RNDr. František Šusta, PhD., www.trainingisialogue.com
“All the so called spells in fact are only a question of balance. To be a good witch you have to become the central stable point in the middle of moving see-saw.“ – Terry Prachet, The Discworld Series
In our job as positive reinforcement trainers, it is always challenging to train more than one animal at a time. Especially in zoos, where it is very rare to have just one animal, most of the time animals are housed in groups. It is still possible to train animals in groups if we understand that the effort of every individual is influenced by relationships inside that group. To set the animals up to succeed, we need to appreciate them as a part of a group, and monitor their individual successes.
In Prague Zoo, where I have worked since 2017, we have trained a group of Ring-tailed lemurs to participate in a demonstration. Check out the video clip below – https://www.zoopraha.cz/zvirata-a-expozice/zvirata-se-uci/seznam-zvirat/7140-lemuri-kata, https://www.youtube.com/watch?v=J1PbZkI1fJU.
The demonstration was presented by one person. The number of animals would change from 8 to 11. The hierarchical structure of the animals was dynamic and their actual relationships strongly influenced the training and participation of each individual. To enable us to objectively assess what is going in the group and continue to set them up to succeed, we have devised the following system:
The basic idea was, that for good training of any group, it is necessary to maintain a “balance inside the group” for each behaviour. That is why each animal, for each behaviour in the demonstration was rated accordingly using the following variables:
1) K – Does animal KNOW the behaviour?
2) R – Expected REWARD – The reward which is expected by the animal. Remember, the animal does not perform the behaviour to reach the reward we choose but to reach the reward it expects to receive. This depends on the reinforcement schedule and many other conditions.
3) PA – POTENTIAL “to be ASSAULTED” – For example the PA increases when submissive animal is located right under the branch where a dominant, aggressive animal is sitting. For another species the PA increases when a submissive animal for example, a thinner sea lion, is laying belly up on the bank when the dominant animal is close by. For hoofstock a high PA is when the weaker individual is inside a smaller paddock where the dominant one is controlling the only escape route. All the designated values of these categories depend on the species ethology, trainers understanding and knowing the animal.
4) D – DIFFICULTY (also control over resources) – How difficult is the behaviour for the animal. From the animal’s point of view, this can be also understood as control over resources (food), because the behaviour is the way to reach the reward – to get control over it. For example, when lemurs are sitting around the trainer, the highest control over resources and so the smallest level of D, can be found by the animal sitting the nearest to the pocket with rewards – the animal has the highest control.
Of course it is logical, that if the animal KNOWS (K) the behaviour, it will probably do the behaviour. And if the animal expects a huge REWARD (R), it is highly motivated to do the behaviour. So by increasing K and R you increase the probability that the behaviour will occur.
On another hand, increasing “POTENTIAL to be ASSAULTED” (PA) during the behaviour decreases the chance that the behaviour will occur, as well as it will increase the DIFFICULTY (D). So increasing PA and D will decrease the chance that the behaviour will occur. Knowing this, we can establish the “Balance index” (IB), which speaks about the probability that the behaviour is running at the moment. We count it as IB = (K×R)/(PA×D). The actual levels of IB for each animal gives us the basic information about balance inside the group. We found that this model, which was developed for lemurs, could be successfully used on other species/groups we train and it can be used to solve many challenges.
How to calculate the IB ?
For each behaviour, we assign each animal the numerical value from the following variables. For each animal we assign it only for the behaviors we actually ask at any particular moment. For example, if we want one animal to jump and the rest of the group just to sit and wait; we assign values of variables for “jumping” for one and for “sitting” for the rest of group.
For counting the IB from numerical values for each variable you can use this prepared form:
After we are able to count the actual IB for any animal in the group, any time during the group training, then the most important rule is: At any time in the group training the IB must be the highest for the most dominant group member. It will then begin to decrease for the others, according to their decreasing position in hierarchy system. If we do not ensure that the IB of the most dominant member is the highest, this imbalance will lead into an imbalance of the behaviour in the whole the group. Even small lost of control over behaviour of the dominant group member, will lead to much bigger changes in lower hierarchically positioned animals.
On another hand, if we ensure that the IB is the highest for the most dominant member and then decreases for the others with their decreasing hierarchy status. We will have a calm and balanced group through the whole training session and we can successfully train behaviours with each of the group members. You can see from the variables above that there are many possibilities. We can increase the IB of the dominant member just by asking him for the behaviour which he knows well (highest K). Or we can give him a jackpot for a simple behaviour and increase his expectation of the reward (R). We can ask him for an easy behaviour or we can assign him a station which is closest to the food resource (trainer or the food bin). Both examples will decrease his D. We can station/train subordinate animals away from the dominant ones, somewhere where they will feel the most safe and have a lesser likelihood of being attacked (decrease their PA). There are many possibilities, you just need to be open minded and flexible.
In 2012 a Czech zoo asked me to solve a problem – “We have 4 adult South American Sea lions and have successfully trained them for many years. The each know the location of their stations during the show and do not leave these positions unless instructed (see the picture). This year Týna started to leave her station many times during the demonstration. What can we do?” The question was “What changed during last year?” and the answer was “with Týna nothing changed, but the dominant female Líza became blind in one eye”. So we know that Líza which was originally the dominant group member, had become physically handicapped. Then a simple test of the hierarchy showed the actual status of each member inside the group. The ranking seemed to be Týna and Mano 1-2, Hugo 3, Líza 4. See the IB results for the time when all the sea lions are stationing.
You can see that during stationing Týna has the greatest distance from the food resource, which gives her the smallest value of the IB for that behaviour. It wasn’t a problem last year, because her hierarchy status was lower. But now that Líza is handicapped and Týnaś hierarchy status has increased, this IB value is not sufficient for her. What can we do with this information? We can increase her K (knows) by giving Týna private training sessions. We can train her to learn that when she doesn’t know what to do, she can always go, without any cue, to her station. This would mean that her station would became her default behaviour – a place that is always reinforcing and her K on this behaviour would be the highest in the group. The other sea lions station on cue but not as default behaviour. We can also increase her R (expected reward) by giving her more reinforcement on this station. And we can decrease the D (difficulty) by changing the position of her station or position of the food bin, both is about distance from resource. Or we can change her station to lower and wider stone, which doesn’t take as much effort to climb.
What does it mean “the most dominant member”?
Many positive reinforcement trainers hate the words “dominance” and “hierarchy”. But they exist in the animal world. The issue comes from too many “animal whisperers” misinterpreting behaviour. Dominance exists and with science we can measure it. We only have to understand one thing – the DOMINATOR is NOT the LEADER. With the IB we account for DOMINATORS, not for LEADERS. The dominator is the one who has the most control over the resources and this title can given to animals not necessarily because they are the strongest member of the group, but many times just because the other members will let them get away with anything, for example, young animals. The leader is the one who has the biggest influence on the behaviour of whole the group and who is voluntarily followed by the others. To allow me to explain what “dominator/leader” means more in-depth, I’d like to use a quote from my book, which you can download for free on my website www.trainingisdialogue.com. The direct address of the book shortcut is here and the text on group training can be found from page 69 of the PDF document. In this section I explain the status of dominator and leader in the group of Prague lemurs in 2012:
Pancho, the youngest of the older lemurs, is physically the strongest and if a heap of tasty food emerges from somewhere, Pancho is able to disperse the surrounding animals by force in order to enjoy the food. I call that a dominator. I wouldn’t call him a leader because if a flock of ducks flies over, there’s a loud noise or a visitor brings a barking dog and the lemurs start to get nervous their eyes turn to Motorka, the oldest group member. If Motorka is afraid, the group takes flight. If Motorka is at ease, the others are at ease too. Nobody in the group is interested in what Pancho, the dominator, does in these moments. You’ll never see Motorka fighting, but he has no problems getting his fair share of the food. He simply arrives and eats, without any conflict. If he doesn’t like a new incoming group member, Motorka just frowns at him and at that moment two other lemurs who always follow Motorka attack the unwelcome newcomer. I can’t remember Motorka ever being injured but Pancho has suffered injuries at the hand of the other lemurs.
Motorka is the animal that determines what the group will do and I don’t think he chose the role himself. The group has simply united around his experience and is following him. Where as Pancho is the individual that achieves privileges thanks to his physical strength. That is how it is now, but the time may come when he becomes weaker, Pancho chose his role of dominator and now he has to go to a lot of trouble to preserve his position. These two lemurs beautifully prove what science knows today. That dominator and leader are two separate roles and often, although not always, two separate individuals.
When you start training with a group:
When we start to train any group of animals with in the zoo, that have never been trained before, we are many times surprised that the first cooperative animal is not always the strongest member. Often it can be the weaker or handicapped one. Especially when the animals are afraid of the trainer. Understanding the rule of IB helps us to explain this phenomenon. When the new trainer is still understood by the group as potential danger, the PA (potential to be assaulted) for each member is so high, that the final IB = (K×R)/(D×PA) is too low to enable any member to cooperate in the training. But then when the trainer is respecting their needs and building trust, the relative PA is decreasing and the final IB increases. For weaker group members it is natural to work on lower level of IB and so when it overcomes their critical level, it enables them to come and participate on first training. But for dominant members the IB at this moment is still too low and so they wait and observe the situation. Then their trust that the trainer is safe decreases actual PA, so that final IB is high enough to let them cooperate. At this moment dominant members many times come and attack the weaker members to steal their chance to work. As trainers we need to be prepared for this situation and prepare already for weaker and dominant members stations and behaviours around the trainer, which will respect their hierarchy status and keep balanced group.
The phases of training a new group look like this:
- No one trains, everyone is terrified. All the animals keep safe distance from trainer and the trust is slowly increasing. Symbol of the trainer with food. See #1
- First animals start to cooperate – trust is increasing, first animals (mostly the weaker members) come to “try their luck”. See #2.
- No one is frightened, everyone is training – dominant members come and trainer has to give them all stations around him/herself as a resource according to their hierarchy status. See #3
- No one is frightened, everyone is training more – trainer has the possibility to improve stations according to his/her own needs, not necessarily reflective of the hierarchy inside the group. It is possible because the animals now know the stationing behaviour (the K in IB model is increasing). See #4
- Everyone knows everything – in this situation (very rare in zoo training) trainer does not need to respect the hierarchy because the animals know their stationing so well (their K is so high), that they do not influence one another. See #5
The IB model can also help with enrichment. It is very common that dominant group members take or control the easiest food based enrichment, anything that doesn’t require too much energy to earn the food. A diploma student devised a test with a group of coatis at Prague Zoo. The question was – if we can predict which puzzle feeders will be monopolised by each particular member of the group, could we give targeted medications without direct feeding or separation of the individuals although the group has not been trained? The results show that it is possible, however, maybe risky as there are too many unforeseen issues that could happen during feeding. You can read the article, which was released by Wellspring (ABMA magazine) in 2015.
Šusta, F, 2015: Training Without the Trainer Through Enrichment Tools and Application of the Balance Index (IB) on Enrichment Use. Wellspring Volume 15, Issue 1: pg 2-12.
Here you can see a mathematical explanation of changes inside the group when dominant group members is not under stimulus control. We will use the same example with lemurs group, the behaviour is basic stationing around the trainer.
Example of ideal situation (IBi):
Action 1: Lemur demonstration at Prague Zoo.
Behaviour: “Basic positioning before the demonstration“ – each animal has to stay on its own station around the trainer.
Rem.: The basic lemurs positions are chosen so that the most dominant animal is on the ground and the others are up on branches – the most submissive are the on the highest branches where they presumably feel safe (small PA). The most dominant animals are also located closest to the pocket with rewards – so they have the best control over resources (the smallest D)
Calculation of IBi (IB ideal): for every individuals K=3 (they are all under stimulus control),R=3 (they all maintain the behaviour for ordinary rewards, no jackpots are needed), PA=1 (they all feel safe thanks to their positions) a D= 1 – 4 depends on utility of every position (needs to hold by the thin branch/staying on the ground) and the distance from pocket with rewards (emoticon), which leads to different value of control over resources. Thanks to this difference in D (difficulty) the IB is highest for the dominant and decreases with decreasing status in hierarchy – which ensures the balance in the group during this behaviour.
This was the ideal situation but now it changes. One day the trainer comes into the lemur group and asks for stations and they can see that Ovisiti (second in hierarchy) is not on his station. What is the reason?
Rem.:The most dominant lemur (Pancho, staying on the ground) is not fully under stimulus control and so he can sometimes jump up. This leads to the situation where the second in the hierarchy sits permanently out of the group and not on its station. Although the trainer does not see the change in IB of the first dominant one, he can see the change in the second one.
The IBr of the most dominant animal is not so different from the ideal level (IBi) and so the animal still participates in the behaviour and the trainer does not see any change. The IBr of the second animal however, has reduced (IBr ≤ IBi) and so the animal does not have enough motivation to stay in the behaviour. This is a testament to how a small change in IB of the dominant animal (IBr =67% of IBi) could lead to much bigger changes in the lower hierarchically posted animal (IBr =30% of IBi). This model practically shows how the success of the group behaviour is affected as a result of the trainer’s control over the dominant animal. Very small changes on this control lead to bigger changes on the other members.
How to solve the situation?
1) – The trainer can change the group’s behaviour by changing the basic position of the second lemur which will lead to a lower PA (the animal is more safe). This will probably increase the difficulty and decrease the control over resources (higher D) so the trainer would probably have to increase the R (expected reward). The final IBr of this individual will return to 89 % of IBi, high enough to make the animal sit at this new station. The animal is also far enough from the rest of group not to decrease their PA.
2) – The trainer will not change the basic positions, but he will focus more on the dominant animal. Starting again – leads to better K variable; higher expectation of rewards by intermittent schedule – makes higher R. The increase of IBr on the dominant animal will lead to better control over him and so better feeling of safety in the second lemur (lower PA). The trainer can also make higher R of the second animal by increasing reward, perhaps with some jackpots and so make the IBr higher.
The IB model gives us following information about group training:
1) Every behaviours of every animal must be understood as a part of a complex group behaviour. We are training the group, not only one individual. See photo #1
2) This group behaviour stays and falls with the control over the dominant member and even small, almost insignificant changes in control over the dominant animal will lead to bigger changes in subordinate members of the group. When seeing trouble in behaviour of any submissive lemur, this is often caused by loss of our control over the dominant animal. See photo #2.
3) To have a well-balanced group behaviour IB has to be the highest for the dominant animal and decrease for the rest of group with their decreasing hierarchy status. If the IB for more submissive animals is higher than for more dominant animals, this will probably lead to less effectively maintained group behaviour. See photo #3.
4) We can establish 4 variables directly influencing the success of each behaviour. Where two of them (K, R) increase and two (PA, D) decrease the probability that the behaviour will be maintained. Two of them (K, D) are also specific for each behaviour in the demonstration, but other two (R, PA) are more specific for the animal than for the behaviour. That is why the actual value of R and PA will influence every behaviour of each animal during the demonstration. See photo #4.
5) All the actual values of those variables for each animal we can simply identify from a group behaviour, where the highest number of group members can participate in one moment, we say the behaviour has “the highest capacity”. For example, when positioning lemurs around a trainer there is capacity of 6 participating animals in one moment. In a behaviour where one animal is “self closing” the door to the transport box, there is a capacity of only one animal. See photo #5.
We have found, that this mathematical model fits not only on lemurs, but can be successfully used on other species in the zoo when trying to determine hidden challenges and motivations in the group. It provided us answers to assist us managing the following:
Przewalskii horse mares when training them to separate and stay alone inside small stalls, when training female Pere’s David Deer to gate into a small paddock and various other behaviours. This also helped to the keepers from other zoos to manage their sea lions during the demonstration and scientists working with experimental macaques to manage the group during training.
The use of this model effectively includes the group dynamic as a part of the reinforcement strategy. BUT PLEASE, DON’T BE SCARED BY THE SCIENCE OF IT. In fact, in most of the cases the keeper or trainer does not have to calculate the model fully. Often a trainer finds out very quickly usually during the first steps of calculation when focusing on one animal and disregarding the rest of group; that the IBr for the dominant animal in the group is lower than for the submissive and that the trainer has to improve the balance. And that is why we developed the model. If the trainers will use it, they have to think about the situation from the viewpoint of the group. In this way, trainers will identify the challenges without our assistance. This model helps trainers understand, that:
A successful GROUP TRAINING session is only a question of balance. To be a good TRAINER you have to become the central stable point in the middle of moving see-saw.
The address of the author
RNDr. František Šusta, PhD.