Gilead and colleagues propose that abstraction underlies the predictive nature of cognition. Functional neuroimaging and lesion mapping studies indicate that abstraction and long-term prediction are both associated with the human prefrontal cortex (HPFC). I suggest that the HPFC is also a good fit for the storage and use of time- and event-related higher order representational knowledge – contributing to predictive processing beyond a single event (Wood & Grafman Reference Wood and Grafman2003). I have no dispute with Gilead et al. using the term hierarchical to describe the dominant role of the HPFC in storing, organizing, controlling, and issuing the intentional commands to execute most daily routines or adaptive behavior. Although this is by no means a new idea (e.g., see Miller et al. Reference Miller, Galanter and Pribram1986; Schank & Abelson Reference Schank and Abelson1977), the recent development of statistical and computational modeling and information theory tools Gilead et al. describe have substantially improved the quantification of hierarchical models and their predictions.
The environment that surrounds us contains events of various durations. One possible way humans have come to represent events of varying durations is by parsing the brain into differing representational sectors with each sector containing representations of event series of differing durations (Radvansky & Zacks Reference Radvansky and Zacks2017). This is a hierarchical schema, with increasing number of events composed of longer durations more likely to be stored in the anterior sectors of the HPFC. Prediction would require retrieving such stored events held in long-term memory (Grafman et al. Reference Grafman, Spector and Rattermann2005). Exemplar features such as complexity, abstractness, frequency of exposure, and relational similarity could emerge from the organizational structure of these representational networks. Flexible parsing rules (see Zacks et al. Reference Zacks, Kurby, Landazabal, Krueger and Grafman2016) allowed events to be represented in structured event complexes. This organizational principle could facilitate switching between complex representations and abbreviated heuristic knowledge contained in posterior cortical/subcortical networks that store more time-compact single event representations conserving the brain's free energy (Parr & Friston Reference Parr and Friston2019).
These HPFC structured event complexes including plans, narratives, abstractions, deliberations, and goal-derived activities would be stored in networks along the same principles that enable words or objects to be stored in a coherent cognitive network. This would, for example, allow for prioritizing structured event complexes performed frequently in the real world by giving their representations cortical space distinctiveness and activation default superiority (Grafman et al. Reference Grafman, Thompson, Weingartner, Martinez, Lawlor and Sunderland1991; Rosen et al. Reference Rosen, Caplan, Sheesley, Rodriguez and Grafman2003). Although this would be true of any stored memory (e.g., words or objects), it is particularly crucial for sustaining the activation of longer duration memory representations that support goal achievement and social navigation. But, why couldn't our brain just be organized around short time duration representations that were dynamically and adaptively repackaged into a sequence depending upon intended or environmentally provoked behavior? That would require neural resources that would be continually engaged in such activity causing laborious multi-tasking, siphoning the brain's energy deposits and cognitive resources.
Using functional neuroimaging, Etienne Koechlin and David Badre have demonstrated that distinct brain regions may support different levels of a hierarchical processing framework (Badre & Nee Reference Badre and Nee2018; Koechlin & Jubault Reference Koechlin and Jubault2006). Although these authors have focused mostly on left dorsolateral prefrontal cortex language mechanisms, their studies have demonstrated the feasibility of using a mathematically constrained hierarchical model to predict differing levels of processing and representation. Other functional neuroimaging and lesion mapping studies of chess players (Nichelli et al. Reference Nichelli, Grafman, Pietrini, Alway, Carton and Miletich1994), processing narratives (Nichelli et al. Reference Nichelli, Grafman, Pietrini, Clark, Lee and Miletich1995), and script decision-making (Sirigu et al. Reference Sirigu, Zalla, Pillon, Grafman, Agid and Dubois1996) have supported the structured event complex (SEC) conceptualization.
But, how are durations first encoded and later parsed? There is some evidence that individual segments or events are first tabulated in childhood, later compiled into sequences that can then be reduced to heuristics and stored posteriorly or deeper in the brain (Rattermann et al. Reference Rattermann, Spector, Grafman, Levin and Harward2001). This hierarchy of duration representation doesn't mean one form replaces another as we mature and learn. Rather, it is likely that all of these forms of representations remain stored across the HPFC with various representational forms activated depending on situational needs.
Cautions
A challenge for comparing hierarchical forms of knowledge to see if they occupy distinct brain sectors is to make sure the compared tasks are psychometrically matched. For example, controlling for the difficulty in retrieving different categories of words based on frequency or age of acquisition of the word is a magnitude easier than controlling for the difficulty level of abstract tasks or event narratives. Chapman and Chapman described the problems and solutions in creating tasks that could be psychometrically comparable when focusing on dissociations decades ago (Chapman & Chapman Reference Chapman and Chapman1978; Reference Chapman and Chapman2001). Also note that much of the literature Gilead et al. cite depends on correlational functional neuroimaging findings in healthy volunteers rather than using causal lesion mapping or non-invasive brain stimulation results. Providing convergent data from studies using different techniques to muster support for a specific idea or theory is important.
To summarize, a complementary perspective to Gilead and colleagues' all-encompassing view about the influence of abstractions in predictive processing is that the HPFC evolved to capture events that occur over longer and longer periods of time. All deliberative thought including complex abstractions, social agency, and narrative explanations that fit within structured event complexes devoted to representing information occurring over multiple events and time periods will be represented in the HPFC. Such HPFC representations will bind via a variety of neural mechanisms to representations stored elsewhere in the brain providing a relatively complete network-based capture of all the information composing that time frame. The large number of exemplars within each hierarchical SEC network would allow abstract representations to emerge that capture the similarities across representational exemplars, but preserve the individuated representations that are critical to remembering episodes. Given that longer time frames impose greater persistence toward the completion of an activity and achievement of a goal, activation of one or more structured event complexes inhibit diversions enabling reinforced and rewarded goal-directed activity in the face of environmental distractors. Other species cannot compete with humans because of the richness and multiplicity of our structured event complexes. Forming predictive abstractions are dependent on the existence of structured event complexes.
You have
Access
Gilead and colleagues propose that abstraction underlies the predictive nature of cognition. Functional neuroimaging and lesion mapping studies indicate that abstraction and long-term prediction are both associated with the human prefrontal cortex (HPFC). I suggest that the HPFC is also a good fit for the storage and use of time- and event-related higher order representational knowledge – contributing to predictive processing beyond a single event (Wood & Grafman Reference Wood and Grafman2003). I have no dispute with Gilead et al. using the term hierarchical to describe the dominant role of the HPFC in storing, organizing, controlling, and issuing the intentional commands to execute most daily routines or adaptive behavior. Although this is by no means a new idea (e.g., see Miller et al. Reference Miller, Galanter and Pribram1986; Schank & Abelson Reference Schank and Abelson1977), the recent development of statistical and computational modeling and information theory tools Gilead et al. describe have substantially improved the quantification of hierarchical models and their predictions.
The environment that surrounds us contains events of various durations. One possible way humans have come to represent events of varying durations is by parsing the brain into differing representational sectors with each sector containing representations of event series of differing durations (Radvansky & Zacks Reference Radvansky and Zacks2017). This is a hierarchical schema, with increasing number of events composed of longer durations more likely to be stored in the anterior sectors of the HPFC. Prediction would require retrieving such stored events held in long-term memory (Grafman et al. Reference Grafman, Spector and Rattermann2005). Exemplar features such as complexity, abstractness, frequency of exposure, and relational similarity could emerge from the organizational structure of these representational networks. Flexible parsing rules (see Zacks et al. Reference Zacks, Kurby, Landazabal, Krueger and Grafman2016) allowed events to be represented in structured event complexes. This organizational principle could facilitate switching between complex representations and abbreviated heuristic knowledge contained in posterior cortical/subcortical networks that store more time-compact single event representations conserving the brain's free energy (Parr & Friston Reference Parr and Friston2019).
These HPFC structured event complexes including plans, narratives, abstractions, deliberations, and goal-derived activities would be stored in networks along the same principles that enable words or objects to be stored in a coherent cognitive network. This would, for example, allow for prioritizing structured event complexes performed frequently in the real world by giving their representations cortical space distinctiveness and activation default superiority (Grafman et al. Reference Grafman, Thompson, Weingartner, Martinez, Lawlor and Sunderland1991; Rosen et al. Reference Rosen, Caplan, Sheesley, Rodriguez and Grafman2003). Although this would be true of any stored memory (e.g., words or objects), it is particularly crucial for sustaining the activation of longer duration memory representations that support goal achievement and social navigation. But, why couldn't our brain just be organized around short time duration representations that were dynamically and adaptively repackaged into a sequence depending upon intended or environmentally provoked behavior? That would require neural resources that would be continually engaged in such activity causing laborious multi-tasking, siphoning the brain's energy deposits and cognitive resources.
Using functional neuroimaging, Etienne Koechlin and David Badre have demonstrated that distinct brain regions may support different levels of a hierarchical processing framework (Badre & Nee Reference Badre and Nee2018; Koechlin & Jubault Reference Koechlin and Jubault2006). Although these authors have focused mostly on left dorsolateral prefrontal cortex language mechanisms, their studies have demonstrated the feasibility of using a mathematically constrained hierarchical model to predict differing levels of processing and representation. Other functional neuroimaging and lesion mapping studies of chess players (Nichelli et al. Reference Nichelli, Grafman, Pietrini, Alway, Carton and Miletich1994), processing narratives (Nichelli et al. Reference Nichelli, Grafman, Pietrini, Clark, Lee and Miletich1995), and script decision-making (Sirigu et al. Reference Sirigu, Zalla, Pillon, Grafman, Agid and Dubois1996) have supported the structured event complex (SEC) conceptualization.
But, how are durations first encoded and later parsed? There is some evidence that individual segments or events are first tabulated in childhood, later compiled into sequences that can then be reduced to heuristics and stored posteriorly or deeper in the brain (Rattermann et al. Reference Rattermann, Spector, Grafman, Levin and Harward2001). This hierarchy of duration representation doesn't mean one form replaces another as we mature and learn. Rather, it is likely that all of these forms of representations remain stored across the HPFC with various representational forms activated depending on situational needs.
Cautions
A challenge for comparing hierarchical forms of knowledge to see if they occupy distinct brain sectors is to make sure the compared tasks are psychometrically matched. For example, controlling for the difficulty in retrieving different categories of words based on frequency or age of acquisition of the word is a magnitude easier than controlling for the difficulty level of abstract tasks or event narratives. Chapman and Chapman described the problems and solutions in creating tasks that could be psychometrically comparable when focusing on dissociations decades ago (Chapman & Chapman Reference Chapman and Chapman1978; Reference Chapman and Chapman2001). Also note that much of the literature Gilead et al. cite depends on correlational functional neuroimaging findings in healthy volunteers rather than using causal lesion mapping or non-invasive brain stimulation results. Providing convergent data from studies using different techniques to muster support for a specific idea or theory is important.
To summarize, a complementary perspective to Gilead and colleagues' all-encompassing view about the influence of abstractions in predictive processing is that the HPFC evolved to capture events that occur over longer and longer periods of time. All deliberative thought including complex abstractions, social agency, and narrative explanations that fit within structured event complexes devoted to representing information occurring over multiple events and time periods will be represented in the HPFC. Such HPFC representations will bind via a variety of neural mechanisms to representations stored elsewhere in the brain providing a relatively complete network-based capture of all the information composing that time frame. The large number of exemplars within each hierarchical SEC network would allow abstract representations to emerge that capture the similarities across representational exemplars, but preserve the individuated representations that are critical to remembering episodes. Given that longer time frames impose greater persistence toward the completion of an activity and achievement of a goal, activation of one or more structured event complexes inhibit diversions enabling reinforced and rewarded goal-directed activity in the face of environmental distractors. Other species cannot compete with humans because of the richness and multiplicity of our structured event complexes. Forming predictive abstractions are dependent on the existence of structured event complexes.