Last week I talked about the case of Clever Hans and the importance of verifying abilities in animals. This week I’m going to expand on that by discussing how different training procedures can cause animals to use different cognitive strategies in a task.
In order to test whether an animal has a particular cognitive ability, we usually have to use tasks that animals don’t immediately understand how to do. We could easily explain such tasks to humans, but since animals don’t understand language, we have to “train” them to do tasks before assessing their performance. Depending on the task and the particular animal being studied, this can take anywhere from a few trials to a few months or more. Sometimes the animal never learns the task, which tells us something important, too.
To illustrate the importance of task training, we’re going to look at a recent study investigating whether a harbor seal could understand the concept of same/different. It would be easy to study this concept in humans because we could just tell them to indicate whether two pictures are the same or different. Not only can we not do the same with animals, but also the concept of same/different is especially complicated because it’s abstract. Understanding same/different involves more than just making a simple association; it requires making a particular cognitive judgment about the relationship between two stimuli, including stimuli never encountered before.
Scholtyssek et al. (2013) investigated whether a harbor seal could indicate whether two pictures were the same or different. The seal began each trial with his head on a “stationing target” in front of a monitor. One picture would appear on the left side of the monitor, followed by another picture on the right side of the monitor. If the pictures were the same, the seal was to keep his head on the stationing target. If the pictures were different, he had to touch his nose to the monitor.
In Stage 1, the researchers trained the seal on one image pair at a time (each image pair formed four problems: two same and two different). The seal worked on these same four problems until his accuracy was at least 80% (to give you an idea of how slow this training can be, it took almost 2000 trials for the seal to reach 80% accuracy!). Then the researchers trained him using a new image pair, either until he reached 80% accuracy or 360 trials.
They found that the seal reached 80% accuracy for five different image pairs, but was unable to do so for a sixth image pair within 360 trial limit. This indicates that, rather than learning the concept of same/different, the seal was actually just memorizing the correct response for every single problem (i.e. Image A & Image A: don’t move nose; Image A & Image B: touch monitor with nose; Image B & Image B: don’t move nose; Image B & Image A: touch monitor with nose). When he couldn’t store any more problems in his memory, he couldn’t accurately do the task.
In Stage 2, the seal only saw 5 trials of each problem (20 total trials) for each image pair, before moving on to the next pair. It took the seal fewer than 18 image pairs (about 350 trials) to reach the 80% accuracy criterion. He was 80% correct even though he had seen fewer than 20 trials of an image pair, and even though he had already seen more than five other image pairs. This suggests that he may have learned the same/difference concept, rather than memorizing the correct response for each problem. However, he could also have just gotten quicker and better at making associations between the specific problems shown and the correct responses.
To rule out this possibility, Scholtyssek et al. went a step further and tested whether the seal could do the same/different task with completely new images on every trial (rather than every 20 trials). For Stage 3, they trained the seal using sets of 15 images to create 30 problems (15 same and 15 different), where each image only appeared in three trials (one same trial and two different trials) and no pair of images was shown twice. Once he reached 80% accuracy, they tested him using completely new images on every trial.
The seal reached 80% accuracy in the fifth image set during training. More importantly, his overall accuracy on the test trials was above 80%, confirming that he had learned the same/different concept.
Overall, this experiment demonstrates the importance of training animals to do a task using the cognitive ability you’re testing for – in this case, learning the concept of same/different. For small numbers of images, it requires less cognitive effort to just memorize each problem than it does to create a concept or rule about them. For the training in Stage 1, which only focused on one image pair at a time, it was actually more cognitively efficient for the seal to memorize the correct responses to individual problems (at least at first). In Stage 2 and especially Stage 3, new images were shown much more frequently, causing the seal to learn a rule (i.e. same/different concept) to generalize to new images, which was much more efficient for those types of training.
Since we can’t give animals instructions for how to do a task (or which type of mental strategy to use when doing a task) we have to be very careful about how we train them – just another factor to keep in mind when studying animal cognition!
(Note: Scholtyssek et al. also had a fourth stage in their experiment, where the “different” problems had two of the same image, but in different shades of gray or with different patterns. As the title of their paper indicates, the seal was able to use his concept of same/different to respond significantly more accurately than chance!)
Scholtyssek, Christine, et al. “A harbor seal can transfer the same/different concept to new stimulus dimensions.” Animal cognition (2013): 1-11.