Discussion and conclusion

The ability to train volunteer dogs (who have the opportunity to opt out) to lie still while undergoing fMRI scans has recently opened up new avenues of research into canine cognition and greatly augmented what we can learn from ethological studies in this field. As reviewed here, fMRI studies of the canine brain have revealed positive and consistent responses in the caudate nucleus, specifically the nucleus accumbens, to objects and stimuli, including human individuals, that dogs liked. Positive emotions are invariably linked to caudate activations; specific parts of the caudate stand out for their consistent activation to many things that humans enjoy, and indeed caudate activation is so consistent that, under the right circumstances, it can predict our preferences for food, music and even beauty.17 Canine responses were of a manner and magnitude similar to caudate responses in humans—a functional homology that may be indicative of dogs experiencing similar emotions to humans.18 Further fMRI investigations, combined with behavioural studies, have demonstrated the sensitivity of dogs to human social interaction/bonding, and that dogs may prefer the company of familiar humans over the company of either familiar or unfamiliar dogs,32 33 and that dogs may have evolved mechanisms especially tuned to social cues and therefore may have specialised neural machinery for face processing.33–35 Dogs seem able to understand a significant amount of human language, including both verbal and emotional content.

In other words, there is formidable and mounting evidence illustrating and underlining the complex cognitive and emotional capacities of dogs. It is important to note that such evidence has, however, existed for some time from ethological studies, and indeed many would argue it has simply been obvious from the most cursory of observations to those who have interacted with dogs that they (and, in fact, many other non-human species) are intelligent, emotional and highly sentient beings. Such ethological investigations have demonstrated, and increasingly illustrate, the depth and breadth of their cognitive and emotional capacities. A recent review (2016) cites observational ethology and problem-solving experiments in the early 1970s and 1980s as being part of a ‘rediscovery’ of canine cognition investigations and comparative studies.36 It cites numerous supporting studies that have increased our understanding of cognition in dogs, for example via tracking of eye movements (elucidating interest, preference, planning, intent, emotional perception and processing), touch-sensitive screens (recognition, categorisation, communication) and others, as well as non-invasive studies of canine neurobiology including electroencephalography (EEG) and fMRI. fMRI data have not, therefore, ‘reinvented the wheel’, but they have, importantly, expanded the methods by which scientists can access the minds of other animals, and this is an important move. Arguably, it should not really be necessary to prove to any greater degree what seems clear, when considering the suffering experienced by dogs undergoing invasive experiments, although the fact that this may be an inconvenient truth to a scientific establishment that uses many tens of thousands of dogs every year as research subjects may be a factor. Nevertheless, it seems—belatedly perhaps—that the deep, advanced, and in some ways ‘human-like’ nature of canine cognition and emotions are finally being recognised by the scientific community, given the degree and type of research being funded and conducted. fMRI data are cementing this, by showing us the mechanisms behind these qualities, which parts of the canine brain are working to achieve them and how. It means that dogs must, and do, suffer much more than is accepted currently within part of the scientific community, as evidenced by the type of dog research being conducted, the number of dogs used and the suffering associated with some procedures (see Introduction). This has profound and serious consequences for the use of dogs in harmful experiments and toxicity testing.

One omnipresent caveat of any discussion of fMRI data, despite more than quarter of a century of human fMRI studies and use, is their reliability, and the question of what fMRI images specifically represent in terms of neural activity (see ref 37 for a discussion). While the exact basis and mechanistic nature of fMRI images are somewhat controversial and remain under scrutiny even now, it is generally accepted, based on much evidence, that the fMRI signal—based as it is on blood flow and oxygen levels (‘blood oxygen-level dependent [BOLD] signals’)—accurately represents neural activity. Another potential criticism is that the fMRI images central to the argument we present in this paper are simply evidencing neural activity associated with various stimuli and behaviours, as opposed to being correlates of conscious experiences. This type of argument is not novel, of course—even for humans—and it is accepted that the evidence herein and its like cannot prove conscious experiences. Nonetheless, there is increasing and convincing evidence from human studies (of course, humans can follow detailed and complex orders, and report verbally to researchers) that neuroimaging data strongly correlate to various mental states. An in-depth review published in 2015 describes how neuroimaging technologies such as fMRI and EEG can reveal detailed aspects of, for example, decisions, intentions, thoughts, imagining tasks, behavioural will and others in humans.38 For example, in humans, fMRI signatures (‘brain activation patterns’) correctly identified intentions to perform specific tasks in 80% of instances,39 revealed movement intentions before their initiation40 and identified the type and duration of task being imagined with a mean accuracy of 95%.41 Further support for the assertion of conscious experience versus stimuli/behaviour-associated neural activity has been provided by human research showing that imagining performing tasks generates highly similar patterns of brain activity to actually performing them, exemplified by imagining limb movement,42 spatial navigation43 and playing tennis.44 In other words, fMRI patterns can be used as neural proxies for behaviour. Further, the evidence above is augmented and fortified by many studies that have been conducted over the past two decades,45 46 comparing, for instance, imaging patterns in (1) healthy individuals and in patients assumed to be in a vegetative state or minimally conscious, and (2) in similar comparative experiments in humans who are awake and asleep. The former have demonstrated that many forms of awareness and cognition exist in some people in vegetative states, in whom specific neuroimaging patterns are indistinguishable from healthy individuals, and which are therefore indicative of actual conscious experiences, even to the degree that people assumed to be in a vegetative state can communicate, understand and respond to spoken commands and questions (eg, ref 47 and see ref 38). For example, persons fulfilling all agreed criteria for being in a vegetative state or minimally conscious show the same fMRI patterns when instructed to perform mental imaging tasks, such as the tennis playing and spatial navigation just described.48 The latter, comparing awake and asleep individuals, have shown that diverse mental states, including discrete emotional states and experiences, have associated ‘brain states’ in the form of ‘coherent, emotion-specific patterns’,49–51 supporting the theory that ‘…posit emotions are represented categorically in the coordinated activity of separable neural substrates’.52 A recent (2016) review53 cites work demonstrating in humans that valenced brain states can be differentiated on the basis of neural activity (eg, refs 54 55), including an experiment showing that nine different emotions could be predicted by fMRI activity, with an accuracy of 84% with the same subject and 70% with different subjects.56 The same review reported the authors’ own work, showing that fMRI could classify six different emotions with 37.3% accuracy, compared with chance levels of 14.3%.53 57 Meta-analyses corroborate and strengthen this. Hamann11 summarised a number of these in his review.11 For example, the five basic emotions of happiness, sadness, anger, fear and disgust are ‘characterised by consistent neural correlates’ and can be reliably discriminated from each other,58 and ‘consistent regional brain activations’ correspond to each basic emotion category.59 More recently, Wager et al60 analysed almost 150 studies involving well over 2000 participants, and concluded similarly that the five basic emotion categories are identifiable via patterns of brain activation, revealed by fMRI, across multiple brain systems such as ‘the cortex, thalamus, amygdala, and other structures’.

The use of dogs in invasive and/or harmful experiments, as for other non-human species, currently rests on the aforementioned HBA (see Introduction)—that harms to the animals used must be minimised, and the benefits to others (often humans) from those experiments must be substantial. It can be argued that HBA analyses are not currently conducted appropriately. Inter alia, they are not conducted transparently; are often, if not always, rudimentary; are performed by individuals who have heavily invested in animal research; and the failure of animal experiments to translate to human benefits is not fully considered.10 The harms—as mentioned in the Introduction—we argue are considerable, yet at the same time underappreciated and underacknowledged, especially so in light of the fMRI information reviewed herein.4 These issues have been of sufficient importance to stimulate research into the living conditions and suffering of dogs in laboratories, as well as changes in these conditions to improve the welfare of these dogs. Significantly, however, attempts to ameliorate laboratory conditions for dogs (and also therefore to improve data from dog experiments) by providing social housing may still be inadequate: some dogs still become stressed, anguished, frustrated and bored, and it appears that at the very least more human social contact may be a sine qua non.61 Evidence overall, therefore, reveals that harms from experimentation to the dogs involved may be much greater, widespread and intractable than previously accepted, skewing the HBA balance and demanding any benefits be much greater to redress it. Yet it increasingly appears that this HBA ratio is skewed even further by the actual value of dog experiments to humans (see, eg, refs 1–3), which, as exemplified in the Introduction, we believe is much less than is claimed by advocates of dog research.

In conclusion, in accordance with our intuitions, simple observation and experience, ethological/behavioural research and fMRI experiments, dogs are able to morally evaluate; don’t just ‘hear’ and have a basic understanding of some words, but also ‘listen’ to human intonation to perceive emotional states and intent; understand to some degree human ‘communicative intent’ and perspective, involving to a large extent the eyes; possess neural machinery for face processing, enabling identification and perception of emotions; and so on. Perhaps most importantly and crucially, dogs are able to experience positive emotions, empathic-like responses and demonstrate human bonding at least comparable with human children, via similar neural mechanisms: ‘…there is a consensus that the mind of a dog is very similar in capacity and behaviors to the mind of a human 2 to 3 year-old’,62 and ‘The ability to experience positive emotions, like love and attachment, would mean that dogs have a level of sentience comparable to that of a human child. And this ability suggests a rethinking of how we treat dogs’.63

It is this degree of (and type of) sentience that is at the crux of this review, and of its conclusions and demands for change. We argue that the evidence for the extent of canine sentience was already out there from ethological studies, but we also show the evidence collated herein comprising recent, ethical fMRI brain imaging of ‘volunteer’ dogs augments and fortifies it. Our inference of these canine fMRI experiments, and the weight of evidence regarding what they indicate, is supported by fMRI experiments involving humans—comparative studies involving healthy, asleep, minimally conscious and vegetative state individuals—showing that imaging ‘brain activation patterns’ are reliable indicators of conscious thoughts, intentions and emotional states/feelings. Because all this evidence shows that dog cognition and sentience are advanced and complex, and that dogs must experience a variety of emotions, including positive emotions such as love/attachment, pleasure and empathy, we argue that this should prompt a more empathic view and response from us—particularly, perhaps, from ethicists and others involved in HBAs and approving, licensing and conducting invasive experiments involving dogs. Because HBAs, by their nature, assume some degree of uncertainty about feelings and emotions in dogs, our review challenges this assumption and demands a change in how HBAs are performed. My colleagues and I would argue, based on what we see as formidable evidence, that HBAs must as a matter of urgency factor in this evidence, and the associated and inescapable conclusion that canine feelings and emotions are real and genuine—and therefore dogs in laboratories and in invasive and harmful experiments suffer much more than we currently accept, resulting in greater harm. This is a position supported by philosophical and ethical deliberations: Peter Singer argues that sentience gives rise to interests, and these involve pursuing and experiencing pleasure, as well as avoiding pain and suffering64; Tom Regan, that lives involving pleasure have intrinsic value65; the ability to experience pleasure leads to a capacity for quality of life66; and others. Harm to animals in laboratories, therefore, results when pain or suffering results directly from experimental procedures, caging and so on, and from denying those animals much of their agency, control over their circumstances and the freedom to pursue their natural behaviours and associated pleasures (eg, exercising, foraging, exploring, choosing social partners and so on13 67 68). Further, there is a strong argument that animal welfare is compromised if an individual cannot flourish emotionally, and that happiness and pleasure (and so unhappiness and suffering) depend on emotions and moods, as well as pleasant and unpleasant experiences.69

We think it is fitting to conclude with neuroscientist Gregory Berns’ statement following his work: ‘The fMRI data makes it harder to deny that dogs have feelings very much like we do and that they deserve a consideration under the law that treats them as more than a piece of furniture. We have raised the bar for treating the dogs as sentient individuals with free will. There are still over 50 000 dogs used in research every year [in the U.S. alone], so it is an uphill battle. Most of these dogs are either bred as ‘laboratory-dogs’ – usually beagles – or are acquired from shelters. I hope that our research will show that dogs have many of the same emotions that we do, and that it will become harder to justify using them as research subjects’.70

The authors believe that regulatory authorities worldwide should recognise the urgency to review and re-evaluate the need, ethics and legalities of the use of dogs in invasive research and testing, given our current knowledge and increasing scientific evidence of dog sentience and cognition.