A Long Term Proposal to Simulate Consciousness in Artiﬁcial Life

Joseph D. Horton

, Michael Francis

and Eckart Sußenburger

Faculty of Computer Science, University of New Brunswick, 550 Windsor Street, Fredericton NB, Canada

Bonn-Rhine-Sieg University of Applied Sciences, 53757 Sankt Augutin, Germany

Keywords:

Artiﬁcial Life, Simulation of Consciousness.

Abstract:

Computers will soon be powerful enough to simulate consciousness. The artiﬁcial life community should start

to try to understand how consciousness could be simulated. The proposal is to build an artiﬁcial life system in

which consciousness might be able to evolve. The idea is to develop internet-wide artiﬁcial universe in which

the agents can evolve. Users play games by deﬁning agents that form communities. The communities have to

perform tasks, or compete, or whatever the speciﬁc game demands. The demands should be such that agents

that are more aware of their universe are more likely to succeed. The agents reproduce and evolve within their

user’s machine, but can also sometimes transfer to other machine across the internet. Users will be able to

choose the capabilities of their agents from a ﬁxed list, but may also write their own powers for their agents.

1 INTRODUCTION

A great deal has been written about conscious-

ness in the last few years. A few of the books:

(Chalmers, 2010),(Deacon, 2012), (Dennett, 1991),

(Grim, 2009), (Pinker, 2009). Yet not much is agreed

upon. A quote we like by Sutherland found in (Gaz-

zaniga, 2011), from the 1989 International Dictionary

of Psychology (Sutherland, 1989).

Consciousness: The having of perceptions,

thoughts, and feelings: awareness. The term

is impossible to deﬁne except in terms that

are unintelligible without a grasp of what con-

sciousness means. Consciousness is a fasci-

nating but elusive phenomenon; it is impossi-

ble to specify what it is, what it does, or why

it evolved. Nothing worth reading has been

written about it.

We are not so pessimistic as the last sentence, al-

though we do not think a great more is either known

or agreed upon now. All the books referenced above

have interesting things to say about consciousness.

But they do not agree upon a deﬁnition, in fact most

do not even try to deﬁne it. The problem as we see it

is that there is very little with which we can actually

do experiments. It is difﬁcult to study consciousness

in living animals. We can only watch and see what

actions agents take in certain circumstances. Not ev-

eryone agrees that animals are conscious.

As for humans, they can tell us what they are con-

scious of, at least some of the time. But from a log-

ical viewpoint, we cannot even prove to one another

that we are conscious, that we are not “zombies” that

are only feigning consciousness without actually be-

ing conscious. People usually assume that others are

conscious, if only from politeness. Now if you believe

that strong AI hypothesis (Searle, 1997) page 9, that

the mind is just a computer program, there is no con-

tradiction. Acting as if conscious and claiming to be

conscious implies consciousness. There is no possi-

bility of such a thing as a Zombie that acts as if it were

a conscious being and says that it is conscious, and

is not conscious. Searle has the opposite viewpoint

and thinks that simulating consciousness in a machine

is irrelevant. “... the behavior by itself is irrelevant”

(Searle, 1997), page 204. We think that if some com-

puter program tries to take over your bank account for

its own purposes, whether it is conscious or not is ir-

relevant. Of course this is a different probem from

what Searle was discussing.

We ignore the “hard problem” of explaining the

conscious experience. We leave that to the philoso-

phers. Our goal is to study consciousness, or at least

the simulation of consciousness, from an experimen-

tal computational viewpoint. We intend to build arti-

ﬁcial universes in which the agents demonstrate many

of the properties of conscious beings, whatever peo-

ple think of as properties of conscious beings. Of

course the animats will not be able to discuss the real

world, but we would like them to become able to dis-

cuss their own artiﬁcial universe, among themselves.

389

D. Horton J., Francis M. and Sußenburger E..

A Long Term Proposal to Simulate Consciousness in Artiﬁcial Life.

DOI: 10.5220/0004332903890394

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 389-394

ISBN: 978-989-8565-39-6

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

This paper is essentially a project proposal, rather

than a scientiﬁc research paper, although we are build-

ing such a system. The project is too large for a small

research team. We need to have help to make it suc-

cessful. The project was originally proposed in (Hor-

ton, 2010).

2 AN OVERVIEW OF THE

PROPOSAL

2.1 Why now?

It may be that computers will be powerful enough to

simulate the human brain in the next few years. We do

not think that we are close to being able to simulate

the human brain, although there are people like the

Blue Brain project, that are trying. It is known that

the human brain contains close to 10

neurons and

that neurons on average have about 10

connections

between them (Azevedo and et al, 2009). Such con-

siderations has led to speculation that a human brain

could be simulated in 10

ﬂoating point operations

per second, which is about the same as today’s fastest

supercomputers. We think that this is likely to be a

low estimate. There has even been some evidence that

not all thinking occurs in the brain itself.

There are several reasons not to be unduly inﬂu-

enced if you think that this is a severe underestimate

of the computational power of the human brain. The

ﬁrst is Moore’s Law type arguments. The computa-

tional power of computers is doubling every year or

so. Just wait a few more years, and the computational

power will be there. Another is that most of the hu-

man brain is not being used for consciousness. Only

19% of the neurons of the human brain are in the cor-

tex (Azevedoand et al, 2009), which is the most likely

part of the human brain in which consciousness is to

be found. A third reason, if you believe that animals

are conscious, is that many relatively intelligent and

social animals, such as the crow, have much smaller

brains than humans, yet have to deal with a very com-

plicated physical world.

Is it possible for a computer system of some kind

to become conscious on their own? Not likely. But as

algorithms become more nondeterministic, and more

and more autonomous agents are being written and

sent to do their jobs online, it would be nice to have

some evidence, however weak, that the computer sys-

tems are not likely to appear to be conscious on their

own. If it turns out that it is relatively easy to sim-

ulate consciousness, that would be a very important

discovery.

2.2 How to Start towards

Consciousness?

The only beings that we currently recognize as con-

scious were produced thru evolution. We propose to

use an evolutionary approach. We insert autonomous

agents into an artiﬁcial life environment, and try to

stimulate them to simulate consciousness. We do not

think that this will be an easy task. It is not at all clear

that there is a good reason for consciousness to have

developed in the real world. What advantage does it

give to an animal, or to a species?

If an animal lives in a community of other animals

that it interacts with in multiple ways, it is certainly

to its advantage to be able to predict what one of its

companions is going to do in certain circumstances.

To do that it is useful for the animal to feel the same

as its companion, and perhaps be able to predict what

it itself would do in the same circumstance.

The ﬁrst steps to move in this direction is to de-

velop the four C’s: conﬂict, co-operation, commu-

nication, community. They may not be essential for

consciousness, but they seem to be a start. These

should be possible to make in our proposed universes.

We have not yet thought much beyond the stage

of building communities. Building communities is a

goal for the intermediate term.

2.3 The Gaming Interface

We do not expect that we will be able to simulate con-

sciousness by ourselves. We hope that we can enlist

others to aid us. Nor do we expect that the goal will

be obtained in a short period of time. We need a great

deal of computational power for many years. It might

take several decades. The internet becomes the only

real source of such resources. Ideally we need mil-

lions of computers working for many hours a day. So

like seti@home or the protein-folding online game,

we want to attract non-scientists to work with us, and

lend us

their computers for us to work on. We need some-

thing to give them in return.

Many computer/internet games have an artiﬁcal

life component. Will Wright’s SPORE is the most

obvious example, and it even has an evolutioinary

component. Several game types could be developed.

Can you evolve a bug, or community, with some

propetiy or properties? One could have contest be-

tween bugs/communities.

A very important point here is that the agents

evolved must be available to the project, so that any

useful property that is evolved or developed can be

made available to everyone. Not only should the

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

390

players allow bugs to evolve on their computers, but

also they should be able to write code, what are

called powers below. Probably a restricted scripting

language is needed. These user-developed powers

should slowly become available to other players. In

fact the possibility of serendipitous interactions is an

essential idea of this paper.

3 THE ARTIFICIAL LIFE MODEL

We are writing a system called SOCIAL, simulation

of consciousness in artiﬁcial life. It has many of the

features that we have thought about, but is far from

complete.

3.1 The Universe

The universe should be as simple as possible, so that

the brains will not require much resources to han-

dle the physics of the universe. What does a uni-

verse require? A being must be located somewhere

in a universe, so locations are required. From a lo-

cation a being must be able to move to some other

locations. This suggests that the universe should be

a graph: locations are nodes; connections between

nodes are edges. Adding geometry just makes the

universe more complicated. Conceivably even sim-

pler universes without locations are possible, with the

bugs only knowing facts about the universe, but this

would be harder to describe as a universe.

A graph structure is very simple, yet it allows for

many different possible universes with many differ-

ent properties. A graph can mimic a two, three or

higher dimensional space, using any ﬁnite tesselation

of space R

. A grid graph could be used, to make

for easy display. Or one could take any random set

of points in the space, and construct the Voronoi di-

agram, with the cells being the nodes. We have ex-

perimented with very simple universes, such as a cy-

cle. Or the universe can be the union of many dif-

ferent types of graphs, possibly with gateway nodes

connecting the different components.

Each node can contain beings and resources.

There are caps upon the contents of a node, decided

by the experimenter. The ﬁrst resource that we added,

we called energy. We have toyed with the idea of

adding conservation laws, but have not done so yet. It

could be done by specifying that whenever a resource

is used, an equal amount of some other resource is

created. We have not considered any entropy law, that

would require the universe to run down.

We have considered making an expanding uni-

verse, in that new nodes could be added at random

times and places, but have not done so yet.

The beings must be given the power to move be-

tween neighboring nodes. Resources as we have

imagined them cannot move, although this is a fea-

ture that could be added.

Time we have decided should be discrete, like

space. In our experiments so far we have allowed the

beings one action per time step.

3.2 Bugs

The beings in the universe are called bugs, beings

whose universe is a graph. One can also think of them

as being like viruses, although viruses are much more

complicated than our bugs so far.

The bugs have multiple powers. Powers are abil-

ities that a bug can use. Each bug has three different

kinds of powers: actions, like move, eat, turn, pickup,

drop, which they can do; sensors, like bugsensor (how

many bugs are there in neighboring nodes?) that re-

turn information about their environment; and a brain,

which decides what action they will try to do next.

The powers that speciﬁc types of bugs have can be

decided by the user. All powers can also change thru

learning or evolution, decided by the user.

Bugs can use the resources at the node where they

are. When a bug performs an action, it consumes

some resource. In most of the simulations that we

have tried, we have only had one resource which we

called energy, but the system has no such restriction,

and we have run simulations with multiple resources.

They also have internal resources, which they can

obtain by eating. They can pick up and carry re-

sources as well. Presently they generally only eat

energy, although in some of the simulations they eat

other bugs. If their internal energy resources falls to

0, then they starve to death.

3.3 Evolution

Evolution can occur in many different ways in the sys-

tem. The bugs do not have genes per se. The set of

powers that the bug has, together especially with the

structure of the brain which decides what the bug does

next, is the “genetic code”. The parameters included

in the powers must also be included. Mutations can

occur in any power, most importantly in the brain. It

is possible that a new power can be added, if this par-

ticular type of bug is allowed to add powers. How

the brain changes during mutation is determined by

the type of brain that the bug has. More importantly,

a bug can inherit different powers from different par-

ents, and thereby get an evolutionary advantage over

both parents.

ALongTermProposaltoSimulateConsciousnessinArtificialLife

391

The experimenter determines when mutations oc-

cur. The most obvious time of mutation is when the

bugs reproduce. Both sexual and asexual are possi-

ble. When bugs reproduce asexually, only mutation

occurs. When two different bugs reproduce sexually,

two different brains must be combined. How this is

done is determined by the type of brain that they have.

Sexual reproduction has not yet been tested.

The bugs also can mutate as they live. One of the

factors that can mutate is their mutation rate. As bugs

become adapted, their internal mutation rate become

much slower. This may appear to be an unusual way

to evolve, but consider jumping genes (McClintock,

1950). The human brain appears to have jumping

genes that change what genes are applied in new neu-

rons, even in adults.

We have considered Lamarckian evolution as well

as Darwinian evolution. If some useful sequence of

actions gets used a lot, then the bug should be able to

increase the chance of that sequence occurring or pos-

sibly make some of the actions less costly. We have

not yet implemented this idea directly, but it is feasi-

ble within the system. This is just a form of learning.

4 AN IMPLEMENTATION

A prototype of SOCIAL has been implemented using

Java. It includes a graphical user interface for build-

ing an experiment. The user can deﬁne the graph,

choose the type, number, and locations of the bugs,

and also of the type and amount of resources. The

user can then start the simulation and watch it play

out if the graph is displayable. They can also watch

a graphical display of data, such as number of bugs,

amount of resouces, average age of each type of bugs

among others.

The implementation presently allow only grid

graphs or graphs imported from a ﬁle to chosen as a

universe. There are twenty-six different actions that

a bug can perform; eight sensors that the bug can

have, most with multiple parameters; thirteen brains

to choose from, also with multiple paramters. Some

of these powers are general purpose, but others were

developed simply to see if the SOCIAL implementa-

tion could copy results in the literature.

4.1 Some Simple Brains

Many different brains have been programmed, but we

only mention those more general ones that might be

useful as parts of future brains. The ﬁrst brain that we

tried was the random chance brain; each action has

a probability of been chosen. The chances of each

action can change thru mutation. The second brain

was the sequential brain that has a sequence of actions

that get repeated in a loop. The sequence can change

by adding/ removing actions, interchanging actions,

or splitting the sequence. This is a relatively effective

brain compared with many others. These two brains

do not use information from the sensors.

Brains can also use facts about the world, to help

decide what to do. Facts are discovered using sensors.

The simplest one is a hierarchical ruleset (Brooks,

1986), where a rule is an if-then construct consist-

ing of a possible fact, which can be true or false, and

an action. More complex approaches include Affect

Logic (Ciompi and Baatz, 2008) where creatures fol-

low plans and the Fungus Eater (Toda, 1982), which

combines routines and urges to create a behaviour.

These three brains have been tested in (Sußenberger,

2013).

Any deterministic brain can be implemented as a

decision tree. Evolving good decision trees does not

seem to be easy, because decision trees with similar

or even the same outcomes, can be quite distant from

each other in a tree edit model. We do not know a

good way to implement a deterministic brain that is

comprehensive (can model any deterministic brain),

stable (does not change too much with an evolution-

ary change) and yet is compact (size is polynomial in

the number of possible facts).

4.2 Some Simulations

We have not yet done any experiments that are of sci-

entiﬁc interest. For the most part we have looked at

other simpulations in the literature, and tested that

SOCIAL could also perform them. We emulated a

simple simulation writen for Repast Simphony (North

et al., 2006), in which “zombies” chase “humans”.

There was no evolution.

We have modeled an ecosystem with three differ-

ent bugs, which we called grass, sheep and wolf. A

grass bug ate energy and could not move, but could

reproduce into a neighboring cell; a sheep could move

and eat only grass; a wolf could eat only sheep. The

bugs evolve somewhat, as bugs with poor brains died

quickly. If the parameters are chosen well, eventu-

ally the system can stablize and keep all three types

of bugs usually. Sheep worked well with sequential

brains; wolves with decision tree brains.

We have tested SOCIAL with some other test

setups, for example the StupidModel (Isaac, 2011),

which is a reference implementation for agent-based

modeling platforms. It supports all the necessary

setup, output and display features to complete that

model. The behaviour of the agents in the model is

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

392

implemented using Actions, and the action sequence

is implemented using a simple Brain. This demon-

strates one of the core features of the SOCIAL model:

these actions can be reused by any other experiment,

or the same experiment can be run with a different

brain. Other models, such as one based on the Prison-

ner’s Dilemma (Kim, 2010) game, have also been im-

plemented. This showed that cooperation can evolve.

4.3 Motivators

Bugs must be motivated to do things. The only such

system that we have yet implemented is something

akin to hunger. The bug knows, or has a sensor that

can detect how much energy that it has, and if the en-

ergy level falls to 0 then the bug dies. So one thing

that the bug has to do to survive is to eat if the energy

level gets low. Successful bugs become motivated to

ﬁnd and eat food, whatever their food is, or at least

they act like they are motivated. Similarly bugs that

reproduce successfully only reproduce if their energy

level is high. Thus a single stored variable gives suc-

cessful bugs something like hunger and a desire to re-

produce under the appropriate conditions.

The action of the energy level is somewhat anal-

ogous to the level blood sugar in animals. The sim-

ulation is much simpliﬁed, but other motivators for

animals can often be driven by the levels of chemicals

in the blood or in the brain. We have not tested this

yet, but we can make general motivators just by giving

the bugs more variables that rise and fall on the basis

of their actions and/or their sensors. Unlike (Grand,

2000), who had a simulated blood stream in his crea-

tures, and simulated chemical levels, we do not intend

to specify what the bugs do with their motivators. We

intend to let evolution decide.

Whether motivators are part of the brain or an en-

tirely different power of the bug is not clear. Every

motivator needs sensors. They also need something

to change the level of the sensors, which could be a

brain of its own.

5 THE FUTURE

We have a great deal of work to do. Here are a few of

our ideas, some much more difﬁcult than others.

1. We need to get SOCIAL to work across multi-

ple machines. There will be security problems to

solve here.

2. In the short term, although the Prisoner’s dilemma

model has shown that the bugs can cooperate in at

least one case, we need more examples in which

the bugs to cooperate and communicate. (Wagner

et al., 2003) provides a comprehensive overview

of recent research in the emergence of communi-

cation in artiﬁcial simulations.

3. Add motivators to the system.

4. Give the bugs memory. (How?)

5. We would like to see a brain based on the Cerebral

Code deﬁned in (Calvin, 1996a; Calvin, 1996b),

or some other darwinian machine.

6. Design games to attract users.

7. Make SOCIAL an open source project, to which

others can contribute.

8. Show the universe from a bug’s viewpoint, as well

as examine every aspect of a bug. This could be

rather interesting in some cases. For example,

a bug wandering in a four-dimensional universe

could be rather interesting.

9. Investigate multicellular creatures, in which mul-

tiple bugs stick together.

10. Have an interface to create new powers for bugs,

maybe a scripting language. Currently powers

have parameters can that can be set, but new pow-

ers need to be programmed.

6 CONCLUSIONS

We are proposing a system to evolve consciousness.

The graph universe is very simple, yet at the same

time will require the bugs to be very ﬂexible if they

move to different graphs.

The bugs themselves are very ﬂexible. A power

written for any bug up to now can be given to any

other bug. Eventually there may be powers that con-

ﬂict. Certainly it is not easy to combine two different

brain types, but otherwise powers do not seem to con-

ﬂict.

The internet is an almost ideal medium for evo-

lution. Evolution occurs best on small island envi-

ronments, but also requires there to be mixing be-

tween them. Individual computers act like small is-

lands. The occasional transfer of individualbugs from

one machine to another can allow the powers and

the brains to combine in novel ways, with possible

serendipity.

REFERENCES

Azevedo, F. A. C. and et al (2009). Equal numbers of neu-

ronal and non-neuronal cells make the human brain

ALongTermProposaltoSimulateConsciousnessinArtificialLife

393

an isometrically scaled up primate brain. Journal of

Comparative Neurology, 513:532–541.

Brooks, R. (1986). A robust layered control system for a

mobile robot. IEEE Journal of Robotics and Automa-

tion, RA-2:14–23.

Calvin, W. H. (1996a). The Cerebral Code. MIT Press,

Cambridge,Mass.

Calvin, W. H. (1996b). How Brains Think. BasicBooks,

New York.

Chalmers, D. J. (2010). The character of consciousness.

Oxford University Press, Oxford.

Ciompi, L. and Baatz, M. (2008). The energetic dimension

of emotions: An evolution-based computer simulation

with general implications. Biological Theory, 3:42–

50.

Deacon, T. W. (2012). Incomplete Nature: How Mind

Emerged from Matter. Norton, New York.

Dennett, D. (1991). Consciousness Explained. Backbay

Books, New York.

Gazzaniga, M. S. (2011). Who’s in Charge? Free Will and

the Science of the Brain. HarperCollins, New York.

Grand, S. (2000). Creation: Life and How to Make It. Har-

vard University Press, Cambridge Mass.

Grim, P., editor (2009). Mind and Consciousness: 5 Ques-

tions. Automatic Press.

Horton, J. D. (2010). An idea for a project: A universe

for the evolution of consciouness. Technical Re-

port TR10-203, Computer Science University of New

Brunswick.

Isaac, A. G. (2011). The abm template models: A refor-

mulation with reference implementations. Journal of

Artiﬁcial Societies and Social Simulation, 14(2):5.

Kim, J.-W. (2010). A tag-based evolutionary prisoner’s

dilemma game on networks with different topologies.

Journal of Artiﬁcial Societies and Social Simulation,

13(3):2.

McClintock, B. (1950). The Origin and Behavior of Mu-

table Loci in Maize. Proceedings of the National

Academy of Science, 36:344–355.

North, M. J., Howe, T., Collier, N., and Vos, J. (2006). A

declarative model assembly infrastructure for veriﬁca-

tion and validation. In S. Takahashi, D. S. and Rouch-

ier, J., editors, Advancing Social Simulation: The First

World Congress. Springer.

Pinker, S. (1997,2009). How the Mind Works. Norton, New

York.

Searle, J. R. (1997). The Mystery of Conscousness. The

New York Review of Books, New York.

Sußenberger, E. (2013). Tracing motivation in virtual

agents. Master’s thesis, Bonn-Rhine-Sieg Univer-

sity of Applied Sciences and University of New

Brunswick.

Sutherland, N. S. (1989). The international dictionary of

psychology / Stuart Sutherland. Continuum, New

York.

Toda, M. (1982). Man, Robot and Society. Nijhoff, The

Hague.

Wagner, K., Reggia, J. A., Uriagereka, J., and Wilkinson,

G. S. (2003). Progress in the simulation of emer-

gent communication and language. Adaptive Behav-

ior, 11:37–69.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

394