New Caledonian Crows Learn the Functional Properties of Novel Tool Types

We carried out the experiment with five wild crows captured on the island of Maré, New Caledonia. Three of the crows (Caesar, Laura and Bess) were adults more than 2 years old and two (Mimic and Pepe) were sub-adults less than 2 years old. Based on sexual size dimorphism [21] , Laura and Bess were female. The crows were housed in a five-cage outdoor aviary close to the location of capture; the cages varied in size but were all at least 8 m 2 in area and 3 m high. Caesar, Laura and Pepe completed the entire series of experiments. Bess was replaced by Mimic after Experiment 1 because of a neophobic reaction to the experimental apparatus. After learning to stone-drop, Mimic participated in Experiments 2–6 and 9–10, but did not take part in Experiments 7 and 8. Again, this was due to a neophobic reaction. All crows were released at their site of capture after testing.

The vertical, clear glass tubes used for the stone dropping tasks were 180 mm high and 50 mm in diameter ( Figure 1 ). Large stones weighed 14 g and small stones 2 g. Polystyrene and rubber blocks were of the same size and colour and weighed 0.25 g and 16 g, respectively. During the search experiments, the tubes where food was hidden across trials were the same size (50 mm in diameter and 70 mm long), shape and colour. For the unrelated tool discrimination tests in Experiments 9 and 10, crows were presented with a crevice made of two Perspex sides (100 mm long×70 mm high×10 mm thick) that were positioned 12 mm apart. The crows could chose between three white tools in Experiment 9 that were 80 mm long and 4 mm in diameter and made of different materials (a 6 g length of metal, a 0.4 g length of plastic and a 0.4 g length of string). In Experiment 10, the crows could choose between two white tools 80 mm long, one with a diameter of 0.4 mm weighing 0.6 g and the other with a diameter of 0.7 mm weighing 1.2 g).

The crows were tested in visual isolation from other crows. Trials began when a bird flew down to the table to investigate the apparatus and ended after food retrieval or 5 minutes. Crows varied in the distance that they could reach into the tube with their bills to obtain food. To ensure food was the same distance out-of-reach for each crow, we initially presented each crow with meat floating at differing heights until the ‘reachable height’ had been established. This was the lowest height at which the crow could remove the meat from the tube with its bill.

Once this height had been established, four crows (Caesar, Laura, Pepe and Bess) were first given five trials, each of 3 minutes duration, with the original Aesop's fable paradigm, to see if they would spontaneously drop stones into water (Figure 1). These four crows were then given ‘shaping’ trials to teach them to drop stones. At this point Bess had a neophobic reaction to the apparatus and was replaced with Mimic, who was also given shaping trials. Once the crows had learnt to drop stones into the tube they were given the three tasks reported in [13]: matching the number of stones to distance to water (which tested whether the crows' action was goal directed), discriminating between large and small stones (which tested whether they were sensitive to the functional properties of the objects involved) and discriminating between sand- and water-filled tubes (which tested whether they were sensitive to the functional properties of the material in the tube). As in [13], crows were given 20 trials with each of these conditions. For these experiments and Experiment 5 and 6, a stone-drop/object-drop was defined as the selection of a stone or object from the table and the dropping of the same stone or object into the tube. The crows were then given two further tests of 20 trials (as in [15]) that examined their understanding of the functional properties of the objects in the experiment. Experiment 5 examined if the crows could discriminate between water and air, and Experiment 6 examined whether the crows could discriminate between heavy and light objects of the same size and colour. Experiment 6, therefore, tested if the crows understood that objects needed to be both large and heavy in order to raise the water level substantially.

Three crows were then given two experiments where they had to search for food hidden in one of two tubes that were next to either the functional or non-functional stone (Experiment 7) or functional and non-functional tube (Experiment 8) (Mimic had a neophobic reaction to the tubes and could not be tested). The crows were given 20 trials with each experiment. These search paradigms allowed us to test whether an arbitrary link between object and outcome, rather than a causal one involving stone dropping, would lead to the same level of performance as in the object and tube choice problems (Experiments 3–6). The findings in Experiments 3–6 could be explained by associative learning if the crows were capable of associating an object with an outcome within one or several trials. That is, when a crow was successful the object involved acquired positive hedonic value, and when the crow made errors the object acquired negative hedonic value. Thus the initially neutral objects involved in the experiments would have become conditional reinforcers or punishers depending on whether the crow was successful or not. In effect, the crows would be following a simple heuristic – ‘always choose what worked before’. The search paradigms in Experiments 7 and 8 acted as an associative learning control because these experiments were essentially re-runs of Experiment 3 (large stone/small stone discrimination) and 4 (water-filled tube/sand-filled tube). The only difference in these experiments was that there was an arbitrary link between object and outcome, rather than a causal link. For example, in Experiment 3 the large stones affected the outcome by substantially raising the water level. Thus, there was a reason why the stones were positive stimuli: they efficiently raised the water level, unlike the small stones which displaced only a small amount of water. In Experiment 7, the large stone was arbitrarily linked to the outcome – there was no functional reason for it to be a positive stimulus when the crows were searching for food in tubes. Associative learning is driven by correlations between initially neutral stimuli and unconditioned stimuli or primary reinforcers (here, access to food). It should be irrelevant to a simple associative account whether these correlations are arbitrary or causal. Therefore, we made two predictions if the crows' behaviour in Experiments 3 and 4 was due to simple associative learning. First, the crows should prefer to search a tube with a large stone or water-filled tube in front of it because these objects would be conditional reinforcers due to their previous association with food in the prior experiments. Second, the crows should have a similar learning pattern – they should link the large stone and water-filled tube to success as quickly in the search paradigm as they had done in the stone-dropping paradigms.

In the last two experiments (Experiments 9 & 10, Figure 2), we examined if the crows' tool behaviour with sticks and stick-like objects was affected by their learning about the novel dropping tools in the previous experiments. In Experiment 9, the crows were given 20 trials where they had to choose between three tools of the same size that differed in weight and/or flexibility. In Experiment 10, they were given 20 trials to choose between two tools of the same length, one of which had a volume three times larger than the other, and a weight double that of the other. If the crows were using a heuristic, such as ‘always pay attention to the weight of objects when food is out-of-reach’, we expected them to continue to discriminate between heavy and light objects in other situations where food was out-of-reach. If the crows had an understanding of the actual mechanics of the task, we expected the transfer of knowledge of physical properties (e.g. weight) to new tasks to occur only when the mechanics of the tool use were the same. Therefore, by changing the structure of the task, but not the relative properties of the objects involved, we could examine if the crows would transfer knowledge of the functional properties of objects into situations where such knowledge was now irrelevant. That is, we could test whether knowledge about a new type of tool use involving stones and similar objects merged in a non-functional way with existing knowledge about stick tools.