examples of epigenetic robots (e.g., see Table 2) as
demonstrating emergence, but not ongoing emergence.
4. Human Infant Developmental
Examples
In contrast to the state of the art in epigenetic robotics,
human infants clearly exhibit ongoing emergence.
Development is an unending process that continually
produces new skills by making use of currently available
skills, environmental conditions, and other resources. In
this section, we provide prototypical examples of ongoing
emergence in infants from three developmental areas:
walking, language, and visual object skills.
4.1 Emergence of Walking
As any one-year-old would acknowledge, walking is
more difficult than it may first appear. To properly walk,
children must achieve the right mix of balance, head
control, and coordinated oscillation of the limbs that have
thousands of muscle fibers and billions of nerves as well
as their own length, mass and transitory inertia. The
degrees of freedom for the task are enormous (Bernstein,
1967).
Further complicating this process, children grow
physically. They begin life top-heavy, which makes
stabilizing this system all the more difficult. Their growth
is also erratic and dramatic—children can go up to 63
days with no measurable change in height and then
suddenly grow up to 2.5cm in a single night (Lampl,
Veldhuis, & Johnson, 1992). As if this wasn’t difficult
enough, children must learn to navigate different slopes
and uneven terrain, perhaps while carrying objects
(Adolph & Avolio, 2000). Yet somehow nearly all
children learn to walk, and continue to walk, despite the
complexity of achieving the right mix of skills, changes
in body morphology, and varying situations.
Current theory (Thelen, 1995) views this process as a
dynamic self-organizing system in which integration of
diverse skills plays a key role. Because the world, the
task, and even children’s bodies are constantly changing,
each component is constantly being weighted differently,
as dictated by the interaction of the nervous system and
the environment. For example, while all infants posses a
stepping and a kicking reflex at birth, the stepping reflex
disappears after a few months. Why? In short, babies
don’t have the strength to keep up this reflex as they
grow heavier—even though the nervous system is still
sending the signals. Stepping and, by extension, walking
must wait for stronger muscles to grow before infants can
take their first steps. If one makes the task easier, by
supporting the infants (on a treadmill or underwater),
even newborns can walk (Thelen & Fisher, 1982). In
contrast, if one makes the task harder by placing weights
on older infants, their walking again approximates that of
younger infants (Thelen & Fisher, 1982). Thus, it is the
dynamic interaction between current skills, the state of
the system, and the environment that allows for walking
behavior to self-organize into coherent patterns across
changes in morphology and task. This illustrates Criterion
1 for ongoing emergence—namely that skills are created
through the integration of resources including
environment and existing skills.
4.2 Emergence of Language
While purely physical skills like walking show ongoing
emergence, skills that are more cognitive also require the
use and integration of multiple developing skills. For
example, word learning can be seen as the product of
social skills (e.g., sensitivity to eye gaze), domain-general
skills (e.g., sensitivity to statistical structure such as
invariances), and linguistic skills (e.g., a bias toward
labeling objects based on shape).
Just as in walking, the weight placed on each of these
skills likely changes with time and situation. In the
beginning, infants may depend on a range of perceptual
biases and statistical relations to establish the meaning of
each new word (Hollich, Hirsh-Pasek, & Golinkoff,
2000). However, as more words are learned, children use
their knowledge of known words to help them learn
additional words. This illustrates another property
(Criterion 2) of systems exhibiting ongoing emergence:
incorporation of new skills into a skill repertoire. For
example, Smith (1999) provides evidence that infants
may notice how particular types of words get extended
(e.g., nouns are generalized, a.k.a. extended, to different
objects on the basis of shape). Infants then use this
knowledge to guide their own extensions of novel words.
When told a U-shaped object is a “dax,” infants will
spontaneously extend that word to other U-shaped
objects. Even so the system is flexible—infants will not
extend a word based on shape if the object happens to
have eyes (or even shoes), suggesting that children have
noticed that living creatures can often change their
overall shape in ways that static objects do not.
Related to the use of multiple skills, the more skills
that an agent can bring to bear, the more fault-resistant
and flexible the system. Loss of one skill does not cause
the system to fail entirely, and the interaction among
skills insures that children can successfully acquire a
language under extremely impoverished conditions. For
example, even deaf children growing up in an area
without exposure to any fully formed language will create
their own language (Sengas, 1995). With both biological
and robotic systems, more pathways to success imply
greater adaptivity and increased likelihood of organism
survival. With regular upheaval at the neurological and
muscular levels, it is no wonder that developmental
architectures are massively fault tolerant with multiple,
redundant skills. Thus, the self-organization of new skills
combined with an increasing skill repertoire can lead to a
process of ongoing emergence.
4.3. Object Skill Developments
At the same time that human infants are developing the
walking and language acquisition skills discussed above,
they also show an ongoing emergence of physical and
mental capabilities related to visual objects. Starting from
birth, infants are able to extract information about object
66