Here’s a plot depicting the average percentage of first person episodic memories over a period of a couple of months of simulation time. The vertical axis is multiplied by a factor of ten. A first person memory is one acquired from direct personal experience, whereas second hand memories are acquired from other apes via the telling and re-telling of anecdotes.

Over time the percentage of first person memories is slowly decreasing as anecdotes (scurrilous or otherwise) spread throughout the population and displace them in a competition for mindshare. This is the beginnings of a sort of persistent culture based upon accumulated shared experience.

 

Curiously, brain probes seem to become highly active for short periods of time. The plot shows the average brain probe activity for all individuals in the population over a period of a couple of months of simulation time. Activity counts as any change in the position or frequency of a probe.

Brain probes are links between the braincode program and the 3D cognitive simulation, which is a reaction-diffusion type system.  They transfer values between these two systems with some frequency in minutes.  The braincode program itself can specify how the probes are configured.

Why these spikes in activity exist I’m not sure.  They might be episodes where apes are grouped close together and chatting frequently (like a conference), or they could be some kind of positive feedback effect.  The spike towards the centre of the graph seems to show an exponential buildup of activity, which might be a particular kind of program spreading between apes via conversation and then halting after some time.

Since the braincode program is largely unconstrained it can support any arbitrary type of computation (including overwriting itself), within the limits of the time and memory space available, so all manner of curious computations can, and no doubt do, appear from time to time.

 

For anyone who is a newcomer to Noble Ape here is a high level overview of the genetics and how it operates.  This system is loosely, although not literally, based upon biological genetics. Currently there are no epigenetic switches, but that also might be added in later versions.

Chromosomes

Each individual has four diploid chromosomes, each consisting of 32bits with 2 bits per base. An example genome looks like the following:

1 2 3 4
GTGTTGAA CACCAATA TTTTTAAT ATTTTGCA
TGGCCAAA CATGATGC TGTGGACA CCGGTTCA

This may not look like much, but the genome could better be thought of as a program or a combinatorial system which can generate a larger set of information from a relatively small genetic origin.

Recombination

At conception we have the familiar crossover and mutation operators, along with the possibility for deletions or insertions. The crossover point for each chromosome is randomly selected.

In many genetic algorithms only point mutations provide a source of novelty, but here in addition a few types of transcription can also occur. With some probability, sections of the genome may be copied from one location in a chromosome to another, or between chromosomes. During transcription a chunk of genetic code may sometimes become inverted whilst in transit. This is a major source of genetic change and differences between siblings.

Genes

Gene values corresponding to, or influencing, phenotypic traits are extracted from the genome by taking pairs of bases from different locations and joining them together to make a single 4 bit value. The locations for the two bases are not necessarily sequential and are specified by regulator genes in separate areas of the genome. This means that phenotypes are a product of a web of inter-connected genes, and that an alteration in one gene might also have cascading effects upon others, resulting in non-trivial dynamics over successive generations. Changes in regulator genes may also have multiple effects on other genes, in line with the idea of punctuated equilibrium in evolution.

Some genes are sex linked. The first chromosome is inherited exclusively down the male or female line, such that a daughter inherits chromosome 1 from her mother and a son inherits chromosome 1 from his father. The outcome is that there can be sexual dimorphism in various characteristics, such as appearance and mating preferences.

Genes influencing physical appearance

These genes are sex linked, so that males and females may have different versions of the gene.

GENE_RATE_OF_GROWTH Influences the rate of the growth in height
GENE_PIGMENTATION Determines the fur pigmentation
GENE_HAIR Determines the fur length
GENE_FRAME Determines the body frame
GENE_EYE_SHAPE Shape of the eyes
GENE_EYE_COLOR Colour of eyes
GENE_EYE_SEPARATION Separation between eyes (binocular effect)
GENE_NOSE_SHAPE Nose shape
GENE_EAR_SHAPE Ear shape
GENE_EYEBROW_SHAPE Eyebrow shape
GENE_MOUTH_SHAPE Mouth shape

Genes influencing psychology

GENE_POSITIVE_AFFECT_FADE Controls the rate of decay of episodic memories with positive affect
GENE_NEGATIVE_AFFECT_FADE Controls the rate of decay of episodic memories with negative affect
GENE_BRAINCODE_SENSORS Determines the initial percentage at birth of braincode instructions related to sensing
GENE_BRAINCODE_ACTUATORS Determines the initial percentage at birth of braincode instructions related to actions
GENE_BRAINCODE_CONDITIONALS Determines the initial percentage at birth of braincode instructions related to if/then conditionals
GENE_BRAINCODE_OPERATORS Determines the initial percentage at birth of braincode instructions related to operators, such as add, divide, move, etc
GENE_BRAINCODE_DATA Determines the initial percentage at birth of braincode instructions related to data storage

Genes influencing behavior

GENE_GROOM Biases the tendency to groom others
GENE_AGGRESSION How likely the ape is to engage in acts of aggression. You can think of this as regulating rate of testosterone production.
GENE_SPEED Influences how fast the ape can move
GENE_STAGGER Determines randomness in locomotion. This is only used as a placeholder if the cognitive simulation is disabled
GENE_SWIM Influences speed of swimming
GENE_HILL_CLIMB Controls how efficient the ape is at climbing
GENE_LATENT_ENERGY_USE Metabolic rate

Genes influencing mating preferences

These genes are sex linked, such that males and females may have different versions of the gene.

GENE_STATUS_PREFERENCE Weights the preference for social status
GENE_PIGMENTATION_PREFERENCE Determines the preference for fur pigmentation
GENE_HEIGHT_PREFERENCE Determines the preference for height
GENE_FRAME_PREFERENCE Determines the preference for body frame
GENE_HAIR_PREFERENCE Determines the preference for fur length
GENE_MATE_SEEK Determines the extent to which the ape seeks out previous mates
GENE_MATE_BOND How strong the bond must be between two apes before mating can occur
GENE_INCEST_AVERSION The degree to which the ape avoids mating with its close relations (based upon pheromone)

Genes related to production of digestive enzymes

The efficiency of nutrient absorption for any food type is based upon a normalized weighted sum of these values. This helps to prevent an uninteresting scenario where all these values are driven to maximum, and creates trade-offs and the possibility for specialist grazers.

GENE_ENERGY_FROM_VEGETABLES Efficiency of energy absorption from vegetables.
GENE_ENERGY_FROM_FRUITS Efficiency of energy absorption from fruits.
GENE_ENERGY_FROM_SHELLFISH Efficiency of energy absorption from shellfish.
GENE_ENERGY_FROM_SEAWEED Efficiency of energy absorption from seaweed.

Inheritance

The double chromosomes also facilitates Mendelian inheritance, such that genes may be inherited but remain latent, with the possibility of them being expressed only in some later generation. This increases the informational carrying capacity of the genome, facilitating biodiversity. It also means that mutations or changes due to transcription may not necessarily be expressed in the current generation, but may turn up later.

Not by genes alone

There’s a fair amount of genetic determinism within Noble Ape, but genes don’t completely determine behavior or survival value.  Largely in parallel to the genetics there is a system of cultural transmission, supported by the cognitive system, which can evolve independently.

 

In a recent addition to the code, rumors about the activities of apes are able to spread in a more systematic manner than before.  Like the actor_index (see brain.c), which indicates which individual within the social graph is the ape’s current focus of attention there is now also an episode_index which points to the location of the episodic memory currently being accessed by the braincode program.

In essence, the ape’s thought process is not only a place where arbitrary code can be executed or overwritten but also there is the potential for a variety of actors and events to enter into the mix in unconstrained and perhaps non-intuitive ways.

There is now an additional braincode command, called ANE (anecdote), which during conversation allows an episodic memory to be transmitted from the speaker to the listener.  This includes the possibility for the message to become corrupted or misheard in transit, with a hard-coded probability.  Each episodic memory also references up to two protagonists.  In most cases the first protagonist is the self (first person perspective), but if the memory gets transmitted as an anecdote then this becomes a second hand memory in the minds of other individuals.  This means that it’s possible for an ape to have not only memories about its own experiences, but about the experiences of others – the beginnings of an oral tradition with the potential for features such as remembered ancestors and (due to accumulated errors in transmission) possibly even fictitious apes which enter into the popular discourse.

A consequence of the ability to harbor second hand memories are phenomena such as hearsay and celebrity.  Previously when two apes met and chatted the initial friend_or_foe value would be largely genetically determined by preferences for features such as height and fur color, but now hearsay familiarity is also weighted into this function.  If an ape has heard about another ape, via the transmission of anecdotes, but without having met in person then a friend_or_foe estimate can be constructed based upon the affect values for the relevant second hand memories (see social.c/social_celebrity).  In this way an ape’s reputation can precede it.

All of this spreading of gossip – whether scurrilous or celebratory – not only creates a complex social environment which the apes inhabit, but also has real consequences for survival, since mate choice is partially based upon factors which include friendliness.  An ape which justifiably or otherwise has gained a bad reputation might lose social status and struggle to find a mate, thereby decreasing its chances of passing on genes to the next generation.  Genetic extinction does not necessarily mean total informational death though, and simian bogeymen or villains could still linger as “ghosts” in the ideosphere of second hand memories.

 

New possibilities for AI

One observation over the course of my development on the Noble Ape simulation is that much of what I’ve done doesn’t feature anywhere within the traditional discourse about AI, and I think this is because usually AI is narrowly focused upon the workings of individual minds often in a way which is context free.  Once you start thinking about multiple minds, and how they might interact in an embodied context which can include birth and death then lots of other factors become more readily apparent.

 

Here’s a TED video about the curious nature of the self.

Anyone who is interested in AI or intelligent machines will inevitably eventually have to confront the nature of the self and its curious malleability.  As Ben Goertzel mentioned in “The Hidden Pattern” intelligence is really something which exists between minds as much as it does within them, and it’s in multi-mind simulators such as Noble Ape that these ideas can be explored.

The above video I think has some relevant points: the idea of the self as a constellation of elements which are not necessarily localized, and the idea of consciousness as an “illusion” or “trick”.  However I don’t think this goes quite far enough.  IMHO the “trick” is really a virtual machine, and the trans-personal nature of intelligence are mobile programs, or “programs in the wild”.  When the elements of the brain are arranged in some way and there is a means of communication with just enough bandwidth and generativity then it becomes a suitable initiator and host of virtual machines which can move around, combine, compete and so on.

As people such as Stephen Wolfram have found, it doesn’t take much to make a Turing machine.  My guess is that that computation is ubiquitous in all human cultures which contain a generative language, but largely unrecognized and unquantified.  Here I’m not referring to digital computers or modern infomatics – which are one example of computation – nor the genome which can be thought of as a Turing machine operating with 3D molecules rather than transistors or tape.  So I think that the self is a kind of virtual machine which may be highly localized (as in a hermit living in a cave) or more typically as a distributed entity across multiple minds, including those minds which have become fossilized within cultural artifacts.

The self also seems to have a scale free nature, such that the identity of a group can be similar to the identity of an individual, or vice versa.

© 2012 Noble Ape Blog Suffusion theme by Sayontan Sinha