh, Novamente will find a hell of a lot of different ways of achieving a
simple goal like that, even in its current form...

Its two main learning modules -- evolutionary learning and probabilistic
inference -- are both quite good at diversity generation...

But actually, your question bespeaks a certain unfamiliarity with standard
GOFAI type technology.  Generating a lot of different alternatives and then
pruning down the space is pretty standard stuff.

What is harder is as follows: if goals G1 and G2 look to be related, and the
system has learned a bunch of ways to fulfill G1; then, this latter fact
should make it easier for the system to find ways to fulfill G2 (than if it
hadn't learned ways to fulfill G1 already).  This is known as "transfer
learning" and has proved more challenging for AI systems than simply
generating diverse plans for the same goal.

Because you can generate diverse plans without arriving at a deep
understanding of the goal and the space within which it is situated; but
transfer learning, except in lucky cases, requires real insight...

Ben G

-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936

Reply via email to