The transition from water to land 380 million years ago transformed vertebrate cognition. Eye sizes tripled, and combined with air’s transparency compared to water, monitored visual space increased by a million-fold (MacIver, Schmitz, Mugan, Murphey, & Mobley, Reference MacIver, Schmitz, Mugan, Murphey and Mobley2017). Even land animals in darkness see significantly further than shallow-water fish in full sunlight (MacIver & Mugan, Reference Mugan and MacIver2020). This “sensory horizon shock” meant that land animals had exponentially more possible futures. Success would belong to those who could learn or plan successful routes through this thicket of options.

In mammals and convergently in birds (Striedter, 2015), the hippocampus emerged as a key brain structure for managing this complexity, functioning simultaneously as a memory, spatial mapping, learning, imagination (Comrie, Frank, & Kay, Reference Comrie, Frank and Kay2022), and planning system (Bakermans, Warren, Whittington, & Behrens, Reference Bakermans, Warren, Whittington and Behrens2025). Its abilities may explain why some vertebrates learn dramatically faster than machine learning systems.

The hippocampus specializes in conjunctive bindings between locations and their associated elements (Eichenbaum Reference Eichenbaum2004, Komorowski, Manns, & Eichenbaum Reference Komorowski, Manns and Eichenbaum2009, Bakermans et al., Reference Bakermans, Warren, Whittington and Behrens2025). For example, healthy individuals can richly imagine common scenes when prompted, while those with hippocampal damage produce impoverished descriptions (Hassabis, Kumaran, Vann, & Maguire, Reference Hassabis, Kumaran, Vann and Maguire2007).

We have designed a rodent-robot interaction arena with naturalistic spatial complexity to test ideas on the role of terrestrialization in the genesis of cognition (Lai et al., Reference Lai, Espinosa, Wink, Angeloni, Dombeck and MacIver2024). A focal point is “non-local activity” in the hippocampus – signals indicating the brain is representing locations away from the current position. We can infer this activity and watch as the brain examines what may be the animal’s possible futures. When mice see approaching predator-like robots, neural signals in the hippocampus appear to reveal imagined trajectories tracing away from their current location. Mice then often move along one of these imagined routes (Pfeiffer & Foster, Reference Pfeiffer and Foster2013). The sequences occur at 10-20x real time, perhaps underlying unconscious planning or consolidation.

This capacity for imagination creates a fundamental challenge: distinguishing it from reality. Fleming and Michel suggest that the brain requires a “perceptual reality monitor” that uses confidence indicators like signal strength to tag external input as reality.

This segregation is essential for survival. If an animal that is mentally rehearsing an attack became confused and enacted an imagined trajectory, it would immediately reveal its position; similarly, if a dimly lit attack is underway, the victim might think it is imagined rather than actual.

The challenge becomes apparent in everyday experience: when drifting to sleep, distant voices can easily be confused with dream onset (which likely involves similar hippocampal activations). As we become conscious and notice that the voices are continuing, we understand these voices are real rather than from a dream.

This theory opens experimental avenues, including testing postdictive effects in non-mammalian systems and examining the balance between internal and external processing via consciousness-altering pharmaceuticals like ketamine. For instance, if an animal learns an internal model of a threat and then becomes biased toward internal content through an injection of ketamine while the threat’s behavior changes, the theory predicts the animal will perform actions based on the outdated internal model. Conversely, if the animal becomes biased toward external content, perhaps by reducing sensory horizon, unconscious high-speed responses should begin to overwhelm the internal model due to lack of time for consciousness (Keramati, Smittenaar, Dolan, & Dayan Reference Keramati, Smittenaar, Dolan and Dayan2016).

From Aquatic Simplicity to Terrestrial Complexity . The journey from simple aquatic reflexes to complex terrestrial cognition illustrates evolution’s response to environmental challenges. Fish exhibit straightforward escape responses when visual stimuli exceed certain thresholds – a solid black circle growing larger triggers escape when it reaches a specific size on the retina (Bhattacharyya, McLean, & MacIver, Reference Bhattacharyya, McLean and MacIver2017).

While mammals also have reactive response systems (Yilmaz & Meister, Reference Yilmaz and Meister2013), the hippocampus represents a fundamentally different approach that is well attuned to the combinatorial explosion of terrestrial possibilities. It enables proactive simulation of multiple possibilities through patterns of non-local activity, arousing all the bindings relevant to guiding the behavior in the current scene. This virtual learning capacity provides enormous advantages, especially in high-stakes predator-prey situations where premature enactment might prove fatal, and there is no previous experience similar enough to cue recall and reuse.

Being able to evoke possible scenes allows animals to estimate all the typical parameters of occurrent reality – whether something feels good, bad, or dangerous – without experiencing the situation. This represents a revolutionary cognitive advancement: the ability to learn from imagined rather than lived experience.

There is a gap, however, as Fleming and Michel mention. Reality monitoring tags real versus imagined, but how do imagined consequences become behaviorally potent, particularly for effortful deliberative planning? Is our awareness of this, like vivid imagery, a failure of the reality monitor? If consciousness is present-oriented, then technologies (Pichaske et al., Reference Pichaske, MacIver, Gratz, Ryan, Nagorka and Pardo2017) or cultural scaffolding (Cerf, Matz, & MacIver, Reference Cerf, Matz and MacIver2023) may be needed to reliably “import” distant consequences into the now. But perhaps we need to understand how consciousness is for content that is stable and structured enough for representation (classic reality monitor function but including constructs of conscious planning) and poised to guide current or future behavior where the content draws on, or must be disambiguated across, representational sources such as sensory input versus imagination versus memory. Whenever we need a metacognitive understanding that this is from perception, that is from imagination, and this is from memory, then it is conscious.

After decades of slow progress in consciousness science, Fleming and Michel have provided a concrete proposal with promising experimental potential across human and non-human model systems. Their framework suggests that consciousness emerged not as a single breakthrough but as a solution to the specific challenge of managing multiple competing representations of reality.

The transition from water to land thus represents more than anatomical adaptation – it marks the onset of evolutionary pressures that drove the development of imagination, planning, and possibly consciousness itself. Gaining a mechanistic understanding of consciousness will help us understand the emergence of cognition through deep time. In a more applied sense, it could lead to new forms of artificial intelligence that learn with the efficiency we observe in conscious mammals.