Friday 24 January 2014

Double Effect and "Purely Positive" Training

Some people argue that there is no such thing as �purely positive� training, because even 100% R+ trainers �use� aversives.  This makes them no better than P+ trainers because in both cases, the dog is suffering during training.  They give examples of dogs �in drive� and so desperate for a reward that withholding it is aversive.  A dog with an incredibly high prey drive and low self-control, for example, might shake, cry, and show classic signs of stress and frustration when deprived of the chance to chase a toy.  If chasing a toy is the reward in an R+ training session, then the dog is not rewarded for non-performance of a desired behavior, so for these kinds of dogs are arguably going to suffer during such a session.  

There are a few things wrong with the practicalities of this scenario, but let�s go with it as a way to test whether the conclusion - that the R+ trainer�s use of aversives makes them �no better� than a P+ trainer - is justifiable.  Although there have already been some excellent discussions of this issue, I want to show how academic ethics can give dog trainers a different set of resources to make the argument. 

Trolleys and Bridges


The Doctrine of Double Effect is a popular idea in academic ethics.  It states that it is permissible to allow harm to happen if it is an unavoidable side-effect of a good action.  The classic example goes something like this:

You are standing by a train track, next to a lever that switches the points from the track on the left to the track on the right.  You notice that a train is coming, and that there are five workmen on the right track, but also one workman on the left track.  Do you switch the points?

Philippa Foot argues that it is permissible to switch the points, because five people dying is worse than one, and - crucially - you�re not killing one person in order to save five people (Foot 1967).  Judith Jarvis Thompson contrasts Foot�s example with this one:

You are standing on a bridge, next to a very overweight man.  You notice that a train is coming, and that there are five workmen on the track in front of you.  There is no way to warn them, but you know that pushing the overweight man onto the track will stop the train and save the workmen.  Do you push the man?

Thompson argues that it is not permissible to kill a man deliberately in order to save others; it isn�t right to use someone�s body like this.  The difference is in whether the death (a moral wrong) is an intentional, or integral part of the action.  In the Trolley Problem, the deaths of the one workman wouldn�t change the fact that five workmen would be saved by switching the points.  In the Bridge Problem, the death of the overweight man is an integral part of saving the five workmen.  Double effect states that it is permissible to do an action with harm as an unintended consequence, but not to do harm as an intentional and integral part of an action.  So, it is okay to flip the switch in the Trolley Problem, but not to push the man in the Bridge Problem. 

How This Applies to R+


The mechanism of operant conditioning that underlies R+ training works whether the dog becomes distressed when the reward is withheld or not.  Something positive being added to a dog who is already happy will work as a reinforcer.  The circuit doesn�t depend on there being any stress or frustration; if it happens that the dog does suffer from a bad lack of self control that causes him to get upset at the absence of a reward, this is something the R+ trainer should try to minimize.  Note that I'm not suggesting that all frustration should be completely avoided; some is unavoidable, especially when we're working with high drive dogs doing the thing they love.  Frustration, if kept to a minimal level, can actually provide additional motivation (Amsel and Roussel, 1952) and removing it entirely from the picture, whilst ethically desirable, is often practically unworkable.

The critic of the R+ trainer is assuming that the R+ trainer�s methods work like the Bridge Problem, where the death of the overweight man is an intended part of saving the five workmen.  If the stress was needed for the training to work, then this would be true.  But in fact, the R+ trainer�s methods work like the Trolley Problem, where the death of the one workman is neither intended or necessary for saving the five.  The R+ trainer�s ideal situation would be a dog who didn�t feel stressed at the prospect of his reinforcement being withheld, since the stress isn�t necessary or intentional.  

So, there is a moral difference between training methods that intentionally use aversives, and methods that have stress or frustration as an unintended, unnecessary side-effect.  It�s not true to say that there is no such thing as a purely positive training method, even though it might be true that some real-life training situations are less than purely positive. This is something R+ trainers should be working to avoid.  

References and Notes


  • Foot, Philippa (1967). "The problem of abortion and the doctrine of double effect". In B. Steinbock and A. Norcross (eds), Killing and Letting Die. NY: Fordham University Press. 
  • Thompson, Judith Jarvis (1976) "Killing, letting die and the trolley problem". The Monist 59, pp 204-17.
  • Amsel, A and Roussel, J (1952) "Motivational properties of frustration: effect on a running response of the addition of frustration to the motivational complex". Journal of Experimental Psychology 43(5), pp. 363-368
  • General notes on drive and frustration in dogs: http://www.training-your-dog-and-you.com/drive.html
  • Another way to approach the fallacy of "purely positive training": http://eileenanddogs.com/2013/07/08/purely-positive-all-positive/

A

No comments:

Post a Comment