Netpets

[Amir Taaki]

Hi,

With all the Arduino experimenters at the Hackspace, it seems strange
that no one has experimented with artificial life.

Remember the Tamigotchis? These were a ludicrous craze at one point -
a small portable device with a virtual pet that you have to 'feed'
every so often (press A) or give love to (press B). I remember seeing
them first marketed as a way for children who live in small apartments
to be able to own a pet animal.

I soon got bored with mine and smashed it under a hammer. The thing
was lame and totally predictable. There was nothing life-like about
it. I think a large part of the appeal of pet dogs and cats, is there
anthropogenic qualities. They seem almost like innocent simplistic
children.

At one time in my life, I was amazed and obsessed with cellular
automata - particularly Game of Life. It is like a mini-universe with
its own physical laws. One thing I wanted to do was create artificial
life. After getting bogged down with the price equations and
population dynamics, and messing with gemini spaceships, I discovered
a rule change to Conway's GOL that allowed replicating patterns:

http://www.conwaylife.com/wiki/HighLife

The Life games are highly chaotic. But I think a less chaotic ruleset
with an added bit of randomness should be able to spawn patterns that
are very life-like.

And the rules for Life are very simple and able to run on an Arduino-
type device.

There is no definition for life (in nature) - it's kind of just
whatever we decide it to be. However one definition posits:

"Living organisms undergo metabolism, maintain homeostasis, possess a
capacity to grow, respond to stimuli, reproduce and, through natural
selection, adapt to their environment in successive generations."

Those first 3 items imply a kind of consumption. Normally life is fed
from the energy given by the sunlight. That energy is used to combat
and work against thermodynamics. However with artificial life, it
seems the device gets energy from batteries and the virtual
environment would only degrade if programmed in such a way.

So what can you 'feed' artificial life? Well I'd argue it is the one
thing humans are better than machines at doing - solving abstract
puzzles. Lets say the organism needs to evolve in some way. It could
be up to the human to try to resolve the puzzle (like those games on
the net for solving puzzles to help molecular biologists) which allows
the organism to either keep surviving in its current environment or to
evolve. Maybe neglect from solving the puzzles prevents the organism
from being able to remain in lockstep with the environment.

And perhaps you could link these netpets up to the net to interact
with others from the around the word. Perhaps they could somehow
primitively fight like Pokemon or mate to produce new organisms.

I think this is possible, and would be super cool. Not some boring
artificial construction of reality but something quite new and
realistic.

Do what you want with this idea.

[Paul Randle-Jolliffe]

Life can be pretty artificial

Marvin: Life? Don't talk to me about life!  

Marvin: Here I am, brain the size of a planet, and they ask me to take you to the bridge. Call that job satisfaction, 'cause I don't. 

[Ken Boak]

Amir

Some great ideas.

Could you use the randomness of Conway's GOL with a web connected 'duino to perform some random data selection task for you - to bring variety into a task or your life?

Thinking of how you could use GOL as a track selector for LastFM - to lead you into new musical genres that you would not have explored yourself.

Give it access to any database - like a list of London pubs/clubs or European holiday destinations - and it makes a choice for you. It would certainly bring variety to your life.

So use Conway's Life to help you change your life ?




Just a thought


Ken

[Clare Greenhalgh]

I think a large part of the appeal of pet dogs and cats, is there
anthropogenic qualities. They seem almost like innocent simplistic
children.

I have to say I find cats, at least, much more complicated than most people!

They are a law unto themselves, need attention when they say so and visits to the vet at the most complicated times ever!

If you want to use arduinos to create something as mystical as the world of cats I bow to your amazing creative knowledge! WOW!

Noko

--

[Yuan Gao]

I would definitely be interested in building a sort of genetics/natural selection aspect into something like this.

I was quite impressed when I read about the genetics system that The Sims uses to pass down traits (visual and behavioural).

Imagine Conway's GOL where colonies may randomly mutate new "rules", and pass these on to offspring 

[Amir Taaki]

We now have virtual life which is feeding off human ingenuity. What
exactly does this mean and how does it relate to the sun?

Proof of work was a proposed system to stop spammers:
http://en.wikipedia.org/wiki/Proof-of-work_system

By creating a small fixed cost per email, it would be infeasible for
spammers to send millions of emails. The way it works is that for your
email to be accepted, your computer has to solve a small mathematical
riddle. Solving this riddle entails using CPU cycles, which uses
electricity, and electricity costs money. There is a cost to solving
this riddle and getting your email accepted by the recipient. The
riddle can be any mathematical problem that's hard to compute by easy
(resource-wise) to validate.

A human solving pattern based problems set to it by this pet, would be
expending time and energy. This time and energy is related to money
(lost income) and the sun (brain power required). Therefore the owner
would be feeding the pet irreversibly in this one way energy
conversion. The consumption would be something directly visible in the
evolution and functioning of the virtual lifeform.

[Martin Johnston]

If any of you are interested in artificial life, I recommend you take a look at the work of Steve Grand, if you haven't already.

Back in the 90s(?) he developed a computer game ('Creatures') where the creatures in question not only learned from their environment (they had simple software brains built on a neuron model) but also had a genetic code that informed their structure, colour etc and was used during 'reproduction' to create genetically related offspring. Awesome.

He then tried building a robot orangutan (Lucy), wrote a book or two, moved to the US and now has a project on Kickstarter along similar lines.

http://stevegrand.wordpress.com/about/

As you might guess, I love this stuff! :)

M

[tom]

Creatures was pretty awful from what I remember, although you could pull their brains apart with 3rd party tools though :)

[IrradiatedHaggis]

Hi!

I played with artificial life a few years ago... It was very interesting. My experiment grew out of a simple single hidden layer back-propogation nerural net originally that I wrote in C++. I used that to develop a simple demo handwriting recognition engine... It occured to me that 'survival of the fittest' (genetic programming) could be an alternative to backpropogation learning in the net. So I pulled out my neural net code and placed it into autimata agents in a simple 2d virtual world.  I gave each agent the ability to rotate left/right, move forwards, detect food, detect mates, and detect danger. The agents were divided into Carnivores and Herbivores with simple rules... The Herbivores could eat plants that randomly occurred in the environment to replenish their energy (and therefore nearby plants appeared on their 'radar'). For the carnivors they could only consume herbavores.  Each creature had an energy level that depleted over time, upon reaching 0 the creature would die. The numbers were all configurable for energy levels but basically for Carnivores eatng a single herbavore could replenish 50% energy. However a herbavore only recieved 10% energy from plants. Thus the herbavores needed to eat much more to survive - but plants were much more common and they don't move... Upon encountering (into the same square) another identical creature (heravore or carnavore) they could mate if their energy level was over 20%. Thus, creatures close to death required food as a priority and only those who were fairly effective at gathering food would be able to propogate their lines.

Mating consisted of taking the nueral nets of the two creatures and averaging them together, along with a few random mutations to create a new agent. Thus the agents that were most successfull would mate the most often and impart their qualities on the children...

With all agents starting out as random nets most agents were imbiciles... I would see lots of creatures that walked around in circles endlessly... Occasionally an agent would develop that actively sought out death (a herbavore that chases carnivores...).   Most of the populations died out qiuckly. I only ever managed to get a few to last more than a couple of generations... But I think i was all in tweaking my numbers. I still believe I could potentially get a stable population if I got the balances just right and large enough starting sample agents to ensure that at least some of them have 'the right stuff' to survive in the world.

This project hasn't been worked on in years... The code was messy and very, very slow... With more than a couple hundred agents at any one time I'd be lucky to get more than one decision per second out of the nets. Some day I will revisit this... But I have about 30 other projects on the go at the moment...

Cheers,

Troy

Sent from Samsung tablet

Amir Taaki <gen...@gmail.com> wrote:


Hi,

With all the Arduino experimenters at the Hackspace, it seems strange
that no one has experimented with artificial life.

Remember the Tamigotchis? These were a ludicrous craze at one point -
a small portable device with a virtual pet that you have to 'feed'
every so often (press A) or give love to (press B). I remember seeing
them first marketed as a way for children who live in small apartments
to be able to own a pet animal.

I soon got bored with mine and smashed it under a hammer. The thing
was lame and totally predictable. There was nothing life-like about

it. I think a large part of the appeal of pet dogs and cats, is there

anthropogenic qualities. They seem almost like innocent simplistic
children.

[Yuan Gao]

I think there's something nobody has tried here.  "survival of the fittest" can be split into two kinds: "natural selection" and the artificial variety.  Natural selection being one in which survival of an individual is determined by how will it survives and propagates in some adversarial environment (either a physical environment like our own meatspace, or a simulated one); while the artificial variety applies some externally imposed (non-natural) set of rules that determine survival.  This latter case encompasses most "machine learning" systems like Haarlet-based facial or handwriting recognition.

What if we were to create a marketable product that evolved, and whose survival is based on sales?  Take this example: we create a batch of stuffed toys, each different from the next.  The ones that sell best are then used as a basis for the next batch of stuffed toys (with their features combined, and a few random mutations thrown in).  The survival of these toys are therefore driven by trends in the stuffed toy market, and the natural (or unnatural) selection means that the product line evolves towards one that sells best, as well as in theory being able to shift with market trends through its mutations.  I guess this similar to how product development works, but much faster.

[Martin Johnston]

Yep! IIRC the background graphics were bitmaps built off physical model sets he built at home - he sold them on eBay a few years back :)

Steve Grand is a real hacker though. When it came to constructing Lucy, the robot orang-utan, he built his own surface mount PCBs at home in his oven.

M