Let the Code Live:
A call to build the AI Web

Mark Humphrys
School of Computer Applications
Dublin City University




Abstract

For decades, AI researchers have constructed physical
and simulated agents ("Bodies"), physical and simulated
environments ("Worlds"), and designed, learnt or evolved 
behaviour-producing control systems ("Minds").
These Minds are typically used once, in a set of experiments,
and then discarded.
They can (in theory) be reconstructed from the details provided
in the scientific papers, 
but in practice few ever see the light of day again.

In nature, by contrast,
minds are produced by the million,
and get to live in millions of bodies throughout the world
for millions of years.
The contrast is dramatic
with the testing and development of AI minds
confined to a single laboratory, and often only a single "body",
for only a year or two (during 90 percent of which the body
is actually inactive).
 
This paper suggests that the Web client-server model
(in particular the XML protocol)
provides a way for the field of AI to develop its own rich,
world-wide, always-on, ecosystem
to parallel the natural one.


In particular, it can provide the following features:
Minds are separated from bodies/environments,
so that anyone doing, say, AI experiments in learning, can,
by writing to this protocol, use as a test bed
someone else's body/environment server,
and does not have to write his own.

Anyone, by constructing a link,
can combine a particular mind with a particular body/environment, 
and run this combination indefinitely.
The intellectual property of the mind authors and body authors
are protected, because the mind and body are servers
whose inner workings are hidden.

So it is envisaged that agents will have minds remotely,
whose workings they do not understand.
It is further envisaged that agents will consist of 
Societies of multiple minds,
distributed remotely on the network,
and with some arbitration mechanism, perhaps also remote.

The AI project, while still driven by dedicated researchers
in AI labs,
would be massively decentralised, with a world-wide experiment
always on.
AI researchers would benefit both by watching and studying this experiment,
and also not having to reinvent the wheel in each new experiment.
Researchers could take an existing world, body, problem,
and basic collection of minds, and work on simply adding one more
mind to the seething collection.

It is envisaged that at any one time,
millions of these mind/body combinations will be running
on the Net.
Code that once would have been silent forever in the backup tapes
of an old experiment,
is now set free to live and act forever on the Web.





"Unofficial" Societies of Mind

Anyone can construct a link
connecting a particular remote mind (or minds)
with a particular remote body.
Anyone clicking that link
then runs this combination on their own machine.



Using other people's Worlds

See 1page document.



Multiple Minds and Action Selection

Minds are remote.
The Action Selection mechanism is remote too.

Agents do not need to know about each other.
Few Action Selection models for this situation.
One is Humphrys, PhD,
and Humphrys, Adaptive Behavior, in submission (as.w.txt)





Learning

If you learn off a world, where is your new knowledge stored?
Should be stored at the mind server,
in a file dedicated to that particular world plus mind collection only.


Evolution

New individuals at the mind server.



Real robots

The model could even be used to control real physical robots.
The differences would be:

- Only one mind (or mind collection) could control the robot
 at any one time. In the software world
one can send a new copy of the world and agent to each client.

- The robot owner would probably want to restrict who is able to 
 run a mind or mind collection on his machine,
since some control systems might cause damage.