When fish are taught to count

fish

Fears have been expressed about Artificial Intelligence (AI), which uses computing power and machine learning to imitate sensitive human activities. At the same time, biological processes, plants and micro-organisms are seen as the answer to dangers that confront the earth because of climate change. Could natural agencies be harnessed to perform computations?

A paper by Jouby, V Schluessel, N Kreuter, I M Gosemann and E Schmidt, from the Institute of Zoology, University of Bonn, published in the journal Scientific Reports, suggests that it is possible. The paper reports that the Zebra mbuna, a species of fish found in Lake Malawi in Africa, and the stingray, have been trained to “add and subtract by one, from numbers from one to five.”

The method used conditioned-response-type training, with a stimulus that called for a mathematical task. In traditional Pavlovian conditioning, when a dog is given food along with the sounding of a bell, the dog begins to salivate every time she hears the bell. This involuntary response can be extended to voluntary actions on command. The dog is given a light tap on the rump while saying 'sit'. Every time the dog happens to sit, she is rewarded. As a result, the dog ‘learns’ to sit when asked to do so.

In place of one stimulus, like a bell, or a word, the researchers used two stimuli. The first was to show a number (of objects) and an instruction, ‘add’ or ‘subtract’. And the second was to show a pair of images — one less and one more object. And the task was, first, to make out which task was to be done, and then carry it out, by making the correct choice.

The experiment

Zebra mbuna or stingray fish were housed in tanks with a trapdoor that led them from the ‘housing’ area to the ‘experiments’ area. Here, they were presented with the first stimuli, power-point slides or flashcards, that showed 2, 3 or 4 triangles, squares or circular dots, in either yellow or blue. If they were blue, the task was to add one to the number of shapes shown. If the colour was yellow, then one must be subtracted. And the next task was to choose from the possibilities shown, as in the picture.

Many of the fish could be successfully trained, the paper says. The training was with stimuli of 2 or 4 objects, with correct answers of 1 or 3 and 3 or 5, respectively. And in the testing, there was the additional stimulus of three objects, with correct answers of 2 or 4.

The range of incorrect responses was also extended, to be off by a count of 2. The sizes of the shapes were varied during the testing, to show that it was indeed the number of objects that the fish had learnt to discern.

The paper speaks of other work with primates, salamander, mockingbird, jungle crows, sticklebacks, guppies or the eastern mosquito fish. And fish have demonstrated impressive cognitive ability. Finding and remembering the way through a maze, for instance. And in groups, fish that learnt the way they can transfer learning to others.

Live applications

An application of transfer learning is in breeding salmon, where billions of salmon are released from hatcheries, to go forth and grow in the open sea. But barely 5% survive to become adults, because they lack the skills to hunt for food and to avoid predators.

To recognise food in the wild, baby salmon can be taught, by exposure, or by introducing a few ‘leaders’ who know the skill. But to recognise predators, the skill is taught with the help of submerged TV screens! On the screens, young fish watch scenes of enemy species devouring salmon, and this evokes a flight response when the predator species is seen later in life.

The paper says that the math exercise represents a new level of competence — short-term memory to remember what the task is and then carry it out. The structure of the fish brain is different from that of primates or birds, and lacks the neo-cortex, which is a part of the brain recently evolved that helps numerical processing by primates, or the similar, avian pallium, which is found in birds. That honey bees and fish are able to deal with numbers suggests that complex brain structure is not necessary for complex cognitive skills, and it may not be right to refer to the animals as primitive or lower vertebrates.

Nonetheless, the paper is inconclusive about increased mathematical competence among the fish used in the trials. The fish are ‘opportunistic feeders and not hunters,’ the paper says, and there is no apparent value for numbers in activities like mating or reproduction. But there may be unrecognised ecological purpose, advantages in complex and changing environments, that have led to the evolution of sophisticated numerical, even mathematical ability, and in species that are anatomically and evolutionarily far separated, according to the paper.

(The author is...)