Welcome! Anthrobotics™ is a privately-funded R&D venture whose purpose for
more than 10 years has been to discover an innovative new approach for the
"strong AI" problem, as typified by machines like the fictional "HAL-9000"
or "C3PO" or "Commander Data". We believe we have achieved that purpose and
can demonstrate that.

This videotape consists of less than two minutes more about Anthrobotics™,
followed by a short and NON-confidential version of our PC computer
demonstration. Further details about us and our technology are included in a
document accompanying this videotape.

Generally, in building our Anthrobotics™ systems, we designed software
simulations of biological-type sub-systems like perception, recognition,
decision, action, etc. We began with pre-mammalian-type systems; and then,
adding to those, mammalian- and primate-type systems; and then, human-type
systems; and then, lastly, language systems.

Our Anthrobotics™ machines include an internal representational map or
setting capable of simulating the infinite variety of the natural world by
various combinations of a surprisingly small set of primitives. This
representational map or setting is constantly compared to a "relevancy"
network to determine current and predicted threats to and opportunities of
the machine.

Overall, the data structures and programs we use provide for extremely
efficient parallel computation, for straightforward learning and "common
sense" programs, and for easy correspondence with natural-language
structures.

Our computer demonstration is run on a slow, common-variety, 286-chip
computer using an EGA screen. The software, all produced directly by
Anthrobotics™, is Turbo Pascal compiling to about one-half a meg, including
point-and-drag interfaces and data.

We made no effort in this research platform to "reinvent the wheel" in side
areas like graphics, voice quality, and robotics, all of which remain
primitive. The software platform consists of many small "simple" systems
operating in simulated parallel processing to produce "complex" results.
Most sense inputs and motor outputs are fully simulated.

Four areas of abilities are demonstrated in order:

     (1) our universal representation abilities and to compute in simulation
     immediate future implications, both of which are required abilities to
     simulate human-like "imagination" and "what-iffing" for
     "understanding";

     (2) a simulated humanoid "living" in a viewable chaotic
     world-simulation and telling "relevant" true stories in English about
     its actual experiences, a required ability for episodic memory and
     "consciousness";

     (3) questions to and answers by the humanoid in English about its many
     reportable kinds of immediate and general knowledge; and

     (4) use of the humanoid's fully-simulated motor output to drive a
     mobile three-dimensional robot face which speaks and shows the
     humanoid's true emotions.

To demonstrate universal representation and simulation abilities, we put on
the screen a 2-dimensional world we call Fetch-World. We simulate a fenced
back yard about the size of the screen, in which yard one or more persons
and dogs are playing fetch with balls. We can move any thing when we want.
And we can delete any thing when we want. We can "run" the simulation
forward when we want to see what implications lie in any situation. We can
stop the simulation "run", move anything, and start again. We can stop and
modify by adding things ---- and then start the "run" again.

We can stop and quickly create or edit other things to put into our
simulated situation by use of a fast all-natural-language editing "meta"
interface. This "meta" interface is supported by another more detailed
natural-language editor used to selectively create the faster interface. No
programming is necessary. Using our meta interface, we can name our thing
and select distinguishments like shape, color, size, and temperature. And we
must do that in this platform since our default is to fully simulate sensory
input systems in the simulation of life forms doing recognition -- and since
our simulated human sends information from its decision system to its
sensory systems about what specific sensory inputs to "look-for" in the
particular circumstances, goals, and plans.

We can select the behavioral tendencies any simulated thing will possess by
using our fast "meta" interface. Going through this particular version, we
can quickly select "complex" behaviors like positions within particular food
chains and activities like ambushing, gardening, picnicking, or playing at
ball-fetching. And we can select uses for inanimate things such as "homes"
and "storage places".

So, to build a simulated lion to add to our back yard Fetch-World, we select
a lion-type shape and sensed distinguishments and then we select some
lion-type behavioral tendencies like eating-metabolism, "high" in a food
chain, inability to follow prey into its home sanctuary, and wandering when
not sleeping or acting with purpose.

Notice that our simulated dogs have been made "low" in the first food chain
and are thus a suitable prey for our lion. By the way, this program
automatically re-creates and randomly re-places any prey that is eaten, so
that steady-state ecologies may be primitively simulated.

We can place this lion in our world, and we would expect, once the dogs are
close enough to notice the lion, that they will care more about getting away
than playing fetch. Let's "run" the simulation and see ......... Since the
lion is now sleeping, we stop and will go to a more complicated world..

We call this world Africa-World. The problem to be demonstrated is to
provide human-like capability in the areas of understanding relevancy and
language well enough to tell "true and interesting" stories, including
feelings, about what's happening; this ability is also a necessary major
step toward human-like "consciousness" in its relation to episodic memory --
stories about one's "personal history".

The open window shows a close-up of just a portion of Africa-world. Another
window being opened temporarily shows this entire world in a smaller version
and permits fast relocation within this world of the viewed close-up
portion.

The things in Africa-World simulate rabbits, deer, trees, grass, natives,
lions, and one "StoryPal" character having many Anthrobotics™ human-like
abilities. In this world's food chains, remember that lions eat StoryPals
and StoryPals eat rabbits. Also remember that StoryPals have safe homes at
trees and that lions will not get too close to natives, who can kill them.
So in the initial conditions of the world as opened, we can move StoryPal
closer to the lion to help the lion catch StoryPal -- or we may move
StoryPal closer to a tree or native to help StoryPal escape the lion. To try
to encourage a "better" story, we'll place StoryPal to make it close either
way.

We'll open a life-signs window to view continuously StoryPal's level of
hurt, hunger, and tiredness, shown by different colors of bar-graphs. That
will help us to know StoryPal's major "metabolic" problems at any time and
judge how that affects plans and stories.

StoryPal, among its abilities, is capable of things like following plans,
doing "what-iffing" in planning, feeling "emotions", and using English (or
any other natural language) to talk about these things and to tell "true"
and relevant stories about it's unpredictable life experiences -- as that
life is happening.

We'll open a window showing StoryPal's face, animated by simulating
StoryPal's motor output to some facial muscle groups, so that we can judge
StoryPal's "true" emotions at any time. This window also has subtitles to
help you hear StoryPal's low-cost voicing system.
 

 Africa-World
Africa-World has run to the point where the lion is a threat to StoryPal.  StoryPal's face reflects both fear and tiredness.  

Notice that StoryPal knows when to start a story, when to end it, and what
not to say in between. As we run Africa-World for a while, notice the
life-signs bar graphs changing appropriately. [Run]

Notice that we can "stop the world" any time we wish and ask StoryPal
questions. We can choose to question StoryPal about the details of what is
now happening -- or about this world generally -- or about what StoryPal
remembers about its past life in this world -- or about StoryPal's usual
plans, decisions, and "look-fors" fed back to sensory systems.

[close-up of "Ask" window] [go through sample "Ask" program]

The last part of our demonstration shows a capability to go from simulated
motor responses to real-world action. One difficult area will be in face
control since we humans are so sensitive to faces. So we use StoryPal's
simulated facial motor output to drive a mobile three-dimensional robot face
which speaks and shows the humanoid's "true" emotions. You may wish to
compare the screen face with even this primitive robot face to judge facial
and emotional similarity as we now turn on StoryPal's robot face and
continue to run this world with story-telling ....

The Head

 The "skull" on the left has inside it the mechanisms that
 enable its parts to reflect the face on the computer screen.

The "face" on the right is the skull with the "skin" on it. 
Not shown is the skin with a silver-color coating on it so that
it looks more uniform than  the several shades of brown
seen on the skin on the right below,which also makes it
 looks more "robotic"

That concludes our brief demonstration of portions of our operating research platform