Is option A good if the chance of success is 50%?

Well, that depends on the success chance of option B. Is it 20% or 80%? If we know that, then the choice between the options is easy. If we are not told the probabilities but instead have to learn them from our previous experience with A and B then it becomes more complicated, but still we humans seem to be able to make optimal choices in this context. To some extent, chimpanzees can as well.

Recent research showed that carrion crows can use this type of statistical inference as well during decision-making.

The researchers first trained the birds by presenting them with different symbols (purple circle, white triangle, etc) on the screen, one at a time. When the crow touched a symbol with its beak they might get a treat. The probability of getting it depended on the symbol. Each symbol had its probability of reward, from 10% to 90%. Over a couple of days, each crow saw each symbol (in random order) more than 500 times, potentially learning the probability of reward for each of them.

Next, the researchers tested the statistical abilities of the birds. They showed two random symbols on the screen and counted how many times each crow touched the symbol with the higher probability of reward. The results showed that the birds learned the probabilities linked with each symbol and could clearly make optimal choices. On average, they chose the symbol with the higher reward probability more than 75% of the time. Not surprisingly, when the difference between the probabilities was large (e.g. 20% vs. 80%) they almost always made a good choice but were less likely to do so when given a choice between similar probabilities (e.g. 50% vs. 60%). What’s more, when tested one month later, the birds still remembered the meaning of the symbols and could make optimal choices.

But were the birds actually comparing probabilities or were their decisions based on the number of treats they got for each symbol? After all, for symbols linked with low probability, the crows got fewer rewards in absolute terms.

To disentangle the effect of the number of rewards from their probabilities the researchers trained the crows with the new symbols, one for 40% reward probability and one for 80%. However, this time they didn’t present the symbols the same number of times, but showed the one linked with lower probability twice as often. That means that in total the birds got the same number of treats for touching each symbol even though each of them was linked with a different reward probability.

When the birds were given the choice between the two symbols, they clearly preferred the one with the higher reward probability. This showed that crows used relative reward frequency to maximize the reward chance.

This experiment shows that crows can flexibly apply statistical inference, that is, can use limited information to draw conclusions about the probability of an event in different contexts. Of course, this was shown in quite artificial conditions in a laboratory. However, such skills are also likely important in the wild. For example, one foraging location can be best in winter when alternative options are poor, but it may not be optimal in spring when conditions change and finding food elsewhere becomes more likely.

While crows in general have high cognitive skills, I wouldn’t be surprised if other animals can also apply statistical inference in foraging and other decisions important for their survival.

Photo: Siegfried Poepperl from pexels.com