Memetic Engineering as a Basis for Learning in Robotic CommunitiesThis paper represents a new contribution to the growing literature on memes. While most memetic thought has been focused on its implications on humans, this paper speculates on the role that memetics can have on robotic communities. Though speculative, the concepts are based on proven advanced multi agent technology work done at NASA - Goddard Space Flight Center and Lockheed Martin. The paper is composed of the following sections : 1) An introductory section which gently leads the reader into the realm of memes. 2) A section on memetic engineering which addresses some of the central issues with robotic learning via memes. 3) A section on related work which very concisely identifies three other areas of memetic applications, i.e., news, psychology, and the study of human behaviors. 4) A section which discusses the proposed approach for realizing memetic behaviors in robots and robotic communities. 5) A section which presents an exploration scenario for a community of robots working on Mars. 6) A final section which discusses future research which will be required to realize a comprehensive science of robotic memetics.
https://ntrs.nasa.gov/citations/20160007482https://ntrs.nasa.gov/api/citations/20160007482/downloads/20160007482.pdfMemetic Engineering
For robots, memes have been defined as sets of instructions
that can be followed to evolve behavior. Instructions can be
encoded as written text and visible or vocal action (Silby,
2000). To allow for memetic learning, memes should also
include observations of the environment. A robot that is able
to observe and intelligently imitate the behavior of others is
able to participate in memetic learning. In order to perform
intelligent imitation, a robot needs to be able to process
memetic information. This process involves evaluating
models, examples, and patterns which the robot observes
(Hougen et al., 2003). In addition, the robot is expected to
analytically compare its current knowledge to the new
information it is observing.
A robot has modified its individual knowledge base when
it learns a new meme. Each robot is expected to evaluate
active memes in the community knowledge base for
strengths and weaknesses when deciding whether to learn
them. It can be expected that each individual robot will
benefit from the aggregation of other robots which are also
participating in memetic learning. When the community
knowledge base and size expands, there is a larger selection
of memes which can be evaluated and learned. With a larger
community knowledge base, a robot has a larger selection
of memes to modify to develop novel memes. All
individuals capable of participating in memetic learning are
able to generate new memes. Also, individuals who are able
to broadcast observations are capable of generating new
memes.
The memes that are in the knowledge base of a robot are
in constant competition with all other memes in the meme
pool (Silby, 2000). The meme pool is the collection of all
existing memes that are accessible to the other individuals
in the community. An individual robot may develop new
memes that become candidates for imitation in the
community meme pool. The community meme pool
increases with the addition of novel memes generated by
individual robots in the community. The connection
between an individual knowledge base and the community
meme pool is similar to the structure of a distributed cloud
network. A distributed cloud network is structured so that
each individual is connected to the cloud where they can have limited abilities to explore large areas efficiently.
NASA has shown increased interest in robotic teams for
exploration and servicing. In these teams different robots
would have various roles and responsibilities. To work
together in a team, robots will need to exchange information
on their current status, needs, capabilities and findings.
Using memes as an unit of transmission could facilitate this
needed information exchange.
Memetics has been a recent subject of interest as a new
method for information exchange (Silby 2000, Hougen et
al., 2003, Aunger 2002, Blackmore 2000, Wilson et al.
2011) . In communities, memes have been studied to
understand and enhance group learning (Hougen et al.,
2003). Richard Dawkins first defined memes as a unit of
cultural transmission (Dawkins, 1989). Essentially, memes
are ideas that evolve according to the same principles as
biological evolution (Silby, 2000). Memetic learning works
by transmitting units of cultural ideas or symbols from one
mind to another. All ideas that exist within an individual's
mind are examples of memes. Memes that are good at
replicating leave more copies of themselves in minds.
Examples of memes are catch phrases, musical themes,
scientific ideas and sayings. In robotics, examples of memes
are algorithms, observations, and instructions.
Figure 1 - Visual Representation of Memes
Memes benefit from having behavior similar to genes
(Gunders, 2010). Like genes, memes are under constant
selection pressures. Memes are in constant competition to
be absorbed and evolved from the collection of memes in
the community. Memes are competing to be learned and
those that are better at reproducing are those that are
successful in accomplishing their intended behavior. Memes
give communities more power and knowledge to
accomplish their tasks. There are still many unknowns
regarding the representation and transmission of memes in
a robotic community.
Memetic Engineering
For robots, memes have been defined as sets of instructions
that can be followed to evolve behavior. Instructions can be
encoded as written text and visible or vocal action (Silby,
2000). To allow for memetic learning, memes should also
include observations of the environment. A robot that is able
to observe and intelligently imitate the behavior of others is
able to participate in memetic learning. In order to perform
intelligent imitation, a robot needs to be able to process
memetic information. This process involves evaluating
models, examples, and patterns which the robot observes
(Hougen et al., 2003). In addition, the robot is expected to
analytically compare its current knowledge to the new
information it is observing.
A robot has modified its individual knowledge base when
it learns a new meme. Each robot is expected to evaluate
active memes in the community knowledge base for
strengths and weaknesses when deciding whether to learn
them. It can be expected that each individual robot will
benefit from the aggregation of other robots which are also
participating in memetic learning. When the community
knowledge base and size expands, there is a larger selection
of memes which can be evaluated and learned. With a larger
community knowledge base, a robot has a larger selection
of memes to modify to develop novel memes. All
individuals capable of participating in memetic learning are
able to generate new memes. Also, individuals who are able
to broadcast observations are capable of generating new
memes.
The memes that are in the knowledge base of a robot are
in constant competition with all other memes in the meme
pool (Silby, 2000). The meme pool is the collection of all
existing memes that are accessible to the other individuals
in the community. An individual robot may develop new
memes that become candidates for imitation in the
community meme pool. The community meme pool
increases with the addition of novel memes generated by
individual robots in the community. The connection
between an individual knowledge base and the community
meme pool is similar to the structure of a distributed cloud
network. A distributed cloud network is structured so that
each individual is connected to the cloud where they can
access the knowledge bases of others in the community
pool. Individuals will be able to quickly access, process, and
analyze the collective knowledge within the cloud (Krutz et
al., 2010).
Figure 2 - Visualization of Memetic Knowledge Base
Related Work
Memes can be utilized for many other applications other
than robotics. One example has been tracking memes to
develop a coherent representation of the news cycle - the
daily rhythms in the news media that have previously been
difficult to perform quantitative analysis (Leskovec,
Backstrom, & Kleinberg, 2008). By using memes, Leskovec
et al were able to develop a framework for tracking short
and distinctive phrases that travel relatively intact in media.
Further, the change in information as it propagates was able
to be observed. Using memetics in this model has provided
a method to see how a particular idea moves within and
between groups.
Memetics also has the potential to enhance our study of
psychology. Memes can be used to discover the origin of
certain psychological conditions. By modeling the mind as
a memetic construct, conditions such as depression or
addiction might be explained by memetic viruses that
influence the behavior of an individual (Silby, 2000). This
method may provide a path to pinpoint methods to prevent
the spread of malicious psychological conditions.
The most powerful adaptation of memetic systems is the
ability to predict behavior and evolution of future memetic
structures. In the future, psychologists may be able to model
memetic learning within communities and predict what will
happen when individuals are exposed to combinations of
memes. With this knowledge, memes will become a driving
force in the study of human behavior (Silby, 2000).
