In the target article, Hoerl & McCormack (H&M) propose a dual-systems account for temporal cognition and argue that non-human animals can only use the temporal updating system whereas humans utilize the temporal reasoning system as well. I am sympathetic to dual-systems approaches in general, but it seems that H&M try to explain too much with too little in animal cognition.
In H&M's view, the model of the current world does all the heavy lifting in animal cognition. The model represents how things are in the current environment; contains information about objects, their features, and locations including goal states; supports single-trial and sequential learning; and can be updated as things change in the environment. Yet this notion of a model is left unpacked. If the model is just for the present state of the world, why not use the world as its own model? Is the model just a cognitive map or a full-blown replica of the world? What is the format in which the information is stored? Is it perceptual or propositional or something else? Is the model constructed componentially? If so, what binds different kinds of representations into a single unit? If not, how could the model be updated so swiftly? Without answers to these kinds of questions it is difficult to assess whether non-human animals operate with a model or not.
Arguably, any attempt to specify what a model of the current world consists of has to include perceptual elements that go beyond an animal's immediate sensory range. H&M seem to agree with this, as they accept that non-human animals can continue to represent an object that they no longer perceive as part of their current environment. These memory-like representations are obtained from the animals’ past experiences and are presumably retained insofar as they are useful for the organism. H&M want to subsume these memory-like representations within an animal's model of the world, but it is possible to be more precise here.
In earlier work (Keven Reference Keven2016; Reference Keven2018), I called these types of representations event memories and argued that we can understand the mnemonic abilities of non-human animals (and young children) with event memory without ascribing them a capacity for full-blown episodic memory. According to the dual-systems thesis that I proposed, event memory is a snapshot-like memory system predominantly in the form of visual images, whereas episodic memory requires additional higher-order inferential processes. The episodic memory system takes event memories as inputs and binds them into a whole by linking multiple events into a temporal sequence, establishing casual relations between temporally separated events and arranging events in a converging structure such that multiple events are bound together to enable an outcome. Unlike episodic memory, temporal, causal, and teleological relationships between events are not specified in event memory. Event memories are fleeting and fragmentary in this sense as they are not bound into a stable whole. Hence, event memories are retained as long as they are relevant for current tasks, otherwise they are rapidly forgotten.
H&M claim that such free-floating representations cannot systematically guide action. Although event memories are not bound into a stable whole, they are still tied to the current goals of the organism and can be activated by task-based cues from working memory. In this respect, event memories differ from Redshaw's (Reference Redshaw2014) uncontextualized representations, as the goals of the organism actually relate event memories to the current context. This is a different kind of contextualizing than what Redshaw seems to have in mind, as it still does not require meta-representational abilities. Instead, the current goals of the organism activate relevant representations that are associated with achieving that goal (Hommel Reference Hommel2009; Hommel et al. Reference Hommel, Müsseler, Aschersleben and Prinz2001). The idea is that when an organism is engaged in a task, task-relevant representations, such as recent events, locations, and other relevant perceptual or semantic information, are activated. If the task is time sensitive, this process could also incorporate temporal information from an interval timer mechanism similar to what H&M envisions. These activated representations can then guide the selection of actions according to their expected outcomes.
To illustrate how this process might work, consider Clayton and Dickinson's (Reference Clayton and Dickinson1998) original study that H&M discuss. Event memory can assist scrub jays by keeping track of caching events (i.e., what did the bird cache where). Because the recovery task is time sensitive, the birds could also use an interval timer mechanism to control how long these event memories would remain task relevant. During the training phase of the study, scrub jays seem to learn that worm-caching events are relevant for the recovery task only for a short time period and there is no need to retain them for longer. In 124-hour trials, then, the birds could actually be operating with only the event memory of caching peanuts, and hence they search for peanuts. In 4-hour trials, however, because the elapsed time is short, event memories for caching peanuts and caching worms would both still be active. In this case, the birds search for worms as their preferred food.
It is important to note that none of these processes require remembering the actual experience of caching the food items, unlike an interpretation based on mental time travel (Salwiczek et al. Reference Salwiczek, Watanabe and Clayton2010). The birds could remember in the same way I can remember where my keys are without remembering the actual experience of where I put them (Malanowski Reference Malanowski2016; Suddendorf & Busby Reference Suddendorf and Busby2003). Event memory is based on automatic perceptual processes and does not require conscious attention at encoding or retrieval.
To conclude, animals need to keep track of what has happened to effectively deal with day-to-day tasks that are extended over time. Event memory can guide animals by providing a record of progress in such tasks.
In the target article, Hoerl & McCormack (H&M) propose a dual-systems account for temporal cognition and argue that non-human animals can only use the temporal updating system whereas humans utilize the temporal reasoning system as well. I am sympathetic to dual-systems approaches in general, but it seems that H&M try to explain too much with too little in animal cognition.
In H&M's view, the model of the current world does all the heavy lifting in animal cognition. The model represents how things are in the current environment; contains information about objects, their features, and locations including goal states; supports single-trial and sequential learning; and can be updated as things change in the environment. Yet this notion of a model is left unpacked. If the model is just for the present state of the world, why not use the world as its own model? Is the model just a cognitive map or a full-blown replica of the world? What is the format in which the information is stored? Is it perceptual or propositional or something else? Is the model constructed componentially? If so, what binds different kinds of representations into a single unit? If not, how could the model be updated so swiftly? Without answers to these kinds of questions it is difficult to assess whether non-human animals operate with a model or not.
Arguably, any attempt to specify what a model of the current world consists of has to include perceptual elements that go beyond an animal's immediate sensory range. H&M seem to agree with this, as they accept that non-human animals can continue to represent an object that they no longer perceive as part of their current environment. These memory-like representations are obtained from the animals’ past experiences and are presumably retained insofar as they are useful for the organism. H&M want to subsume these memory-like representations within an animal's model of the world, but it is possible to be more precise here.
In earlier work (Keven Reference Keven2016; Reference Keven2018), I called these types of representations event memories and argued that we can understand the mnemonic abilities of non-human animals (and young children) with event memory without ascribing them a capacity for full-blown episodic memory. According to the dual-systems thesis that I proposed, event memory is a snapshot-like memory system predominantly in the form of visual images, whereas episodic memory requires additional higher-order inferential processes. The episodic memory system takes event memories as inputs and binds them into a whole by linking multiple events into a temporal sequence, establishing casual relations between temporally separated events and arranging events in a converging structure such that multiple events are bound together to enable an outcome. Unlike episodic memory, temporal, causal, and teleological relationships between events are not specified in event memory. Event memories are fleeting and fragmentary in this sense as they are not bound into a stable whole. Hence, event memories are retained as long as they are relevant for current tasks, otherwise they are rapidly forgotten.
H&M claim that such free-floating representations cannot systematically guide action. Although event memories are not bound into a stable whole, they are still tied to the current goals of the organism and can be activated by task-based cues from working memory. In this respect, event memories differ from Redshaw's (Reference Redshaw2014) uncontextualized representations, as the goals of the organism actually relate event memories to the current context. This is a different kind of contextualizing than what Redshaw seems to have in mind, as it still does not require meta-representational abilities. Instead, the current goals of the organism activate relevant representations that are associated with achieving that goal (Hommel Reference Hommel2009; Hommel et al. Reference Hommel, Müsseler, Aschersleben and Prinz2001). The idea is that when an organism is engaged in a task, task-relevant representations, such as recent events, locations, and other relevant perceptual or semantic information, are activated. If the task is time sensitive, this process could also incorporate temporal information from an interval timer mechanism similar to what H&M envisions. These activated representations can then guide the selection of actions according to their expected outcomes.
To illustrate how this process might work, consider Clayton and Dickinson's (Reference Clayton and Dickinson1998) original study that H&M discuss. Event memory can assist scrub jays by keeping track of caching events (i.e., what did the bird cache where). Because the recovery task is time sensitive, the birds could also use an interval timer mechanism to control how long these event memories would remain task relevant. During the training phase of the study, scrub jays seem to learn that worm-caching events are relevant for the recovery task only for a short time period and there is no need to retain them for longer. In 124-hour trials, then, the birds could actually be operating with only the event memory of caching peanuts, and hence they search for peanuts. In 4-hour trials, however, because the elapsed time is short, event memories for caching peanuts and caching worms would both still be active. In this case, the birds search for worms as their preferred food.
It is important to note that none of these processes require remembering the actual experience of caching the food items, unlike an interpretation based on mental time travel (Salwiczek et al. Reference Salwiczek, Watanabe and Clayton2010). The birds could remember in the same way I can remember where my keys are without remembering the actual experience of where I put them (Malanowski Reference Malanowski2016; Suddendorf & Busby Reference Suddendorf and Busby2003). Event memory is based on automatic perceptual processes and does not require conscious attention at encoding or retrieval.
To conclude, animals need to keep track of what has happened to effectively deal with day-to-day tasks that are extended over time. Event memory can guide animals by providing a record of progress in such tasks.