Do rats feel regret? If so, how is it represented or experienced in the brain? Recent work (Steiner & Redish, 2014) suggests that rats do indeed feel regret, and has even proposed a mechanism for how rats – and by extension, people – feel regret.
On Regret and Regretting
Regret is a fundamental part of the human condition. Many classic works of literature highlight the pain and torment of regret: in Macbeth, the titular character is driven mad by regret for his crimes (“Will all great Neptune’s ocean wash this blood clean from my hand? No, this my hand will rather the multitudinous seas incarnadine, making the green one red”, Act II, Scene II); in Oedipus Rex, Oedipus gouges out his own eyes because of his mistakes; and in Rime of the Ancient Mariner, the mariner comes to deeply regret his decision to kill the albatross (“The very deep did rot: O Christ // that ever this should be”).
Importantly, regret is distinct from the emotion of disappointment. When you’re disappointed in how something turned out, it means you wish something else had happened. Regret means that you realize this worse-than-expected outcome is at least partially a result of your own mistake – thus, you wish that you had done something differently. Both emotions require counterfactual reasoning (Narayanan, 2009), but regret places the self at the center of the counterfactual.
Narayanan (2009) and others (de Vega et al, 2007; de Vega and Urrutia, 2011; de Vega & Urrutia, 2012) suggest that one mechanism for considering alternative actions is via simulation – recruiting neural circuits that are involved in perceiving or performing the action (Bergen, 2012). Most of the research supporting this theory of counterfactual understanding, however, was behavioral in nature, so it was difficult to identify the actual neural mechanisms involved.
More recently, researchers at the University of Minnesota (Steiner & Redish, 2014) investigated whether rats can feel regret about their decisions, and what neural mechanisms underpin the feeling of regret.
Do rats feel regret?
To test whether rats experience regret, Steiner and Redish (2014) developed a neuroeconomic task called Restaurant Row. In this task, the rat ran along a looped track; different zones of the track have different “vendors” supplying various food flavors (banana, cherry, etc.), which the rat could sample after waiting for some period. Each zone also had a different delay period, the timing of which was announced to the rat via a high-pitched tone of varying frequencies. The task was economic in nature, because the decision to wait longer in one zone inherently prevented the rat from sampling food in another zone; thus, the rat had to make decisions to maximize his food intake. Here, the “cost” of the reward was the amount of time the rat had to wait.
A “regret-inducing situation” was one in which the rat encountered a “high-cost” zone – one with a waiting time higher than the rat’s preferred delay threshold – just after skipping a “low-cost” zone (e.g. a cheap, low-delay zone). If a rat finds itself in these situations, it has made an economic mistake, and should feel regret.
As a side note, we can probably all imagine similar situations from our own lives. Perhaps we passed up on a parking spot expecting to come across a better one, but found only slim pickings; or perhaps, like the rat, we opted not to go to the cheaper, lower-quality diner with no wait, and were forced to wait half an hour to be seated at the more expensive gourmet restaurant down the street.
The researchers compared both behavioral and neurophysiological data from “regret-inducing” situations to control situations. In terms of behavior, “regretful rats” were more likely to look back towards the previous, low-cost option – as if overtaken by a wistful contemplation of what could have been.
The neurophysiological analysis is a little more complicated. During the task, the authors recorded activity from neurons in two regions of the brain – the orbitofrontal cortex (OFC) and the ventral striatum (vStr). They then used something called a Bayesian decoding algorithm to try to differentiate between the activity of neurons based on the different rewards and zones in the task. From this, the researchers can estimate p(Reward) and p(Zone) – the probability that the rat is encountering a reward or particular zone, given the neural activity.
Crucially, in regret-inducing situations, the researchers found a strong representation of p(Zone), where “Zone” refers to the previous low-cost zone that the rat had chosen to skip. In other words, the rat is thinking about the zone it just passed over. The representation of the previous zone was much stronger than the representation of the missed reward, which the authors take as evidence for the rat considering alternative actions it could have taken (e.g. regret instead of mere disappointment). The rat isn’t simply unhappy that it missed the low-cost reward; it’s thinking about what it could have done differently in the past.
People make mistakes. There’s nothing quite like the sinking feeling of realizing we’ve done something wrong – that a past decision is “coming back to haunt us”. Those of us who tend towards rumination will often chew over these past mistakes again and again, staring at the ceiling as we try to sleep.
Being a human is different from being a rat, of course. But rats are smarter than we give them credit for – and, as it appears, even rats are not immune from the crushing weight of regret. There’s something a little tragic about the predicament these lab rats found themselves in: their experience revolved around a series of alternatingly low-/high-cost options, which were specifically designed to induce this crushing feeling. It certainly gives new meaning to the term “rat race”.
Importantly, Steiner and Redish (2014) show that this feeling of our past actions “coming back to haunt us” is, in a way, physically instantiated in the brain: when we ruminate on a poor decision that led to an undesirable outcome, we are simulating what it was like to make that decision, and perhaps imagining what we could have done instead. Perhaps, then, we are doomed to relive our mistakes again and again – if not in actuality, then in the simulations of our brain.
Narayanan, S. (2009). Mind changes : A simulation semantics account of counterfactuals Counterfactual reasoning. Draft, 1–44.
Bergen, B. K. (2012). Louder than words: The new science of how the mind makes meaning. Basic Books.
Steiner, Adam; Redish, D. (2014). Behavioral and neurophysiological correlates of regret in rat decision-making on a neuroeconomic task. Nature Neuroscience, 17(7), 995–1002. http://doi.org/10.1038/nbt.3121.ChIP-nexus
de Vega, M., & Urrutia, M. (2012). Discourse Updating after reading a counterfactual event. Psicologica.
de Vega, M., Urrutia, M., Riffo, B., Laguna, L., & Islands, C. (2007). Canceling updating in the comprehension of counterfactuals embedded in narratives. Memory & Cognition, 35(6), 1410–21. http://doi.org/10.3758/BF03193611
de Vega, M., & Urrutia, M. (2011). Counterfactual sentences activate embodied meaning : An action – sentence compatibility effect study Counterfactual sentences activate embodied meaning : An action Á sentence compatibility effect study. Journal of Cognitive Psychology, 5911(September). http://doi.org/10.1080/20445911.2011.590471
Georgopoulos, A. P., Caminiti, R., Kalaska, J. F., & Massey, J. T. (1983). Spatial coding of movement: a hypothesis concerning the coding of movement direction by motor cortical populations. Exp Brain Res Suppl, 7(32), 336.
McDannald MA, Lucantonio F, Burke KA, Niv Y, Schoenbaum G. Ventral striatum and orbitofrontal cortex are both required for model-based, but not model-free, reinforcement learning. J Neurosci. 2011; 31:2700–2705. [PubMed: 21325538]
Camille N, et al. The involvement of the orbitofrontal cortex in the experience of regret. Science. 2004; 304:1167–1170. [PubMed: 15155951]
 This is, obviously, a very small sampling of the relevant literary works. Many tragedies deal with the problem of regret, since a tragedy generally involves (by definition) catastrophe due at least in part to a mistake or character flaw of the protagonist.
 Counterfactuals are often expressed in language through the pluperfect subjunctive tense, such as: if it had rained today, practice would have been cancelled. To understand this statement, you need to consider an alternative “world state”, in which it had rained (rather than not raining), and infer a causal link between the raining and the cancelation of practice. Narayanan (2009) points out that this sort of reasoning is implicit in many human emotions, including: regret, disappointment, blame, hope, anxiety, and more.
 The logic is that the rat should wait for valuable offers, but skip “expensive” offers.
 As mentioned in the introduction, regret is different from disappointment, so it was important to control for situations that were merely disappointing – e.g. ones in which a worse-than-desired outcome occurred, but it was not the rat’s fault. In one control condition, the rat took the initial low-cost offer, and then encountered the same high-cost offer. This is disappointing but should not induce regret, since the rat took the earlier opportunity. In the other control condition, the rat encountered two high-cost zones in a row. The second high-cost zone should, again, be disappointing, but should not induce regret – “because the rat’s actions were consistent with its revealed preferences” (pg. 6).
 Notably, humans with damage to their OFC experience diminished (or no) regret (Camille et al, 2004).
 The ventral striatum also seems to be involved in deciding the value of various outcomes, such as after making a decision (MacDannald et al, 2011).
 In general, the idea behind neural decoding models is to try to predict input from the neural activity it causes. For example, using activity from neurons in a region of the motor cortex, you can predict which direction a monkey’s arm is moving (Georgopoulos et al, 1983). If you can reliably predict something about the external environment – purely from information about which neurons are firing, when they are firing, and to what extent they are firing – that suggests this population of neurons plays an important mechanistic role in either representing some piece of information or coordinating some sort of action.
 Technically, this is p(Zone | spikes) and p(Reward | spikes) – the conditional probability – but as in the paper, I’ve reduced it for simplicity.