Fish cognition and welfare

Fish are the largest category of vertebrate that humans kill, and — until recently — the one drawing the least ethical attention. A single modern trawler can pull more individual animals from the sea in a night than a large slaughterhouse processes in a year. The science of fish minds has moved faster in the last two decades than in the century before it, and it has moved in one direction: toward taking fish seriously as sentient beings.

The scale problem

Land-animal slaughter is typically counted in billions. Fish slaughter is counted in trillions, and not always precisely, because the global fishing industry reports catch by tonnage rather than by individual.

The most cited estimate is the work of Alison Mood and Phil Brooke at fishcount.org.uk, originally published in 2010 and updated in 2019. By dividing reported catch weights by mean species body weights, they estimate that between roughly 0.79 and 2.3 trillion wild fish are caught from the oceans each year, with a best central estimate near 1–1.2 trillion (Mood and Brooke, 2019). Fishcount’s parallel estimate for farmed finfish is between 78 and 171 billion slaughtered annually. Neither figure counts the many hundreds of billions of shrimp and other decapods, nor the wild fish ground into feed to raise the farmed ones.

The FAO’s State of World Fisheries and Aquaculture 2022 confirms the underlying tonnages: global aquatic animal production reached roughly 178 million tonnes in 2020, of which aquaculture contributed 87.5 million tonnes — the first year farmed aquatic production exceeded wild capture for food use (FAO, 2022). Aquaculture is now the fastest-growing food-production sector on Earth.

Do fish feel pain?

The modern debate opens with a specific empirical question and a specific methodology for answering it. Sneddon, Elwood, Adamo, and Leach (2014) proposed criteria for identifying pain in any animal: the presence of nociceptors, central processing of noxious stimuli, physiological responses to injury, behavioral changes that go beyond reflex, modulation of those changes by analgesics, avoidance learning, and trade-offs against other motivations.

Applied to teleost fish, the criteria are largely met. Sneddon’s (2015) review in the Journal of Experimental Biology summarises the evidence: rainbow trout and other species possess A-delta and C fibre nociceptors on the face and body, show sustained (not merely reflexive) behavioural disruption after noxious stimuli, display rocking and rubbing behaviours analogous to those seen in injured mammals, and resume normal behaviour when administered morphine or other analgesics. In motivational trade-off experiments, zebrafish will choose an otherwise-aversive environment if doing so provides pain relief — a decision reflex alone cannot make.

A dissenting line exists. Brian Key’s 2016 paper Why fish do not feel pain argues that fish lack the specific neocortical architecture that generates conscious pain in mammals, and so behavioural responses cannot imply felt experience (Key, 2016). The counter-view, held by most researchers in the field, is that cortex-equivalent function can be implemented in the teleost pallium and other structures, and that a cortex-shaped organ is not the only substrate consciousness can ride on. The 2024 New York Declaration on Animal Consciousness sided with the majority position: there is “at least a realistic possibility” of conscious experience in all vertebrates, fish included.

What fish can do

Culum Brown’s 2015 review Fish intelligence, sentience and ethics in Animal Cognition is the single most cited synthesis of fish cognition. Brown’s argument is that fish intelligence has been “dramatically underestimated”, and that on specific tasks fish perform on par with — and sometimes exceed — non-human primates (Brown, 2015).

The capacities documented across species include:

• Spatial cognition. Many fish build and use detailed cognitive maps of their environments, remembering the locations of food, shelter, and social partners for months.
• Social learning and tradition. Fish learn migration routes, foraging sites, and predator responses from conspecifics; these traditions persist across generations.
• Individual recognition. Cleaner wrasse, groupers, and many reef species distinguish individual conspecifics and specific human divers.
• Cooperative hunting. Groupers and moray eels coordinate hunts using referential gestures — a behaviour once thought unique to mammals and birds.
• Tool use. Several wrasse species use rocks as anvils to crack shellfish.
• Numerical ability, transitive inference, and mirror self-recognition have been demonstrated in at least some teleost species.

Brown’s conclusion is that the ethical treatment of fish has not caught up with what is known about them.

Aquaculture welfare

Farmed fish live under conditions that have been compared, in welfare terms, to those of the most intensive land-animal systems. The European Food Safety Authority’s 2009 scientific opinion on stunning and slaughter of farmed fish reviewed welfare across the major commercial species (salmon, trout, carp, seabass, seabream, tuna, eel, turbot) and documented substantial welfare problems at slaughter under most prevailing methods (EFSA, 2009).

Routine welfare concerns in aquaculture include:

• High stocking densities that compromise water quality, cause chronic stress, and drive aggression and fin damage.
• Sea lice and disease pressure in open-net salmon farms, which spill into wild populations.
• Starvation before slaughter — fish are typically held without food for days to empty the gut, a standard practice with poorly studied welfare costs.
• Slaughter methods that often fail to stun before killing — asphyxiation in air or on ice, carbon dioxide narcosis, exsanguination without stunning. EFSA (2009) identified electrical and percussive stunning as welfare improvements, but adoption is uneven and slow.

Wild capture: trawling, bycatch, and dying on deck

Industrial wild capture adds a different set of welfare harms. Bottom trawls and purse seines haul fish from depth over minutes, causing barotrauma — swim-bladder rupture, eye prolapse, organ damage — before the catch reaches the surface. Most fish then die slowly by asphyxiation on deck or crushing under the weight of the catch. Stunning before killing is rare to nonexistent in commercial fisheries.

Bycatch — the unintentional capture of non-target animals — compounds the harm. Reviews estimate that roughly 10 per cent of global marine catch is discarded, with far higher rates in some fisheries; bycatch includes non-target fish, sharks, turtles, seabirds, and marine mammals, most of which die in the process (Gilman et al., 2019). The number of animals killed by fisheries is therefore larger, perhaps much larger, than the number landed for sale.

The ethical upshot

The facts line up in a way that is uncomfortable for ordinary practice. Fish plausibly feel pain. Fish have cognitive lives richer than the folk-biological picture suggests. Humans kill them in numbers no other form of animal use approaches. And the methods — suffocation, crushing, slow exsanguination, uncontrolled depressurisation — are, by any mainstream welfare standard, among the worst inflicted on any animal at industrial scale.

The case for extending the vegan ethical circle to fish does not rest on a single disputed experiment. It rests on a convergent literature (Brown, 2015; Sneddon, 2015; Sneddon et al., 2014; EFSA, 2009) and on a number — one to three trillion — that most people have never had the chance to sit with.

See also animals, sentience, and the environment pillar for the ecological cost of the same industry that produces these welfare harms.