intrin infants

Investigating the Origins of Intrinsic Motivationin Human Infants

Matthew Schlesinger

Department of PsychologySouthern Illinois University Carbondale

[email protected]

Abstract. One of the earliest forms of intrinsic motivation is visualexploration. In this chapter, I highlight how the development of this ca-pacity is influenced not only by changes in the brain that take placeafter birth, but also by the acquisition of oculomotor skill. To providea context for interpreting these developmental changes, I then surveythree theoretical perspectives that are available for explaining how andwhy visual exploration develops. Next, I describe work on the develop-ment of perceptual completion, which offers a case study on the devel-opment of visual exploration and the role of oculomotor skill. I concludeby discussing a number of challenges and open questions that that aresuggested by this work.

1 Introduction

In contrast to many other species, including gorillas, chimpanzees, and monkeys,human infants are born comparatively helpless and physically immature (Antin-ucci, 1989; Bjorklund, 1997). The fact that humans are unable to survive afterbirth without assistance from others has a number of important consequences.For example, humans have evolved a complex set of caretaking behaviors thatpromote attachment and emotional bonding between parents and offspring(Bowlby, 1969). Another important consequence is that human infants spendthe first several months of life relatively immobile: reaching and grasping do notbegin to emerge until age 3 months, while crawling develops approximately 5months later. As a result, there is a strong developmental bias in young infantstoward visual exploration of the physical world.

The goal of the current chapter is to highlight the development of visualexploration in human infants, not only as a case study in intrinsic motivation,but also as part of a more general claim that visual exploration helps to establishmany of the essential components of intrinsic motivation (e.g., curiosity, novelty,surprise, etc.). Accordingly, the rest of the chapter is organized into foursections. In the first section, I briefly survey the core developmental changesthat occur in visual perception during the first year of infancy, including theshift from exogenous to endogenous control of visual attention. The secondsection reviews some of the major concepts and theoretical approaches that

2 Matthew Schlesinger

are available for linking the development of visual exploration to the studyof intrinsic motivation in infants (e.g., assimilation and accommodation). Inthe third section, I focus on how visual selective attention-a critical skill thatunderlies visual exploration-begins to emerge and develop in early infancy. Inparticular, this section presents recent empirical work that utilizes behavioraland computational methods as an integrated approach for studying the de-velopment of visual exploration in young infants. The fourth and final sectionraises a number of fundamental challenges and open questions concerning thedevelopment of intrinsic motivation.

2 Developmental Components of Visual Exploration

Parents often note that when their newborn infant is in a quiet, alert state, theyare remarkably ”focused” or ”attentive.” The infant may become engrossedwith a mobile hanging above the crib, the mother’s face, or perhaps their ownhand as it moves through the visual field. In this context, it is tempting towonder: what does the infant see? How clear is their vision, and how do theyinterpret or make sense of their visual experience? In addition, what is drivingor motivating this process? Are infants systematically studying and absorbingthe features of their visual world, or is this behavior simply an instance of”tuning out” with their eyes open?

The question of what engages young infants’ attention, and what influencestheir visual exploratory activity can be framed as a developmental task. Muchlike sucking, grasping, crawling, and other basic motor behaviors, learning tosee is a sensorimotor skill that infants develop in the first year of life. From thisperspective, the sequence of changes that occur in visual perception betweenbirth and 12 months is in many ways a process of acquiring increasingly complexand hierarchically-organized or integrated behaviors. These behaviors not onlyimprove infants’ ability for information pick-up (e.g., efficient visual scanning),but they also provide essential data for the process of conceptual development(e.g., object recognition, object permanence, perception of causality, etc.).

To help illustrate the kinds of changes that occur in visual explorationduring early infancy, Figure 1 presents two samples of gaze behavior in younginfants from a study of visual encoding in 2- and 12-week-olds (Bronson, 1991).The sample on the left (Figure 1A) highlights two important aspects of efficientscanning behavior that are typical of 12-week-olds: (1) individual fixations(the small dots) are relatively well-distributed, spanning most of the stimulus,and (2) fixation times–which are indicated by the size of the dots–are fairlybrief (i.e., less than 1 second each). In contrast, 2-week-old infants tend toproduce scanning patterns like those on the right (Figure 1B), which includecomparatively long fixations (i.e., 2-3 seconds) that only span a small portion

Investigating the Origins of Intrinsic Motivation in Human Infants 3

Fig. 1. Gaze patterns produced by two infants (Bronson, 1991) The infant on theleft (A) has many brief fixations, distributed over the stimulus while the infanton the right (B) has several long fixations focused on the lower right portion ofthe stimulus.

of the stimulus.

These data indicate that a fundamental shift in oculomotor control occursduring early infancy. In this section, I highlight changes at two levels that arepart of this shift, and which play a critical role in the development of visualexploration during the first year: (1) maturation of the neural substrate, and (2)the transition from exogenous to endogenous control of attention.

2.1 Maturation of the Neural Substrate

The neural pathway from the retina to the visual cortex is not fully functionalat birth, but instead undergoes a number of important changes during the firstfew months of post-natal life (Banks & Bennett, 1988; Johnson, 1990). Whilethe development of this pathway is largely constrained by maturational factors,visual input also plays a significant role. Greenough and Black (1999) describethe process as experience-expectant development. In particular, a basic neuralpathway is biologically ”hardwired” and then as a series of typical experiencesoccur (e.g., visual or auditory input), structures and connections within thepathway become more specifically shaped or tuned. In the case of early visualprocessing, synapses between neurons are initially ”overproduced” and thenselectively pruned as a function of input into the visual system (e.g., stimulationof the left and right eyes).

One of the most important examples of growth in the early visual systeminvolves changes in the structure of the fovea. In this case, it is also an example


that has direct impact on the development of visual exploration, because thechanges that occur are associated with not only an increase in visual acuity(i.e., detail vision), but also the emergence of color perception. As Banks andBennett (1998) highlight, there are several anatomical differences betweenfoveal cones in the newborn infant and those in adults. For example, theouter segments of these cones (which are responsible for ”capturing” photonsand channeling them toward the inner segments), are significantly larger innewborn infants. As a result, neonate visual acuity is highly nearsighted (e.g.,approximately 20/400). Over time, the outer segments gradually shrink, whichincreases the packing density of cones and leads to an improvement in visualacuity. In addition, Banks and Bennett (1999) note that color perception alsoimproves as the likelihood of capturing photons increases.

While the previous example emphasizes anatomical changes at the sensorylevel, there are also several changes taking place within the cortex that supportthe development of visual exploration. For instance, Canfield and Kirkham(2001) highlight maturation of the prefrontal cortex (PFC), and in particular,the frontal eye fields (FEF). In adults, FEF activity is associated with anticipa-tory eye movements, which typically occur prior to (or during the occlusion of)a visual target and suggest that the movement is not only voluntary, but alsoplanned or programmed on the basis of pre-target information. When does thisarea begin to develop in infants? The evidence available to answer this questionis somewhat mixed, and varies depending on the paradigm used to measuresaccade planning and voluntary eye movements. However, as I note in thenext section (Exogenous to Endogenous Control), a conservative estimate forevidence of FEF activity in infants is age 6 months (e.g., Johnson, 1990), whilea more liberal estimate is 3-4 months (e.g., Haith, Wentworth, & Canfield, 1993).

A third neural substrate that plays an essential role in visual explorationis the development of visual cortex, and in particular, the growth of horizontalconnections in area V1. Lesion and single-cell studies with non-human primatesprovide converging evidence that these connections support the perception ofcontours (e.g., Albright & Stoner, 2002; Hess & Field, 1999). The ability to per-ceive contours is a critical prerequisite for visual exploration, in at least twoways. First, as Figure 1 highlights, efficient and systematic scanning of a visualstimulus cannot occur unless fixations are distributed over the entire stimulus.Without contour perception, the stimulus may appear as a disorganized collec-tion of features (e.g., patches of high and low contrast), and the observer may beunable to determine whether a particular region has already been fixated or not.Second, contour perception also enables the observer to integrate surfaces of anobject that are separated due to an occluding surface (e.g., an animal standingbehind the trunk of a tree). In effect, horizontal connections between V1 neu-rons create a perceptual ”fill-in” mechanism, in which objects are perceived ascoherent wholes despite being partially occluded (e.g., Albright & Stoner, 2002).


2.2 Exogenous to Endogenous Control

In parallel with underlying changes in neural structure and function, oculomotoractivity also develops systematically during early infancy. There are numerousexamples that provide support for a broad, consistent developmental trend: (1)between 0 to 2 months, infants’ visual activity is largely stimulus-driven, andcontrolled by exogenous factors such as luminance and motion, and (2) afterage 2 months, endogenous control of vision begins to emerge, enabling infantsto deploy their visual attention in a deliberate or goal-directed manner (e.g.,Haith, 1980). I briefly describe here a few important examples of this shift fromexogenous to endogenous control of vision.

First, a key hallmark of endogenous control of vision is that it is predictive oranticipatory. For example, the visual expectation paradigm (VEP) investigatesthe development of predictive eye movements by presenting infants withsequences of images that appear at specific locations (e.g., Haith, Hazan, &Goodman, 1988). The ability to learn simple sequences, such as images thatalternate consecutively at two locations (e.g., A-B-A-B), does not appear todevelop until age 2 months, while predictive scanning of more complex sequences(e.g., A-A-B-A-A-B) begins to develop by age 3 months (Haith, Wentworth, &Cantfield, 1993; Wentworth & Haith, 1998).

As Figure 1A illustrates, a second important feature of endogenous visionis that it is systematic or exhaustive. For example, Maurer and Salapatek(1976) recorded the gaze patterns of 1- and 2-month-old infants as they viewedpictures of faces. The younger infants tended to spend less time looking atthe faces than the older infants, and their fixations were often limited to theouter edges of the faces. In contrast, the older infants distributed their fixationsmore evenly over the pictures, and in particular, they spent significantlymore time viewing the internal features of the faces (e.g., eyes, nose, andmouth). It is important to note not only that systematic scanning behavior isa skill that continues to develop beyond age 2 months, but also that there isvariability in the rate at which individual infants develop. I return to this issuelater in the chapter, where I discuss the development of visual selective attention.

Finally, a third component of endogenous control of vision that develops inearly infancy involves competition between multiple, salient objects or locationsin the visual field, and the ability to adaptively distribute attention to theselocations. Dannemiller (2000) investigated the development of this skill betweenages 2 and 5 months, by presenting infants with a moving bar that was embeddedwithin a field of stationary bars of different colors. Two major findings emergedfrom this work. First, Dannemiller found that detection of the moving targetsignificantly improved with age. Thus infants, older infants were more sensitiveto the motion of the target object, and less distracted by the stationary bars.Second, Dannemiller also found that the spatial distribution of the colored, sta-tionary bars had a stronger influence on the older infants. In particular, older


infants benefited when relatively salient, stationary bars were positioned on thesame side as the moving target, while the performance of younger infants wasrelatively unaffected by the placement of the salient, stationary bars.

3 Linking Visual Exploration and Intrinsic Motivation: ATheoretical Bridge

The review presented thus far highlights both the biological and behavioralchanges that underlie the development of visual exploration. Neural growthand physiological development help make possible more efficient and adaptivemodes of information pick-up, and at the same time, visual scanning strategiesbecome progressively more complex, endogenously controlled, and sensitive tovisual features in the environment. While these changes provide infants with themeans to explore the world in increasingly powerful ways, however, they remainsomewhat descriptive. Thus, although they are necessary components or toolsof visual exploration, they do not provide direct answers to two fundamentalquestions. First, why do infants explore the visual world? In other words, isthe exploratory process intrinsically motivated, and if so, what is the basisfor this motivation? Second, how does the intrinsic motivation (IM) for visualexploration arise, and how does it develop?

In this section, I survey some of the prevailing answers that developmentalresearchers have proposed to address these questions. In particular, I describethree distinct theoretical approaches: (1) evolutionary theory, (2) Piagetian the-ory, and (3) information-processing theory.

3.1 Evolutionary Theory

According to evolutionary theory, newborn infants’ gaze behavior is a productof mammalian and primate evolution. By this view, infants are born withan innate behavioral ”program” that directs visual attention. This programhas the implicit goal of providing optimal visual stimulation, which promotesthe growth and development of the neural pathways that support vision. Forexample, Haith (1980) proposes the high firing rate principle, which includesa set of basic heuristics that guide visual activity, such as (1) when thereis no visual input, move the eyes, and (2) if an edge (i.e., adjacent regionsof high and low luminance) is encountered, move the eyes over the edge. Asimilar approach is proposed by Sporns & Edelman (1993), who emphasize theconcept of biological or intrinsic value systems. According to this view, valuesystems are species-specific neural circuits that evolve through the process ofnatural selection, and become activated during sensorimotor activity. In thecase of visual exploration, for example, eye movements create stimulation onthe retina, which modulates neural activity throughout the visual stream (e.g.,lateral geniculate nucleus, occipital cortex, etc.). These patterns of sensorimotoractivity result in increased value (e.g., release of dopamine), which strengthens


the neural circuits that are active during visual exploration.

Evolutionary theory provides a clear answer to the question of why infantsexplore their environment visually: because this behavior promotes brain devel-opment. It also provides a preliminary answer to the question of how it develops:due to its adaptive significance, visual exploration was selected as an innate be-havior in humans through the process of evolution and natural selection. How-ever, where evolutionary theory falls short is in providing a detailed account ofhow visual exploration develops after birth, and in particular, how the value ormotivation system interacts with specific environmental experiences over timeto produce the developmental trajectory observed in human infants.

3.2 Piagetian Theory

A highly-influential approach to understanding the link between IM and visualexploration is offered by Piaget’s theory of cognitive development (1952).Two central concepts are assimilation and accommodation. Assimilation isthe tendency to use familiar or preexisting behaviors when interacting withan object. For example, after young infants master the ability to nurse,they will often examine new objects by exploring them with their mouth.Accommodation, meanwhile, is the modification of preexisting behaviors orschemes, or the invention of new schemes, which gradually become coordinatedand integrated with previous schemes. Imagine, for example, that an infantwho tends to explore objects with the mouth is given a piece of ice. Atfirst the ice goes in the mouth (assimilation), but it is quickly spit out. Theinfant subsequently discovers that the ice slides when pushed (accommoda-tion), and so a novel scheme begins to form as the infant plays with their new toy.

It is important to note that Piaget proposed several distinct forms ofassimilation. Functional assimilation, for example, is the tendency to ”exercise”an existing behavior or scheme. Piaget suggested that functional assimilationplays a fundamental role in the acquisition of new skills (i.e., as the scheme isbeginning to form). This view is consistent with the observation parents oftenmake while their infants develop a new capacity, such as learning to reach orcrawl, as the infant appears entirely focused on ”practicing” or deliberatelyrehearsing the ability over and over again. Generalizing assimilation, meanwhile,is the tendency to extend schemes that have been acquired with familiar objectstoward novel or unfamiliar ones.

Taken together, functional and generalizing assimilation provide a straight-forward answer to the questions of why and how visual exploration develops.First, functional assimilation means that visual exploration is the tendency toexercise the schemes associated with visual activity (e.g., fixations, saccades,pursuit of moving objects, etc.). Given an object to look at, an infant willautomatically engage their visual schemes. According to Piaget, the functioningof these schemes is analogous to the biological activity of physical organs, such


as the relation between food and the digestive system, or air and the respiratorysystem. While this view overlaps with the approach offered by evolutionarytheory, an important distinction is that Piaget’s account also predicts thetransition from exogenous to endogenous control of vision. In particular, Piagetsuggests that the ”automatic” or reflexive exercising of visual activity–that is,functional assimilation–should diminish over time, as the new skill becomesmastered.

Second, generalizing assimilation also clearly defines the role of IM in thedevelopment of visual exploration. Specifically, Piaget proposes the moderatenovelty principle as a core feature of visual exploration (1952; Ginsburg & Op-per, 1988). According to this principle, as infants generalize preexisting schemesto new objects, they do not select those objects arbitrarily. Instead, infants tendto orient toward new objects which are neither too easily assimilated, nor thosethat require entirely new schemes (i.e., completely familiar or unfamiliar objects,respectively). Instead, generalizing assimilation tends to favor objects that aremoderately novel, and therefore largely fit preexisting schemes, but may requiresome modest degree of accommodation (i.e., modification of ongoing activity) inorder to be perceived, used, or manipulated. By this view, infants’ oculomotoractivity patterns or schemes–where and how they explore the visual world–is in-timately tied to their preexisting perceptual categories. Thus, Piaget proposes adynamic and reciprocal connection between seeing and knowing: infants’ visual-exploratory skill level determines the objects they perceive and recognize, whiletheir current perceptual categories influence how they seek out new visual fea-tures. In other words, the process of exploratory skill development leads to thediscovery of new perceptual features and categories, and vice versa.

3.3 Information-Processing Theory

A third approach that relates IM and visual exploration is offered byinformation-processing theory, which both overlaps with and is complementaryto Piaget’s theory. However, there are two important distinctions. First,information-processing theory uses the modern computer as a metaphor forunderstanding mental processes, and consequently, tends to focus on a relativelynarrow set of phenomena, including perception, attention, and memory. Second,and more importantly, while Piaget relies on the somewhat abstract processof equilibration (i.e., the balance between assimilation and accommodation)as the primary mechanism for driving development, information-processingtheory instead emphasizes maturation of the underlying neural substrate (i.e.,”hardware”) and the acquisition of new and more efficient or powerful strategiesfor gathering and using information (i.e., ”software”).

One of the most direct ways that information-processing theory linksIM and visual exploration is through the concepts of curiosity, novelty, andsurprise. For example, template-matching theory (e.g,. Charlesworth, 1969;Sokolov, 1963) proposes that as infants view an object or event, they compare


the visual input produced by that experience with a corresponding internalrepresentation (e.g., mental image or memory). By this view, the amount ofinterest or curiosity an infant shows in an object–which is operationalized andmeasured by their looking time–is a function of how well-represented the objectis: completely familiar objects have robust internal representations, and so thecomparison process terminates quickly (i.e., the infant rapidly loses interest).In contrast, objects that are moderately novel have weak or incomplete internalrepresentations, and as a result, generate a longer comparison process as therepresentation is updated with new sensory data (i.e., mild curiosity or interest).

In many ways, this account is comparable to Piaget’s moderate noveltyprinciple. However, an interesting difference between the theories arises inthe case of a highly novel object. While Piaget might predict that a highlynovel object should generate a low level of interest, there is some evidence tosuggest that such objects may in fact be strongly aversive to young infants. Forexample, Fox, Kagan, and Weiskopf (1979) proposed that stranger wariness oranxiety–which emerges around age 6 months and is marked by distress, fear,and crying in the presence of unfamiliar adults–is the result of improvement inlong-term memory. By this view, the fear is not in response to perceived danger,but rather is a reaction to the inability to retrieve an image from memory thatcorresponds to the face of the stranger. In order to evaluate their hypothesis,Fox, Kagan, and Weiskopf (1979) conducted a longitudinal study of infantsbetween ages 5 and 14 months, and as predicted, found that improvements inmemory reliably preceded the development of stranger anxiety.

Finally, an interesting issue raised by information-processing theory is thequestion of whether individual differences in processing or encoding speed dur-ing infancy can accurately predict later intelligence. For example, Bronson (1991)noted considerable variability between infants at age 3 months in the strategiesused to scan objects. Information-processing theorists have proposed that infantswho scan visual stimuli rapidly and efficiently are thus able to create internalrepresentations more effectively than infants who scan slowly and unsystemati-cally (e.g., Bornstein & Sigman, 1986), and that this difference between infantsmay be diagnostic of more general differences in intelligence or intellectual abil-ity. Indeed, in a meta-analysis of 23 studies, McCall & Carriger (1993) foundsubstantial support for a predictive relation between visual processing duringinfancy and intelligence in later childhood. In effect, infants who are faster vi-sual processors grow up to be more intelligent children (as assessed by standardIQ tests). While these data do not provide a direct link between IM during in-fancy and intelligence in older children, they do suggest that early forms of IMsuch as visual exploration differ in quality and quantity from child to child, andthat these early differences may have an influence on intellectual growth later inchildhood.


4 Visual Selective Attention: A Case Study in IMDevelopment

As I noted earlier, one of the primary goals of the current chapter is to proposethat visual exploration is an early manifestation of IM that provides a founda-tion for several essential components. In order to support this claim, I presentin this section a detailed description of a research project that investigatesthe development of oculomotor skill and object perception in young infants(Amso & Johnson, 2006; Johnson, 2004, Johnson, Slemmer, & Amso, 2004;Schlesinger, Amso, & Johnson, 2007a; Schlesinger, Amso, & Johnson, 2007b).The project incorporates and integrates many of the themes and ideas presentedthus far in the chapter, including: (1) the role of an underlying neural substratethat constrains the development of visual attention, (2) systematic changes inoculomotor behavior that reflect the transition from exogenous to endogenouscontrol of vision, and (3) the dynamic interplay between visual salience andvisual novelty, which work together to shape visual exploration in real time.

The specific component of visual exploration that we are investigating isvisual selective attention, which is defined as the ability to deploy attention ina goal-directed manner while ignoring irrelevant stimuli. In particular, our workseeks to identify the specific mechanisms–both neural and behavioral–that makevisual selective attention possible. Our approach is designed to address a numberof questions. First, how do changes in visual selective attention during earlyinfancy give rise to new (and more effective or efficient) patterns of scanningbehavior? Second, as new scanning strategies emerge, what effect, if any, dothey have on the development of object perception? And finally, how does thedevelopment of visual selective attention help explain the emergence of IM inyoung infants?

Fig. 2. Illustration of the stimuli used to test the development of perceptualcompletion in young infants. (A) The occluded-rod display, (B) the solid-roddisplay, and (C) the broken-rod display.


As a specific case study, consider the developmental problem of learning toperceive partially-occluded objects as coherent wholes. To help illustrate theproblem, Figure 2A presents a snapshot from an animation event, in which agreen rod moves laterally behind a large, blue screen (the arrows indicate thedirection of motion). In the unity perception task, infants are first presentedwith this occluded-rod display until they habituate (i.e., their looking timedecreases by 50%). After habituating, infants then watch the displays illustratedin Figures 2B and 2C on alternating test trials: during the solid-rod display(2B), a single rod moves laterally, while two smaller, aligned rod segments movesynchronously during the broken-rod display (2C).

The tendency to look longer at one of the two test displays is assumed toreflect a novelty preference (e.g., Gilmore & Thomas, 2002), and provides abasis for inferring whether infants perceive the occluded rod as a single object,or two disjoint segments moving synchronously. Thus, infants who look longer atthe broken-rod test display are referred to as perceivers, because their lookingpattern suggests that they perceive the occluded rod as a single, coherentobject. Alternatively, infants who look longer at the solid-rod test display arereferred to as non-perceivers, because their looking pattern suggests insteadthat they perceive the occluded rod as two disconnected segments.

How does unity perception develop? Interestingly, between birth and age2 months, infants look longer at the solid-rod display (e.g., Slater, Johnson,Brown, & Badenoch, 1996), and therefore appear to lack unity perception. Byage 4 months, this preference switches, and infants begin to look longer at thebroken-rod display (Johnson, 2004). This developmental pattern suggests thatunity perception is not present at birth, but instead rapidly develops in the firstfew months of post-natal life.

An important question–suggested by Piagetian theory–is whether the devel-opment of unity perception depends on improvements in oculomotor skill. Forexample, do perceivers and non-perceivers scan the occluded-rod (habituation)display in a comparable manner, or are there measurable differences in howthey distribute their attention? Johnson, Slemmer, and Amso (2004) addressedthis question by tracking the eye movements of 3-month-olds as they viewed theoccluded-rod display. Each infant in the study then viewed the solid-rod andbroken-rod test displays, and was subsequently assigned to either the perceiveror non-perceiver group, depending on which of the two test displays the infantpreferred to view.

Figure 3 presents examples of scan plots from two infants in the study(each dot represents a single fixation). Figure 3A, which was produced by aninfant in the perceiver group, includes a large proportion of fixations towardthe rod segments, and comparatively fewer toward the occluding box. Incontrast, Figure 3B was produced by an infant in the non-perceiver group. Note


Fig. 3. Gaze patterns produced by two 3-month-old infants while viewing theoccluded-rod display. The infant on the left (A) focused on the movement of therod, while the infant on the right (B) distributed their attention to other partsof the display, including the occuding box and the background dots.

that this second scan plot illustrates the opposite pattern: many fixations aregenerated toward the occluding box, and relatively few are generated towardthe moving rod. This qualitative pattern was investigated in a follow-up studyby Amso and Johnson (2006), who divided the occluded-rod display into aseries of regions-of-interest (ROI). Figure 4 illustrates the 6 ROIs, includingthe top and bottom portions of the occluded rod (ROIs 1 and 2) and the fourquadrants of the background and occluding box (ROIs 3-6). They defined rodscans as gaze shifts within or between the rod segments, and vertical scansas gaze shifts from the upper to lower quadrants, or vice versa (e.g., from 6 to 3).

Fig. 4. The six regions of interest (ROIs) used by Amso and Johnson (2006) tomeasure which aspects of the occluded-rod display were fixated by 3-month-oldinfants.


Based on the findings from Johnson, Slemmer, and Amso (2004), Amso andJohnson (2006) predicted that 3-month-old perceivers would produce signifi-cantly more rod scans than 3-month-old non-perceivers. Since vertical scansdo not provide information about the occluded rod, they also predicted thatperceives and non-perceivers would not differ in the percent of vertical scansproduced. Figure 5A presents the proportion of rod scans and vertical scansproduced by the two groups of infants. As predicted, 3-month-old perceiversgenerated significantly more fixations to the rod segments than non-perceivers,while there was no significant difference in the percent of fixations betweenthe upper and lower halves of the occluded-rod display. Taken together, thefindings from these two studies not only show that perceivers are more effectiveat scanning the occluded-rod display, but they also provide support for theidea that effective scanning is a necessary prerequisite for unity perception.However, the data do not provide a direct explanation for why infants make thetransition from non-perceivers to perceivers between ages 2 and 4 months. Whatunderlying developmental mechanism drives this process? In particular, whatenables young infants to scan the occluded-rod display more systematically, sothat they subsequently detect and attend to the key features of the display?

Fig. 5. Proportion of rod scans and vertical scans produced during the occluded-rod display. (A) Behavioral data acquired from 3-month-old infants, and (B)simulation data generated by an eye-movement model.

In order to explore these questions, we used a computational model of earlyvisual processing, originally designed by Itti and Koch (2000, 2001), to simulatethe development of early visual processing in young infants (Schlesinger,Amso, & Johnson, 2007a; Schlesinger, Amso, & Johnson, 2007b). As Figure 6illustrates, the model represents visual processing through a series of four opticaltransformations: (1) each image of the animation display is first projected ontoa simulated retina (6A), (2) a set of four feature-detection systems extractintensity, color, motion, and oriented edges (6B), (3) individual feature maps


are created by summing together the separate maps in each feature channel(6C), and the feature maps are combined into a unified salience map (6D).

In order to produce a simulated sequence of eye movements, the model isthen presented with the entire animation display, one frame at a time. Thesalience map is updated as each frame of the animation is presented, androughly 5 times per second, a highly-active (i.e., perceptually-salient) locationon the salience map is selected as a target for fixation. Gaze shifts are thenproduced by moving the model’s fixation point from one salient location toanother.

A critical parameter in the model is used during the second processingstage (see Figure 6B). The value of this parameter, which varies from 0 to10, controls the duration of an iterative process in which features at differentlocations within the same channel or dimension (e.g., color) compete foractivation. To help demonstrate this process, Figure 7 presents an input image,followed by the corresponding salience map that is produced after running thespatial-competition algorithm for 0, 2, 5, 8, and 10 steps. As Figure 7 illustrates,low values of the spatial-competition parameter (e.g., 0 or 2) are associated withshallow salience maps and multiple activation peaks. In contrast, high values(e.g., 8 or 10) are associated with steeper maps and fewer, more pronouncedactivation peaks.

The spatial-competition parameter has direct relevance for understandinghow IM and visual selective attention are related during the developmentof visual exploration. In particular, this parameter plays a fundamental rolenot only at the computational level, but also at the level of neurophysiology.First, note that because parameter values near 0 produce salience maps withnumerous peaks, the resulting gaze patterns include frequent fixations toinformation-poor areas of the display, such as the center of the occluding boxand the background texture dots. Conversely, parameter values near 10 resultin fixations to highly-informative areas of the display, such as the rod segmentsand the edges of the occluding box. Therefore, at a computational level, changesin the value of the spatial-competition parameter are clearly related to adaptivechanges in the scanning behavior of the model.

Second, the computational process of spatial competition can also beinterpreted as an analog for neural processing in the posterior parietal cortex(PPC), an area of the brain that is associated with the encoding of visualsalience and the modulation of visual attention (e.g., Gottlieb, Kusunoki, &Goldberg, 1998; Shafritz, Gore, & Marois, 2002). In particular, the featuremaps produced during the second processing stage (Figure 6B) correspond tothe retinotopic maps found in PPC, which represent salient locations, objects,and prospective targets for action (e.g., an eye movement, reach, etc.). Giventhis correspondence, the spatial-competition parameter can be interpreted as


Fig. 6. Overview of the model used to simulate infants’ visual exploration. Pro-cessing in the model is divided into four stages: (A) The retinal image, (B)feature abstraction and spatial competition, (C) pooling of feature maps withineach feature channel, and (D) summation of the feature maps into a saliencemap.


Fig. 7. Illustration of the spatial-competition algorithm. The original input im-age is presented, followed by the corresponding salience map that is created after0, 2, 5, and then 10 iterations of the spatial-competition algorithm.


controlling the temporal duration of recurrent processing in the PPC: low valuescorrespond to rapid processing, with brief pauses between synchronous burstsor ”pulses” of activity, while high values correspond to slower processing andlonger pauses between neural pulses.

It then makes sense to ask: does the duration of recurrent parietal processingdevelop systematically in early infancy, thereby modulating the quality of visualexploration? In order to investigate this question, we systematically hand-tunedthe value of the spatial-competition parameter from 0 to 10, and then measuredthe proportion of rod scans and vertical scans produced by the model during theoccluded-rod display for each value of the parameter. As Figure 5B illustrates, akey finding from this simulation study was that we replicated the scan patternsproduced by non-perceivers when the parameter was set at 3 iterations, whilethe scan patterns produced by perceivers was reproduced when the parameterwas set at 4 iterations (Schlesinger, Amso, & Johnson, 2007a).

Thus, the model not only succeeds in capturing the real-time behavior of bothperceivers and non-perceivers, but it also suggests that maturational growth ordevelopment in the posterior parietal cortex may play a central role in the shiftfrom exogenous to endogenous control of vision during early infancy. An impor-tant limitation of the model, however, is that our results were generated with theuse of a hand-tuned model. A logical next step, then, is to determine whetherthe same developmental pattern will emerge when the spatial-competition pa-rameter is allowed to vary freely within an adaptive-learning paradigm (e.g.,reinforcement learning, supervised learning, etc.).

5 Summary and Open Questions

I began the chapter with two questions: what engages young infants’ visual-exploratory activity, and how does this activity develop in the first few monthsof life? In this section, I begin by summarizing the main ideas and researchfindings that I raised. I then conclude by highlighting three open questions thatserve as key challenges to the study of IM in young infants, and suggestingpotential strategies for addressing these questions.

First, the development of visual exploration is strongly influenced by thegrowth of the neural systems that support oculomotor control. For the first 2to 4 months, infants are learning to focus their eyes, to trace complex contoursand follow moving objects, to shift their gaze rapidly and accurately towardtargets of interest. These skills are made possible in part by neural growththat includes structural changes not only at the retinal level, but also withinand between cortical areas (e.g., PFC, V1, and PPC). Second, and in parallelwith changes in the underlying neural substrate, infants are also establishingcontrol over their eye movements at the behavioral level. In particular, theirgaze patterns are becoming predictive or anticipatory, and they are also more


efficient at scanning visual stimuli. These changes suggest that the mechanismresponsible for driving visual exploration in young infants undergoes a majorqualitative shift in early infancy. Thus, while newborn infants look at thingsin a somewhat reflexive or stimulus-driven manner, by age 3 months infantslook at things in a more purposeful or deliberate manner. In other words, thenewborn looks because it has to, while the older infant looks because it wants to.

Although it is clear that both physiological and behavioral factors play animportant role in the development of oculomotor skill, neither of these levelsprovides a complete answer to the question of what motivates visual exploration.In other words, what is the function or purpose of visual-exploratory activity?I approached this question from three theoretical perspectives. First, accordingto evolutionary theory, visual exploration is a behavior pattern that has evolvedin the human species to promote brain development. By this view, visualexploration is analogous to an instinct or innate behavior. Second, while Piagetagrees that visual activity is initially a reflex-like behavior, he suggests thatwith practice and repetition it becomes an elaborate sensorimotor skill. Inparticular, he proposes that infants are constantly searching for new objectsto assimilate, and that this search process is biased toward objects that aremoderately novel. Finally, information-processing theory complements Piaget’sapproach, but also extends it by suggesting specific internal mechanisms (e.g.,template-matching theory) through which novelty motivates visual exploration.

After reviewing the theoretical approaches to IM and visual exploration,I then provided a detailed overview of recent work on the development ofvisual selective attention in young infants. Three major findings were described.First, at age 3 months, some infants have developed the ability to perceive apartially-occluded object as a coherent whole (i.e., perceptual completion), whileothers have not yet reached this milestone. Second, systematic comparisons ofhow infants scan occluded objects suggest that perceptual completion is theresult of efficient, systematic visual exploration. Finally, a series of simulationstudies provide additional support for the role of skilled oculomotor behavior inthe development of perceptual completion, and more importantly, also suggestthat growth in an area of the brain associated with spatial processing and visualattention (i.e., PPC) may underlie changes in visual exploration during earlyinfancy.

The ideas and work discussed thus far provide an important foundation, notonly for the study of visual exploration, but also more broadly for the studyof IM. Nevertheless, there are a number of issues that remain unaddressed. Ihighlight here three key questions.

1. Is visual exploration in early infancy the earliest form of IM? There area number of similarities between how young infants explore the worldvisually, and how older infants, children, and adults explore through moreelaborate forms of action. Indeed, Piaget proposed that the same underlying


mechanism can be used to explain infants’ exploratory activities, both inearly infancy and then again at later stages. Thus, he noted that between12 and 18 months, infants engage in a pattern of behavior he called atertiary circular reaction: at this stage, infants systematically manipulateobjects, in order to discover novel properties or functions. For example, achild might discover that dropping an object from a high chair makes aninteresting sound, and then attempt to drop different objects, one afteranother, in order to figure out which ones make the loudest sounds. Other,later-developing behaviors might also be included within the spectrum ofIM, such as the emergence of pretend or symbolic play in early childhood,and the development of academic skill and mastery motivation in middlechildhood. The possibility that IM begins to take root in early infancyis an important question for developmental psychologists. If in fact thesedifferent forms of IM are related, then it should be possible to predict thequality of later forms on the basis of how prior ones emerge and develop.For example, recall that individual differences in visual-processing speedand efficiency during infancy are associated with comparable differences inIQ during later childhood. These findings provide support for the idea thatcuriosity and novelty-seeking behavior in infants have a measurable impacton learning and cognitive skill in older children.

2. Is IM a general attribute, or is it skill- and domain-specific? A second,related question concerns the knowledge structures that are formed in theprocess of exploratory behavior. One possibility is that IM is a generalattribute. According to this view, a particular child can be characterizedas preferring a certain level of novelty or familiarity. For example, aninfant may show a strong preference for familiar objects, situations, andactions. The idea that there is a single, unified drive or mechanism for IM,which spans a diverse range of knowledge domains, has been proposed bySchmidhuber (2009). Alternatively, IM may be highly domain-specific. Inthis case, an infant may show a preference for familiarity in one domain orskill context (e.g., when interacting with people), while showing a preferencefor novelty in other domains (e.g., food, toys, books, etc.).

3. Where is the ”best place” to explore? The final question is relevant not onlyto developmental psychologists, but also to researchers in machine learningas well. In particular, Piaget’s moderate novelty principle suggests a strategyfor guiding the exploratory process: exploration, and as a result, learning isbest achieved at the ”edges” of the state space. In other words–rather thanrandomly probing the environment, or alternatively, exhaustively sweepingthrough all possible states–Piaget’s theory advocates for a relatively conser-vative approach to exploration, in which new experiences are sought thatmodestly extend known regions of the environment (i.e., hierarchical, cumu-lative learning). Interestingly, it should be noted that this approach comple-ments Vygotsky’s notion of the zone of proximal development, which is the


space between what a child can do alone, and what they can do with theassistance of others.

References

Albright, T.D., & Stoner, G.R. (2002). Contextual influences on visualprocessing. Annual Review of Neuroscience, 25, 339-379.

Amso, D., & Johnson, S. P. (2006). Learning by selection: Visual search andobject perception in young infants. Developmental Psychology, 42, 1236-1245.

Antinucci, F. (1989). Cognitive structure and development in nonhumanprimates. New Jersey: Erlbaum.

Bjorklund, D.F. (1997). The role of immaturity in human development.Psychological Bulletin, 122, 153-169.

Banks, M.S., & Bennett, P.J. (1988). Optical and photoreceptor immaturi-ties limit the spatial and chromatic vision of human neonates. Journal of theOptical Society of America, 5, 2059-2079.

Bornstein, M.H., & Sigman, M.D. (1986). Continuity in mental developmentfrom infancy. Child Development, 57, 251-274. Bowlby, J. (1969). Attachment.New York: Basic Books, Inc.

Bronson, G.W. (1991). Infant differences in rate of visual encoding. ChildDevelopment, 62, 44-54.

Canfield, R.L., & Kirkham, N.Z. (2001). Infant cortical development andthe prospective control of saccadic eye movements. Infancy, 2, 197-211.

Charlesworth, W.R. (1969). The role of surprise in development. In D.Elkind & J. Flavell (Eds.), Studies in cognitive development: Essays in honor ofJean Piaget (pp. 257-314). Oxford, UK: Oxford University Press.

Dannemiller, J.L. (2000). Competition in early exogenous orienting between7 and 21 weeks. Journal of Experimental Child Psychology, 76, 253-274.

Fox, N., Kagan, J., & Weiskopf, S. (1979). The growth of memory duringinfancy. Genetic Psychology Monographs, 99, 91-30. Gilmore, R. O., & Thomas,H. (2002). Examining individual differences in infants’ habituation patternsusing objective quantitative techniques. Infant Behavior & Development, 25,399-412.

Ginsburg, H.P., & Opper, S. (1988). Piaget’s theory of intellectual develop-ment. New Jersey: Prentice Hall.


Gottlieb, J. P., Kusunoki, M., & Goldberg, M. E. (1998). The representationof visual salience in monkey parietal cortex. Nature, 391, 481-484.

Greenough, W.T., & Black, J.E. (1999). Experience, neural plasticity, andpsychological development. In N.A. Fox, L.A. Leavitt, & J.G. Warhol (Eds.),The role of early experience in infant development (pp. 29-40). New York:Johnson & Johnson Pediatric Institute.

Haith, M.M. (1980). Rules that babies look by: The organization of newbornvisual activity. New Jersey: Erlbaum.

Haith, M.M., Hazan, C, & Goodman, G.S. (1988). Expectations and antic-ipation of dynamic visual events by 3.5-month-old babies. Child Development,59, 467-479.

Haith, M.M., Wentworth, N., & Canfield, R.L. (1993). The formation ofexpectations in early infancy. Advances in Infancy Research, 8, 251-297.

Hess, R., & Field, D. (1999). Integration of contours: New insights. Trendsin Cognitive Sciences, 3, 480-486.

Itti, L., & Koch, C. (2000). A saliency-based search mechanism for overtand covert shifts of visual attention. Vision Research, 40, 1489-1506.

Itti, L., & Koch, C. (2001). Computational modelling of visual attention.Nature Reviews Neuroscience, 2, 194-203.

Johnson, M.H. (1990). Cortical maturation and the development of visualattention in early infancy. Journal of Cognitive Neuroscience, 2, 81-95.

Johnson, S.P. (2004). Development of perceptual completion in infancy.Psychological Science, 15, 769-775.

Johnson, S.P., Slemmer, J.A., & Amso, D. (2004). Where infants lookdetermines how they see: Eye movements and object perception performance in3-month-olds. Infancy, 6, 185-201.

McCall, R.B., & Carriger, M.S. (1993). A meta-analysis of infant habituationand recognition memory as predictors of later IQ. Child Development, 64, 57-79.

Piaget, J. (1952). The origins of intelligence in children. New York: Interna-tional Universities Press.

Schlesinger, M., Amso, D., & Johnson, S.P. (2007a). Simulating infants’ gazepatterns during the development of perceptual completion. In L. Berthouze,


C.G. Prince, M. Littman, H. Kozima, & C. Balkenius (Eds.), Proceedings of theSeventh International Workshop on Epigenetic Robotics: Modeling CognitiveDevelopment in Robotic Systems (pp. 157-164). Sweden: Lund UniversityCognitive Studies.

Schlesinger, M., Amso, D., & Johnson, S.P. (2007b). The neural basis forvisual selective attention in young infants: A computational account. AdaptiveBehavior, 15, 135-148.

Schmidhuber, J. (2009). Simple algorithmic theory of subjective beauty,novelty, surprise, interestingness, attention, curiosity, creativity, art, science,music, jokes. Journal of SICE, 48, 21-32.

Shafritz, K. M., Gore, J. C., & Marois, R. (2002). The role of the parietalcortex in visual feature binding. Proceedings of the National Academy ofSciences, USA, 99, 10917-10922.

Sokolov, E. (1963). Perception and the conditioned reflex. New York:Pergamon.

Sporns, O., & Edelman, G.M.. (1993). Solving Bernstein’s problem: Aproposal for the development of coordinated movement by selection. ChildDevelopment, 64, 960-981.

Wentworth, N., & Haith, M.M. (1998). Infants’ acquisition of spatiotemporalexpectations. Developmental Psychology, 34, 247-257.

intrin infants

Documents