Agent and VR Specifications
Agent absolute minimum requirements:
has a unique identification
ability to perceive local environment
ability to manipulate local environment limited by physical laws
ability to communicate with nearby agents via text
rule-base list, ability to access list and add rules
initial rule-base list:
i am agent number: XXX
i can move
i can sense my local environment
i can manipulate my local environment
i can communicate with others nearby
i can add rules to this list
2-512 registers with 3-512 symbol capacity each
2 visualization registers of 27-10^12 bits each
connectivity pattern associated with each set
connectivity pattern between that and agent's rule-base
Conspicuously lacking are: motivation, morality, goal-list, and reality-models. The purposes for omission are: simplicity / ease of implementation - and - discovery of components critical to consciousness. The main purpose of this GP scenario is the latter .. Initially, i considered adding two rules to the base: i must interact with my local environment - and - i must learn from those interactions. But, that forcefully defeats the flexibility of the GP scenario. All agents must be free to do nothing (as all humans are free). Agents must have a minimal capacity as designed above. Initially, i was prepared to propose a flexible connectivity pattern between registers and senses including number of senses but.. That starkly restricts/narrows implementation themes. Admittedly, i have adamantly stated at least two cross-verifying senses are required for consciousness but.. We will use this chance as a process of discovery.
Virtual reality specifications:
it must have a rich and varied landscape including:
ponds, buckets, plants, seeds, sheep, manure, altar, knives, rocks, boulders,.. all obeying the laws of physics. Rules that can be learned: seeds grow on mature plants, plants require time to mature, seeds can be planted to grow more plants, seeds require appropriate water to grow, seeds require appropriate manure to grow, sheep need to eat plants daily to live, sheep can be killed, sacrificing a sheep at the altar produces a God's gift, sheep can make lambs under right conditions, too much water kills plants, too much manure kills plants, plants require appropriate spacing to grow properly, sheep produce manure regularly, manure decomposes over time, and both sheep and plants eventually die.
Obviously, if agents sacrifice all sheep at the altar, they'll learn that no manure = no new plants. So the village of agents must successfully manage the herd to have a successful village. The list of available rules above is not comprehensive .. Perhaps a later implementation could include a seasonal river flood that could be averted by cooperative behavior (such as placing boulders alongside the river). But that would be to investigate social values; this scenario is to investigate consciousness.
Individuals at the 'end' of a village (no plants, no seeds, no sheep, or successful management), can be recycled into restarted scenarios by reconstituting their structures individually or GP style, combining their structures with other successful structures - effectively taking the best of Adam's and Eve's qualities and making a new generation to observe.
Of course, realistic rules such as: an overly confined herd gets diseases and die, sheep normally need to drink water daily to live, manure takes time to compost,.. may be implementable but may be overly problematic if added initially. A sufficiently rich environment (but not overly complicated) is required for agent development. i see above as minimally complex/rich; simple enough for a child to understand and care for but complex enough to be valuable. Essentially, these agents will be virtual children - expected to learn to care for their village 'from scratch' (with no real memories or lessons learned from previous generations). Again, this is to evaluate conditions for consciousness. Please forgive the 'harsh morality' implied (thrive or die). Unfortunately, it's not much different from real life and our DNA.
At this point, i'm requesting simulation implementation assistance / independent collaboration. i've downloaded SeSAm which i believe will 'do the job' required above - all except multi-generation GP which must be 'by hand' with SeSAm. SeSAm is available at:
http://www.simsesam.de/ This link includes full documentation and executables. i've not received word yet from the 'GP crew' what public-domain software is available that does both GP and multi-agent simulation.. Again, if there's any interest whatsoever, please follow the link and join the group: http://tech.groups.yahoo.com/group/distributed_mind_project/