The capacity to entertain meta-representations has long been central to theorising about mental time travel (Suddendorf & Corballis Reference Suddendorf and Corballis1997). In its original and most narrow sense, meta-representation is defined as representing the representational relation between a representation and its referent (Pylyshyn Reference Pylyshyn1978). Hoerl & McCormack (H&M) argue that this capacity may not necessarily be implicated in grasping temporal relations between different models of reality from the same time line (sect. 3.2, para. 9) – such that tonight's dinner will follow the breakfast you are eating now. Still, we contend that meta-representations play a fundamental role in grasping relations between different models of reality from alternative time lines.

H&M allude to alternative time lines when they write about “bifurcation points in linear time” (sect. 5, para. 8), where multiple possible outcomes of an event diverge (e.g., a coin flip). Yet H&M's model includes no mechanism for how such bifurcation points could be understood. We suggest that the capacity to form meta-representations is critical. A meta-representational agent can understand that representations of events are just representations and that they do not always map onto referent events in a one-to-one fashion (Perner Reference Perner1991). With this understanding, the agent can infer that a single representation of an event outcome only maps onto one of all possible outcomes for that event (Redshaw Reference Redshaw2014). The agent can also tag such representations with evaluations, such as how likely they are to transpire, whether they are best or worst case scenarios, and whether they may be exaggerations (Suddendorf Reference Suddendorf, Corballis and Lea1999; Suddendorf & Corballis Reference Suddendorf and Corballis1997; Reference Suddendorf and Corballis2007b).

The capacity to form meta-representations can explain why humans not only envision and prepare for likely and desirable future outcomes, but also implement contingency plans (Redshaw & Suddendorf Reference Redshaw and Suddendorf2016). When planning an overseas holiday, for example, we might book flights and hotels, but we typically also buy travel insurance just in case things take a turn for the worse. Meta-representations may also underlie certain cases of delayed gratification that are not easily explained by the two mechanisms H&M invoke. When beginning a diet, for example, we may imagine a future in which we eat only healthy food. But when we meta-represent that this is just one possible future, we can make it more likely to manifest by removing temptations such as the junk food in our pantry. To generalise, meta-representations enable at least two pervasive types of future-oriented human behaviour: (1) preparing for multiple possibilities to compensate for the fact that we do not know exactly what will happen, and (2) structuring our current environment in such a way so as to increase the likelihood that desirable versions of the future will transpire. In Bulley et al. (Reference Bulley, Redshaw, Suddendorf and Abrahamin press), we introduce the term metaforesight to describe the cognitive processes that drive such behaviour.

The recursive capacity to embed representations within other representations also allows humans to entertain hierarchically nested levels of temporal reasoning. To illustrate: At the first level, we recognise that a future event can have multiple possible outcomes. At the second level, we might reflect on the fact that a past event {level 2} once had such multiple possible future outcomes {level 1}, which is central to the emotional experiences of regret and relief (Hoerl & McCormack Reference Hoerl, McCormack, Michaelian, Klein and Szpunar2016). At the third level, we might imagine a future situation {level 3} in which we will reflect on the fact that a past event {level 2} once did have multiple possible future outcomes {level 1}. This understanding is central to emotional experiences like anticipated regret, which can drive actions such as putting away that second bottle of wine to prevent a regret-filled hangover tomorrow. Recursive operations are in principle unbounded, and so humans may be limited primarily by working memory constraints on the number of levels of temporal reasoning they can entertain. Accordingly, alongside increases in working memory capacity (Alloway et al. Reference Alloway, Gathercole and Pickering2006), children become able to prepare for multiple future possibilities (Beck et al. Reference Beck, Robinson, Carroll and Apperly2006; Redshaw et al. Reference Redshaw, Suddendorf, Neldner, Wilks, Tomaselli, Mushin and Nielsen2019) before they appear to experience regret (O'Connor et al. Reference O'Connor, McCormack and Feeney2012), which in turn appears to develop before they can anticipate regret (McCormack & Feeney Reference McCormack and Feeney2015).

Notably, these nested levels of temporal reasoning are not restricted to situations where future and past perspectives are both represented. Rather, we can recursively imagine future situations in which the future will be imagined, such as when we worry that we will keep worrying (Wells Reference Wells2005). So too, can we reflect on past occasions in which the past was remembered.

Non-human animals’ foresight may be restricted by deficits in various components contributing to mental time travel (Suddendorf Reference Suddendorf2013; Suddendorf & Corballis Reference Suddendorf and Corballis2007b). If they cannot form meta-representations (Carruthers Reference Carruthers2014; Suddendorf Reference Suddendorf, Corballis and Lea1999), then they might find it difficult to prepare for multiple possible versions of uncertain future events (Redshaw Reference Redshaw2014). Preliminary evidence is consistent with this prediction, with studies failing to find any evidence that non-human great apes can spontaneously and consistently prepare for mutually exclusive outcomes of even a very simple, immediate future event (Redshaw & Suddendorf Reference Redshaw and Suddendorf2016; Suddendorf et al. Reference Suddendorf, Crimston and Redshaw2017). In one recent study (Lambert & Osvath Reference Lambert and Osvath2018), chimpanzees were more than seven times more likely to prepare for two certain future outcomes than for two uncertain future possibilities – even though the optimal response across conditions was exactly the same (see Table 1 of that article). Performance in both conditions was relatively poor, however, and so further work with other paradigms is warranted.

Evidence for other forms of complex prospection in animals also remains controversial (Suddendorf et al. Reference Suddendorf, Bulley and Miloyan2018; Suddendorf & Redshaw, Reference Suddendorf, Redshaw, Vonk and Shacklefordin press). Although some high-profile studies have claimed that great apes (Mulcahy & Call Reference Mulcahy and Call2006) and corvids (Kabadayi & Osvath Reference Kabadayi and Osvath2017) can select objects with the intention of using them to obtain rewards in future situations, these have been criticised for failing to control for low-level explanations such as associative learning (Hampton Reference Hampton2018; Lind Reference Lind2018; Redshaw et al. Reference Redshaw, Taylor and Suddendorf2017; Suddendorf & Corballis Reference Suddendorf and Corballis2008). H&M state that criticising such studies on individual “ad hoc” bases may be unconvincing, but we maintain it is important for progress to clarify in each case why the conclusions are not compelling and what kind of controls could provide stronger evidence.