Elizabeth Spelke (Spelke, Reference Spelke2022) championed in her book – and throughout her scientific career – the idea that research on infants and (non-human) animals should synergize and thrive. In particular, research on human newborns and animals of precocial species (which, differently than those of altricial species, are at mature sensory and motoric levels at birth) would allow a direct approach to fundamental questions: What do organisms know at the time when their learning begins? And what makes their learning go so well, that is, so quickly?
Here I want to remark on some crucial insights that arose from a comparative approach such as this. The first two examples show up as a sort of clash of evidence from human and non-human studies, and their resolution yields, I believe, real progress in scientific understanding.
The first example is described by Spelke in detail in the book, and it is related to recognition of partly occluded objects, but I will consider it as regards one neglected aspect of it. When we published, more than 35 year ago (Regolin & Vallortigara, Reference Regolin and Vallortigara1995), our results showing that newborn chicks are capable of recognizing partly occluded objects (or “amodal completion” as Michotte, Thinés, & Crabbé [Reference Michotte, Thinés and Crabbé1964] had dubbed the phenomenon), and then Lea, Slater, and Ryan (Reference Lea, Slater and Ryan1996) duplicated our results using a procedure more similar to that employed by developmental psychologists, we were faced with an apparent paradox. Even taking into account that chicks are precocial and human infants altricial, it appeared that at least 4 months were needed for young humans to develop a capacity for amodal completion that chicks showed at the onset of life (Vallortigara, Reference Vallortigara2021). It is tempting in these cases to argue that humans, differently than other animals, rely more on learning rather than on innate predispositions. But this is a weird argument in my view: If the machinery for amodal completion is available at the start as an evolutionarily granted mechanism (based on rules associated with the way in which junctions of objects' boundaries occur in visual scenes), why should an organism accept the cost of acquiring such a capacity by trial and error learning, which is likely to be very long and is open to risks of mistakes? The solution, as Spelke tells us in the book, was provided by Valenza, Leo, Gava, and Simion (Reference Valenza, Leo, Gava and Simion2006), who showed that using stroboscopic motion, rather than slow, gradual motion, as human newborns do, shows evidence for amodal completion at birth. Claiming innateness is obviously not the same as claiming that a certain capacity is operational at birth, for this would depend on the overall pattern of development of a species. Precocial species such as chicks have mechanisms for motion perception which are mature soon after hatching, whereas in humans, an altricial species, these mechanisms will mature later on. In the latter species, it may be difficult to reveal the presence of amodal completion at birth, in spite of the fact that the mechanism is already there, and it is indeed innate, for it simply does not show up until proper testing conditions are used (such as those measuring stroboscopic motion).
The second example is more recent. Starting from the classical work of Francis Galton (Reference Galton1880), the idea that humans have a mental number line has obtained several confirmations (e.g., Dehaene, Bossini, & Giraux, Reference Dehaene, Bossini and Giraux1993). Number and space seem to be inherently associated, and these kinds of phenomena (also referred to as SNAs, as in Space-Number Associations) have been extensively investigated. Yet the debate regarding their nature and origin remains hot. For example, in recent years, two papers have been published, among others, trying to unveil the origin of SNAs, that reached diametrically opposite conclusions: One showing that chicks exhibit something similar to an ordered mental number line (Rugani, Vallortigara, Priftis, & Regolin, Reference Rugani, Vallortigara, Priftis and Regolin2015), the other showing that a traditional human population lacking any formal arithmetic does not show any significant left-to-right bias (Pitt et al., Reference Pitt, Ferrigno, Cantlon, Casasanto, Gibson and Piantadosi2021). On the basis of second paper and several other anthropological studies, it can be concluded that SNAs are determined by cultural habits dependent upon literacy (e.g., the direction of reading/writing), as they were absent in preschoolers and indigenous people from oral tribes. On the basis of the first paper, one can conclude, on the contrary, that SNAs (and in particular, a left-to-right mapping) are biologically, and not culturally, determined; indeed, such associations are observed in pre- or non-verbal subjects, such as human newborns (Di Giorgio et al., Reference Di Giorgio, Lunghi, Rugani, Regolin, Dalla Barba, Vallortigara and Simion2019) and in several non-human animal species (e.g., monkeys: Drucker & Brannon, Reference Drucker and Brannon2014; honeybees: Giurfa, Marcout, Hilpert, Thevenot, & Rugani, Reference Giurfa, Marcout, Hilpert, Thevenot and Rugani2022).
Recently, we aimed to resolve these apparently conflicting findings (Eccher et al., Reference Eccher, Josserand, Caparos, Boissin, Buiatti, Piazza and Vallortigara2023). We conducted two behavioral experiments in three populations of different ages and cultural backgrounds: Italian adults, Italian preschoolers, and adults from the Himba tribe (an indigenous African tribe with no writing system). Our results showed that when tested with explicit tasks, only Italian adults show a consistent SNA, while when tested with implicit tasks, all the three populations exhibit a common and consistent left-to-right-oriented SNA.
These results support the hypothesis that the SNA phenomenon is dissociable into two different components: One which is acquired and cultural-dependent, and one which is biologically predisposed (note that the underlying mechanisms can be nonetheless quite different from those of a proper number line and rely instead on brain asymmetry, see, for a specific hypothesis, Vallortigara, Reference Vallortigara2018). But apart from the particular case in point, what seems interesting to me is that there is a special value in this sort of comparative research, namely the fact that evidence in non-human species forced us to reconsider our hypotheses on human nature.
The third insight concerns the proper way to build up intelligence in artificial systems. Plasticity seems to be a magic word nowadays in neuroscience and also in artificial intelligence (AI; but see Marcus, Reference Marcus2018). Comparative research on core knowledge tells us a different story. Consider what we learned from newly hatched chicks. These animals seem to be predisposed to orient toward objects that possess features associated with animate objects, such as biological motion, changes in speed, and face-like configurations (review in Di Giorgio et al., Reference Di Giorgio, Loveland, Mayer, Rosa-Salva, Versace and Vallortigara2017; Vallortigara, Reference Vallortigara2021 see also Vallortigara, Regolin, & Marconato, Reference Vallortigara, Regolin and Marconato2005). These are unlearned priors that help chicks to orient toward the mother hen and their siblings, thus facilitating and guiding a robust learning process called filial imprinting. (Similar mechanisms have been documented in human newborns [Lorenzi & Vallortigara, Reference Lorenzi, Vallortigara, Kaufman, Call and Kaufman2021; Vallortigara, Reference Vallortigara2012, Reference Vallortigara2021], even though in this species, proper control of past experience and access to neural substrates are limited for obvious reasons.)
The issue then is how can young organisms orient toward the “right” stimulus in the absence of any previous experience? In contrast to machine-learning systems, biological organisms do not require explicit reinforcement, supervised learning, or thousands/millions of examples to feed learning. They are equipped with dedicated orienting mechanisms that work as adaptive priors that imply some assumptions about the external world that guide learning (Versace, Martinho-Truswel, Kacelnik, & Vallortigara, Reference Versace, Martinho-Truswel, Kacelnik and Vallortigara2018). The priors are sufficiently general to allow errors. For instance, early preferences of chicks are not strictly species-specific but apply equally to hen face-like or polecat face-like features or to the biological-motion appearance of either a hen or a cat (review in Vallortigara, Reference Vallortigara2021). There is a profound biological reason for that. The predisposed orienting mechanisms cannot be too specific for the individual features, given that these are to some extent unpredictable from the genetic repertoire (because of variability between adults within a species and due to changes in the appearance of even a single individual).
However, high plasticity coupled with prior assumptions is not enough. In biological organisms, both early predispositions and high plasticity are transient phenomena that end either with some maturational processes or when the necessary information has been acquired. The existence of critical and sensitive periods has been documented in several domains and functions (Hensch & Quinlan, Reference Hensch and Quinlan2018). However, a very important recent finding is that critical periods do not apply only to the plasticity associated with learning but also to the periods of expression of the priors themselves. For instance, cues of animacy associated with speed changes are expressed in chicks during a restricted period in early life and can be reopened by the administration of certain substances (Lorenzi, Lemaire, Versace, Matsushima, & Vallortigara, Reference Lorenzi, Lemaire, Versace, Matsushima and Vallortigara2021).
There are costs associated with neural plasticity, and this is the reason why, after a certain age, learning new languages and solving amblyopia is so hard, and why early experiences are important for subsequent stages of life. Thus, the plasticity of the nervous system is actively reduced by molecular “brakes” that promote the stabilization of mature brain function (Hensch & Quinlan, Reference Hensch and Quinlan2018).
Elizabeth Spelke's book seems to suggest, among other things, that AI systems could benefit from being equipped with a set of priors that offer a guidance system and a way to speed up plasticity associated with learning mechanisms. I believe, however, that plasticity without critical periods of expression for these priors might have costs that prevent effective learning and cognitive functions.
You have
Access
Elizabeth Spelke (Spelke, Reference Spelke2022) championed in her book – and throughout her scientific career – the idea that research on infants and (non-human) animals should synergize and thrive. In particular, research on human newborns and animals of precocial species (which, differently than those of altricial species, are at mature sensory and motoric levels at birth) would allow a direct approach to fundamental questions: What do organisms know at the time when their learning begins? And what makes their learning go so well, that is, so quickly?
Here I want to remark on some crucial insights that arose from a comparative approach such as this. The first two examples show up as a sort of clash of evidence from human and non-human studies, and their resolution yields, I believe, real progress in scientific understanding.
The first example is described by Spelke in detail in the book, and it is related to recognition of partly occluded objects, but I will consider it as regards one neglected aspect of it. When we published, more than 35 year ago (Regolin & Vallortigara, Reference Regolin and Vallortigara1995), our results showing that newborn chicks are capable of recognizing partly occluded objects (or “amodal completion” as Michotte, Thinés, & Crabbé [Reference Michotte, Thinés and Crabbé1964] had dubbed the phenomenon), and then Lea, Slater, and Ryan (Reference Lea, Slater and Ryan1996) duplicated our results using a procedure more similar to that employed by developmental psychologists, we were faced with an apparent paradox. Even taking into account that chicks are precocial and human infants altricial, it appeared that at least 4 months were needed for young humans to develop a capacity for amodal completion that chicks showed at the onset of life (Vallortigara, Reference Vallortigara2021). It is tempting in these cases to argue that humans, differently than other animals, rely more on learning rather than on innate predispositions. But this is a weird argument in my view: If the machinery for amodal completion is available at the start as an evolutionarily granted mechanism (based on rules associated with the way in which junctions of objects' boundaries occur in visual scenes), why should an organism accept the cost of acquiring such a capacity by trial and error learning, which is likely to be very long and is open to risks of mistakes? The solution, as Spelke tells us in the book, was provided by Valenza, Leo, Gava, and Simion (Reference Valenza, Leo, Gava and Simion2006), who showed that using stroboscopic motion, rather than slow, gradual motion, as human newborns do, shows evidence for amodal completion at birth. Claiming innateness is obviously not the same as claiming that a certain capacity is operational at birth, for this would depend on the overall pattern of development of a species. Precocial species such as chicks have mechanisms for motion perception which are mature soon after hatching, whereas in humans, an altricial species, these mechanisms will mature later on. In the latter species, it may be difficult to reveal the presence of amodal completion at birth, in spite of the fact that the mechanism is already there, and it is indeed innate, for it simply does not show up until proper testing conditions are used (such as those measuring stroboscopic motion).
The second example is more recent. Starting from the classical work of Francis Galton (Reference Galton1880), the idea that humans have a mental number line has obtained several confirmations (e.g., Dehaene, Bossini, & Giraux, Reference Dehaene, Bossini and Giraux1993). Number and space seem to be inherently associated, and these kinds of phenomena (also referred to as SNAs, as in Space-Number Associations) have been extensively investigated. Yet the debate regarding their nature and origin remains hot. For example, in recent years, two papers have been published, among others, trying to unveil the origin of SNAs, that reached diametrically opposite conclusions: One showing that chicks exhibit something similar to an ordered mental number line (Rugani, Vallortigara, Priftis, & Regolin, Reference Rugani, Vallortigara, Priftis and Regolin2015), the other showing that a traditional human population lacking any formal arithmetic does not show any significant left-to-right bias (Pitt et al., Reference Pitt, Ferrigno, Cantlon, Casasanto, Gibson and Piantadosi2021). On the basis of second paper and several other anthropological studies, it can be concluded that SNAs are determined by cultural habits dependent upon literacy (e.g., the direction of reading/writing), as they were absent in preschoolers and indigenous people from oral tribes. On the basis of the first paper, one can conclude, on the contrary, that SNAs (and in particular, a left-to-right mapping) are biologically, and not culturally, determined; indeed, such associations are observed in pre- or non-verbal subjects, such as human newborns (Di Giorgio et al., Reference Di Giorgio, Lunghi, Rugani, Regolin, Dalla Barba, Vallortigara and Simion2019) and in several non-human animal species (e.g., monkeys: Drucker & Brannon, Reference Drucker and Brannon2014; honeybees: Giurfa, Marcout, Hilpert, Thevenot, & Rugani, Reference Giurfa, Marcout, Hilpert, Thevenot and Rugani2022).
Recently, we aimed to resolve these apparently conflicting findings (Eccher et al., Reference Eccher, Josserand, Caparos, Boissin, Buiatti, Piazza and Vallortigara2023). We conducted two behavioral experiments in three populations of different ages and cultural backgrounds: Italian adults, Italian preschoolers, and adults from the Himba tribe (an indigenous African tribe with no writing system). Our results showed that when tested with explicit tasks, only Italian adults show a consistent SNA, while when tested with implicit tasks, all the three populations exhibit a common and consistent left-to-right-oriented SNA.
These results support the hypothesis that the SNA phenomenon is dissociable into two different components: One which is acquired and cultural-dependent, and one which is biologically predisposed (note that the underlying mechanisms can be nonetheless quite different from those of a proper number line and rely instead on brain asymmetry, see, for a specific hypothesis, Vallortigara, Reference Vallortigara2018). But apart from the particular case in point, what seems interesting to me is that there is a special value in this sort of comparative research, namely the fact that evidence in non-human species forced us to reconsider our hypotheses on human nature.
The third insight concerns the proper way to build up intelligence in artificial systems. Plasticity seems to be a magic word nowadays in neuroscience and also in artificial intelligence (AI; but see Marcus, Reference Marcus2018). Comparative research on core knowledge tells us a different story. Consider what we learned from newly hatched chicks. These animals seem to be predisposed to orient toward objects that possess features associated with animate objects, such as biological motion, changes in speed, and face-like configurations (review in Di Giorgio et al., Reference Di Giorgio, Loveland, Mayer, Rosa-Salva, Versace and Vallortigara2017; Vallortigara, Reference Vallortigara2021 see also Vallortigara, Regolin, & Marconato, Reference Vallortigara, Regolin and Marconato2005). These are unlearned priors that help chicks to orient toward the mother hen and their siblings, thus facilitating and guiding a robust learning process called filial imprinting. (Similar mechanisms have been documented in human newborns [Lorenzi & Vallortigara, Reference Lorenzi, Vallortigara, Kaufman, Call and Kaufman2021; Vallortigara, Reference Vallortigara2012, Reference Vallortigara2021], even though in this species, proper control of past experience and access to neural substrates are limited for obvious reasons.)
The issue then is how can young organisms orient toward the “right” stimulus in the absence of any previous experience? In contrast to machine-learning systems, biological organisms do not require explicit reinforcement, supervised learning, or thousands/millions of examples to feed learning. They are equipped with dedicated orienting mechanisms that work as adaptive priors that imply some assumptions about the external world that guide learning (Versace, Martinho-Truswel, Kacelnik, & Vallortigara, Reference Versace, Martinho-Truswel, Kacelnik and Vallortigara2018). The priors are sufficiently general to allow errors. For instance, early preferences of chicks are not strictly species-specific but apply equally to hen face-like or polecat face-like features or to the biological-motion appearance of either a hen or a cat (review in Vallortigara, Reference Vallortigara2021). There is a profound biological reason for that. The predisposed orienting mechanisms cannot be too specific for the individual features, given that these are to some extent unpredictable from the genetic repertoire (because of variability between adults within a species and due to changes in the appearance of even a single individual).
However, high plasticity coupled with prior assumptions is not enough. In biological organisms, both early predispositions and high plasticity are transient phenomena that end either with some maturational processes or when the necessary information has been acquired. The existence of critical and sensitive periods has been documented in several domains and functions (Hensch & Quinlan, Reference Hensch and Quinlan2018). However, a very important recent finding is that critical periods do not apply only to the plasticity associated with learning but also to the periods of expression of the priors themselves. For instance, cues of animacy associated with speed changes are expressed in chicks during a restricted period in early life and can be reopened by the administration of certain substances (Lorenzi, Lemaire, Versace, Matsushima, & Vallortigara, Reference Lorenzi, Lemaire, Versace, Matsushima and Vallortigara2021).
There are costs associated with neural plasticity, and this is the reason why, after a certain age, learning new languages and solving amblyopia is so hard, and why early experiences are important for subsequent stages of life. Thus, the plasticity of the nervous system is actively reduced by molecular “brakes” that promote the stabilization of mature brain function (Hensch & Quinlan, Reference Hensch and Quinlan2018).
Elizabeth Spelke's book seems to suggest, among other things, that AI systems could benefit from being equipped with a set of priors that offer a guidance system and a way to speed up plasticity associated with learning mechanisms. I believe, however, that plasticity without critical periods of expression for these priors might have costs that prevent effective learning and cognitive functions.
Financial support
This work was supported by funding from the European Research Council under the European Union's Horizon 2020 research and innovation program (Grant Agreement 833504 SPANUMBRA).
Competing interest
None.