Making order from chaos: the misguided frontal lobe
Richard Ivry
& Robert T. Knight
The authors are at the Department of Psychology and the Helen Wills Neuroscience Institute, 3210 Tolman Hall, University of California at Berkeley, Berkeley, California 94720, USA. ivry@socrates.berkeley.edu; rtknight@socrates.berkeley.edu
The brain continually attempts to extract patterns from environmental events. A new report suggests that this process depends on prefrontal cortex.
Memory is, in essence, a pattern-recognition process, adaptive in allowing us to detect predictive cues to guide behavior. To examine neural mechanisms of learning and memory, researchers typically present a set of stimuli repeatedly and examine how the brain's response to this information changes over time, or examine differences in neural activation to this stimulus set compared to a new stimulus set. In this issue, Scott Huettel and colleagues take a novel approach to the study of pattern recognition1, creating a situation that could lead participants into committing the gambler's fallacy (Fig. 1), the belief that chance events actually form coherent patterns. Using functional MRI to track blood flow to active neural regions, the researchers demonstrate that a distributed set of regions in prefrontal cortex are exquisitely sensitive to the presence of such patterns, and as importantly, to deviations from these patterns. Whether these neural responses reflect the operation of processes involved in short-term memory, novelty detection, or the generation of explicit predictions about forthcoming events remains to be determined.
Figure 1. We are often tempted into making predictions about upcoming events based on apparent patterns in the world around us, even when these patterns are random.
For example, a gambler may feel his or her best bet is to place money on black during a game of roulette because the last several runs came up black. Huettel et al. report in this issue that these moment-to-moment predictions involve activation of the prefrontal cortex.
The study is elegant in its simplicity. Participants were required to respond as quickly as possible to each stimulus in a series, pressing keys to indicate whether it was a square or a circle. The experimenters explicitly instructed participants that the events were randomly determined. However, over an extended period, even random sequences of stimuli will exhibit brief periods of seemingly 'non-random' patterns. The authors focus on two such series: runs of the same stimulus (such as six consecutive circles) and runs in which the two stimuli alternated (circle−square−circle−square−circle−square). Reaction times were influenced by these patterns. Up to repetitions of eight, responses became faster when the same stimulus was presented on successive trials, and the reaction time to a stimulus deviating from this pattern (for instance, a square following a series of circles) was considerably longer. Similar results were obtained for a series of alternations, although the behavioral effects here were less pronounced. These results suggest that the participants primed their motor behavior based on belief that the brief randomly emerging 'pattern' would continue, despite explicit instructions that the stimuli were determined randomly. Their incorrect predictions were associated with a clear behavioral cost.
The authors used an event-related fMRI design, focusing on the hemodynamic response triggered by a specific stimulus breaking these randomly occurring mini-patterns. For example, the neural response to a square that follows a series of five circles can be compared to a square that does not end a run of circle repetitions or one that perturbs a series of alternations (square−circle−square−circle−square−square); thus, the comparison is between trials involving the same stimulus and response. Repetition violations were marked by increased activation in a number of prefrontal foci, including middle and inferior frontal cortex bilaterally, as well as in cingulate and insular cortex in the right hemisphere. Subcortically, neural correlates of deviation were observed in the basal ganglia. Moreover, the magnitude of the signal was a direct function of the number of repetitions. The prefrontal regions also responded to violations of alternating sequences, although those blood flow changes were restricted to the right hemisphere.
The authors emphasize that the prefrontal regions "evaluate predictive mental models for upcoming events on a moment-to-moment basis," linking short-term pattern recognition processes to working memory operations associated with prefrontal cortex. Exactly how these prefrontal regions contribute to the evaluation process remains to be specified. We can think of at least three mechanisms, none of which precludes the others, which might contribute to the observed prefrontal activations. Indeed, all three may occur in the task used by the authors.
First, prefrontal activity could reflect a short-term memory mechanism, developing and maintaining a transient representation of the current context. A series of successive circles or an alternating square−circle pattern defines a particular context, and breaking this short-term memory template constitutes a violation of that context. The increased activation could be viewed as dishabituation of short-term memory mechanisms under control of prefrontal cortex, that is, a signal corresponding to the re-engagement of the short-term memory process in response to the change in context. There is evidence supporting this role of prefrontal cortex in short-term memory. The mismatch negativity (MMN) is an auditory event-related potential with a latency of 150−200 milliseconds that provides a marker of echoic memory2. The MMN is generated in auditory cortices by a break in either a repetitive or an alternating pattern, and this process is modulated by prefrontal cortex3. Thus, the current fMRI results could represent neural activity associated with transient maintenance of short-term memory processes. Electrophysiological, neuroimaging and neuromodeling evidence support such a role of prefrontal cortex in top-down modulation of posterior association cortex4,
5,
6. The neural representation of short-term context could be either explicit, in a form that the participants could verbally express, or implicit, developing outside awareness. The authors seem to favor an implicit interpretation, although the report does not provide definitive supporting evidence.
A second hypothesis is that the observed activations reflect a novelty response to changes in a pattern. In this view, the prefrontal cortex is engaged by deviance from a pattern. The idea is that prefrontal cortex is recruited to examine whether the break in local context deserves further processing. There is ERP evidence that this process is reflected by a longer-latency ERP occurring at 300−400 milliseconds after occurrence of a deviant event7. The novelty hypothesis need not assume that prefrontal regions are involved in the actual detection of a local emerging pattern or the break in this pattern. The representation of these patterns themselves may be in other cortical or subcortical areas, as is true for the auditory MMN. There is extensive intracranial recording, lesion, electrophysiological and fMRI evidence implicating the prefrontal cortext in novelty detection8. One way to distinguish these two hypotheses would be to record ERPs in the task. If a long-latency novelty ERP is observed, the second hypothesis would be supported.
Third, the prefrontal activations may represent an explicit hypothesis-generation process. Consider the participants, lying in the scanner, faithfully performing this simple task over and over again. Given the low processing demands of the task, it seems reasonable that they might monitor the series of events. Indeed, given the human propensity to find pattern among randomness, they may not be able to help themselves from engaging in such monitoring. This monitoring could well lead to the development of explicit hypotheses, such as "there is now a run of squares" or "the stimuli are alternating." As noted by the authors, the ability to generate and act on predictions can also have a cost. We may know that the next event in a series is randomly determined—be it the sequence of circles and squares in the current study or the next spin of the roulette wheel. But we are easily seduced by the immediate context, inferring causation or predictability even when none exists. Hypothesis generation is an essential feature of higher-level cognition. It should not be surprising to find that we engage in such operations despite explicit instructions regarding the random determination of the events.
Although the results of Huettel et al.1 provide clear evidence of prefrontal activation following pattern deviation, they do not discriminate among these three hypotheses. Moreover, it is conceivable that the distributed activations reported in this paper reflect prefrontal contributions to all three of the operations outlined above. Short-term memory mechanisms within prefrontal cortex may automatically modulate representations of local context in other cortical regions, as has been observed in echoic memory research. This short-term memory mechanism may then signal deviations from these patterns, providing a feedback signal to prefrontal regions to view a deviance as potentially biologically significant and deserving of additional inspection. Such representations may also be accessible for systems involved in the explicit monitoring of behavior and generation of hypotheses. Evidence from neuroimaging, neuropsychology and electrophysiology supports a role of prefrontal regions in all these cognitive operations3,
8,
9
The current work can be compared to more traditional studies of pattern recognition. One approach has involved trial-and-error learning, in which participants are explicitly instructed to try to learn a response sequence10,
11. Another has used the serial reaction-time (SRT) task, in which participants make speeded responses to successive stimuli that vary along a dimension such as spatial position or color. The SRT task is similar to that of Huettel et al.1, except that the number of stimulus−response alternatives is increased, and comparisons are made between blocks of trials in which the events follow a fixed sequence or occur randomly12,
13. This task has been used to study both explicit and implicit learning. In the former condition, participants are either taught the sequence in advance or extract it over the course of the experiment. In the latter, a distractor task is interleaved with the button-pressing task to distract the participant's attention and thus reduce awareness of the sequential nature of events.
These studies have yielded a consistent picture regarding prefrontal activation during sequence learning. When learning is explicitly guided, or when participants become aware of the sequence, prominent activation is observed in prefrontal cortex, including the areas identified by Huettel et al. In contrast, when learning is implicit, no changes are found in lateral prefrontal cortex, even though performance measures clearly indicate that the participants have learned the sequence of stimuli and/or responses. Under such conditions, pattern recognition occurs without hypothesis generation. Thus, the sequence-learning literature is consistent with the claims of Huettel et al.1 regarding a role for prefrontal cortex in the perception (or production) of patterns, but also suggests that this role may be limited to situations in which the participants are able to explicitly express these expectations.
Future work in which the level of the participants awareness is monitored should provide a direct test of this hypothesis. We would expect a marked reduction of the prefrontal response in the current task if the participants were engaged in a distractor task that disrupted their ability to generate hypotheses. "An idle mind is the Devil's workshop," goes an old English proverb. For the gambler impressed by the run of black spots in roulette, an idle mind may result in unwarranted predictions that can lead to misguided actions.
