X-Message-Number: 25557
References: <>
From: Peter Merel <>
Subject: The Limpinwood X-Prize
Date: Sun, 16 Jan 2005 22:44:48 +1100

I was sitting in my a donkey paddock below my teahouse at Limpinwood 
this afternoon thinking about the ANN material I'd waded through in the 
morning when I happened to notice a particularly amazing performance by 
a pair of butterflies. Bright orange they were and the aerobatics of 
all sorts they pulled were really astonishing. When they settled down 
on the herbage by my feet I was amazed to see them disappear - the 
outer coats of their wings were identical, absolutely identical, with a 
dried up leaf. All the brilliant orange was on the inside.

Now my paddock has grass up to my chin and wildflowers and weeds of all 
sorts. Tweed shire gets more rain than almost anywhere else in 
Australia, and my valley gets more rain than almost anywhere else in 
Tweed. This time of year the paddock grows over a foot a week; it 
really wants to turn back into a rainforest.

The pair of butterflies took off again and did another series of 
arabesques, loops, twists, and flings. Then they landed right back 
where they were before. The exact same grass blades. And this in a 
stiff breeze after a thunderstorm.

With the breeze there's no way the butterflies could have tagged their 
grass blades with pheromones to find their way back. Nor could they 
have anticipated the breeze's perturbation of their flight. Nor could 
they have used some kind of genetic map of the terrain - it looks 
completely different to just a week ago. They had to use their EYES and 
BRAINS to orient themselves in an incredibly complex and dynamic 
environment - my paddock - and somehow find their way back to their 
perch.

I read that butterfly nervous systems have less than 3,000 neurons. Not 
1 billion, like De Garis's planned "brain". Not even 75,000 like his 
Xilinx models. Just 3,000, and most of them don't even control flight 
muscles. But their interconnections can be mapped, if you like. It just 
takes a microscope and some patience. Plus the willingness to undertake 
the bad karma necessary to kill and dissect one of those little 
beauties.

It seems to me that the job of the butterfly orienting itself in a 
breeze in a paddock is considerably easier than the job of an autodoc 
orienting itself in a Brownian motion field in a living cell. After all 
the autodoc can't use vision. Still it would be a convincing step on 
the way to real AI / MNT. A lot more convincing than some animated cat 
stumbling about on a featureless surface.

Therefore I'm offering a prize to the first artificial neurologist who 
can achieve this task with under 3,000 neurons.

I'll offer you a night in my B&B here at Limpinwood. I'll even throw in 
dinner. You can think of it as the ANN version of the Ansari X-prize. 
You might say that a night in a B&B isn't worth a lot, but of course 
there'd be great gobs of military funding available to anyone who could 
do this. As for me I'm setting up the prize to try to focus ANN onto 
some kind of realistic step toward real AI. Or get the practitioners to 
give up and do something productive with their lives.

You don't even have to use butterfly wings - an RC helicopter would be 
as good. It just has to take off from a random spot in a breeze in my 
paddock orienting itself only by cameras, flit around the paddock for a 
couple minutes, then land back within 1 inch the same spot, all without 
guidance from any biological critter. Or using signs made with lasers, 
dye-bombs, colored tape, or any other distinctive external mark to 
orient itself or find the spot.

For a bunch of super-smarties like the ANN crowd this should be a 
pretty straightforward task. Unless, just possibly, ANN has nothing to 
do with the ontology of real neural networks.

Good luck to all!

Peter Merel.

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=25557