1. Introduction
Launched by Aristotle (Back Reference Back2014), interest in abstraction resurged in psychology (Burgoon et al. Reference Burgoon, Henderson and Markman2013; Reed Reference Reed2016) and spread to artificial intelligence (Saitta & Zucker Reference Saitta and Zucker2013) because it is important for understanding and problem-solving in variable environments. The target article proposes a theory of abstraction as a basis for mental time travel. In this theory, representations are parsed into three categories (modality-specific, multimodal, and categorical representations, emerging from perceptual similarity, spatiotemporal contiguity, and social interactions, respectively). These are building blocks, abstracta, for the construction of complex structures (episodes, scripts, and hierarchies) predicated by language at increasingly higher levels of abstraction. The structures are sources of predictive processing. Substitutivity, the recognition that two or more elements may stand for the same referent, is the fundamental mechanism of abstraction, underlying generalization, reduction, categorization, and analogical induction. Other theories stress other mechanisms as the basis of abstraction, such as identifying invariant central characteristics (Burgoon et al. Reference Burgoon, Henderson and Markman2013) and discrimination (Reed Reference Reed2016). We argue that the proposed theory of abstraction (sect. 1 and 2) is weak in several respects, thereby failing to account for predictive cognition (sect. 3 and 4).
2. Abstraction in learning and development
Abstraction is partly indeterminate: its very operation changes its subsequent state and products, rendering future abstractions different. Therefore, any theory of abstraction must account for how learning, in the short-term, and development, in the long-term, change abstraction. This theory does not involve any such provisions: it does not explicate how modality-specific abstractions are formed at the first place nor does it account for their integration into multimodal and categorical representations. Simply naming the origins of substitutivity is not enough. We need to know how innate abstractions emerge out of interactions with the physical properties of the world at the first place, how they are redefined by personal experience, and how they are reshaped by social interactions. We also need to know how and when abstraction processes change.
The brain evidence invoked as supportive (sect. 5) is as global as the abstraction theory itself. Localizing different forms of abstraction in different brain regions, even if accurate, says nothing about abstraction itself. We need to know how brain regions operate and speak to each other when forming abstractions. Optimum connectivity defines the precision of abstraction (Raju, Reference Rajuin preparation); brain rhythms may be the lexicon and their coordination the syntax of brain language (Buzsaki Reference Buzsaki2010; Demetriou & Spanoudis Reference Demetriou and Spanoudis2018).
Cognitive development is the development of abstraction. Thus, it was central in all cognitive developmental theories. In Piaget's (Reference Piaget2001) theory, abstraction is the engine of equilibration, the central mechanism of cognitive development. In our theory, abstraction is part of a tripartite system involving, additionally, alignment processes generating relations feeding abstraction, and cognizance, awareness of mental processes and their products, allowing metarepresentation yielding abstracta (Demetriou et al. Reference Demetriou, Makris, Kazi, Spanoudis and Shayer2018a). Levels of cognitive development reflect the ontogeny of abstraction. The proposed theory must explicate how modality-specific, multimodal, and categorical abstractions emerge at successive cognitive developmental cycles. In infancy, before language, modality-specific abstractions generate the primary material that in toddlerhood, with language, will be weaved into complex multimodal realistic representations. Later, in primary school, these multimodal representations are organized into rule-based categories increasingly predicated by language. In adolescence, rules and predications are meta-represented by principles predicating truth, protecting from deception (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018).
Bayesian inference underlying abstraction dominates early in learning or development (Tenenbaum et al. Reference Tenenbaum, Kemp, Griffiths and Goodman2011); logical mechanisms in analogical and deductive reasoning dominate later (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018). Their precise proportion and overall representational profile involving modality-specific, multimodal, and categorical representations at different phases of learning and development is not specified in the target article. Also, it is important to specify how reflection integrates modality-specific abstractions into multimodal abstractions and how awareness of abstraction processes underlies categorical predication. Learning research shows that guided and reflected upon relational processing generates abstraction in different domains (Jee & Anggoro Reference Jee and Anggoro2019; Papageorgiou et al. Reference Papageorgiou, Christou, Spanoudis and Demetriou2016).
3. Abstraction in individual differences
In classical theory of intelligence, individual differences in intelligence reflect differences in abstraction. The very notion of general (Jensen Reference Jensen1998; Spearman Reference Spearman1904) or fluid intelligence (Cattell & Horn Reference Cattell and Horn1978) is basically abstraction coming under different names (e.g., Spearman's eduction of relations and correlates). Individual differences in intelligence reflect differences in how far individuals progressed along the developmental course of abstraction outlined above (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018). Therefore, higher intelligence reflects more increasingly flexible, to-the-point, abstractions employed for long-term predictions about the world.
Processing and representational efficiency constrain abstraction. For instance, attention guides abstraction to relevant information and working memory provides the field where it occurs (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018; Demetriou et al. Reference Demetriou, Makris, Kazi, Spanoudis, Shayer and Kazali2018b). Attention and working memory lapsuses may misdirect or disorganize abstraction. The paper is again silent about how abstraction interacts with these processes.
4. Mental time travel
In conclusion, the theory presented in the target article does not specify how abstraction is done at different levels of development, ability, or learning, what is the difference in abstraction between different levels of intelligence, and how we learn to abstract in different contexts or contents. Thus, the theory is weak in explicating mental time travel in relation to different types of representations. For instance, the episodic representations of the pre-language infant allow some time of prospection: Infants have predictive models of their behavior vis-à-vis their environment that protect them from falling or colliding with objects; however, they may dangerously err in unfamiliar environments. The realistic mental representations of the toddler allow social prospection: toddlers predict others’ behavior based on their knowledge of others’ mental states; however, they may seriously err if others’ behavior is based on different values about these mental states, which they do not know. Primary school children use abstraction to foresee their daily activities. Adolescents build prospective models of their life as adults. The target article is silent about how different types of representations and related abstractions engender different types of perspective in mental time travel at different phases of life.
1. Introduction
Launched by Aristotle (Back Reference Back2014), interest in abstraction resurged in psychology (Burgoon et al. Reference Burgoon, Henderson and Markman2013; Reed Reference Reed2016) and spread to artificial intelligence (Saitta & Zucker Reference Saitta and Zucker2013) because it is important for understanding and problem-solving in variable environments. The target article proposes a theory of abstraction as a basis for mental time travel. In this theory, representations are parsed into three categories (modality-specific, multimodal, and categorical representations, emerging from perceptual similarity, spatiotemporal contiguity, and social interactions, respectively). These are building blocks, abstracta, for the construction of complex structures (episodes, scripts, and hierarchies) predicated by language at increasingly higher levels of abstraction. The structures are sources of predictive processing. Substitutivity, the recognition that two or more elements may stand for the same referent, is the fundamental mechanism of abstraction, underlying generalization, reduction, categorization, and analogical induction. Other theories stress other mechanisms as the basis of abstraction, such as identifying invariant central characteristics (Burgoon et al. Reference Burgoon, Henderson and Markman2013) and discrimination (Reed Reference Reed2016). We argue that the proposed theory of abstraction (sect. 1 and 2) is weak in several respects, thereby failing to account for predictive cognition (sect. 3 and 4).
2. Abstraction in learning and development
Abstraction is partly indeterminate: its very operation changes its subsequent state and products, rendering future abstractions different. Therefore, any theory of abstraction must account for how learning, in the short-term, and development, in the long-term, change abstraction. This theory does not involve any such provisions: it does not explicate how modality-specific abstractions are formed at the first place nor does it account for their integration into multimodal and categorical representations. Simply naming the origins of substitutivity is not enough. We need to know how innate abstractions emerge out of interactions with the physical properties of the world at the first place, how they are redefined by personal experience, and how they are reshaped by social interactions. We also need to know how and when abstraction processes change.
The brain evidence invoked as supportive (sect. 5) is as global as the abstraction theory itself. Localizing different forms of abstraction in different brain regions, even if accurate, says nothing about abstraction itself. We need to know how brain regions operate and speak to each other when forming abstractions. Optimum connectivity defines the precision of abstraction (Raju, Reference Rajuin preparation); brain rhythms may be the lexicon and their coordination the syntax of brain language (Buzsaki Reference Buzsaki2010; Demetriou & Spanoudis Reference Demetriou and Spanoudis2018).
Cognitive development is the development of abstraction. Thus, it was central in all cognitive developmental theories. In Piaget's (Reference Piaget2001) theory, abstraction is the engine of equilibration, the central mechanism of cognitive development. In our theory, abstraction is part of a tripartite system involving, additionally, alignment processes generating relations feeding abstraction, and cognizance, awareness of mental processes and their products, allowing metarepresentation yielding abstracta (Demetriou et al. Reference Demetriou, Makris, Kazi, Spanoudis and Shayer2018a). Levels of cognitive development reflect the ontogeny of abstraction. The proposed theory must explicate how modality-specific, multimodal, and categorical abstractions emerge at successive cognitive developmental cycles. In infancy, before language, modality-specific abstractions generate the primary material that in toddlerhood, with language, will be weaved into complex multimodal realistic representations. Later, in primary school, these multimodal representations are organized into rule-based categories increasingly predicated by language. In adolescence, rules and predications are meta-represented by principles predicating truth, protecting from deception (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018).
Bayesian inference underlying abstraction dominates early in learning or development (Tenenbaum et al. Reference Tenenbaum, Kemp, Griffiths and Goodman2011); logical mechanisms in analogical and deductive reasoning dominate later (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018). Their precise proportion and overall representational profile involving modality-specific, multimodal, and categorical representations at different phases of learning and development is not specified in the target article. Also, it is important to specify how reflection integrates modality-specific abstractions into multimodal abstractions and how awareness of abstraction processes underlies categorical predication. Learning research shows that guided and reflected upon relational processing generates abstraction in different domains (Jee & Anggoro Reference Jee and Anggoro2019; Papageorgiou et al. Reference Papageorgiou, Christou, Spanoudis and Demetriou2016).
3. Abstraction in individual differences
In classical theory of intelligence, individual differences in intelligence reflect differences in abstraction. The very notion of general (Jensen Reference Jensen1998; Spearman Reference Spearman1904) or fluid intelligence (Cattell & Horn Reference Cattell and Horn1978) is basically abstraction coming under different names (e.g., Spearman's eduction of relations and correlates). Individual differences in intelligence reflect differences in how far individuals progressed along the developmental course of abstraction outlined above (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018). Therefore, higher intelligence reflects more increasingly flexible, to-the-point, abstractions employed for long-term predictions about the world.
Processing and representational efficiency constrain abstraction. For instance, attention guides abstraction to relevant information and working memory provides the field where it occurs (Demetriou & Spanoudis Reference Demetriou and Spanoudis2018; Demetriou et al. Reference Demetriou, Makris, Kazi, Spanoudis, Shayer and Kazali2018b). Attention and working memory lapsuses may misdirect or disorganize abstraction. The paper is again silent about how abstraction interacts with these processes.
4. Mental time travel
In conclusion, the theory presented in the target article does not specify how abstraction is done at different levels of development, ability, or learning, what is the difference in abstraction between different levels of intelligence, and how we learn to abstract in different contexts or contents. Thus, the theory is weak in explicating mental time travel in relation to different types of representations. For instance, the episodic representations of the pre-language infant allow some time of prospection: Infants have predictive models of their behavior vis-à-vis their environment that protect them from falling or colliding with objects; however, they may dangerously err in unfamiliar environments. The realistic mental representations of the toddler allow social prospection: toddlers predict others’ behavior based on their knowledge of others’ mental states; however, they may seriously err if others’ behavior is based on different values about these mental states, which they do not know. Primary school children use abstraction to foresee their daily activities. Adolescents build prospective models of their life as adults. The target article is silent about how different types of representations and related abstractions engender different types of perspective in mental time travel at different phases of life.