Hacker News Reader

Comment by seanwilson | Hacker News Reader

Is this not fairly well understood? e.g.

https://en.wikipedia.org/wiki/Clever_Hans

> Clever Hans (in German: der Kluge Hans) was an Orlov Trotter horse that was claimed to have performed arithmetic and other intellectual tasks.

> After a formal investigation in 1907, psychologist Oskar Pfungst demonstrated that the horse was not actually performing these mental tasks, but was watching the reactions of his trainer. He discovered this artifact in the research methodology, wherein the horse was responding directly to involuntary cues in the body language of the human trainer, who had the faculties to solve each problem. The trainer was entirely unaware that he was providing such cues.

View on HN

This particular horse is mentioned in the very first sentence of the introduction of the paper.

I think it's a worthwhile study, especially considering such dogs are currently still used in law enforcement. It's important to be aware of the shortcomings and biases of this methodology, and to do so in a sound, scientific manner.

Is there any proof that "the trainer was entirely unaware that he was providing such cues." Obviously the trainer has an overwhelming incentive to lie.

It should be easy to train the dogs in an environment where the trainer does not know the intended result and must investigate each claim from the dog. All the dog training vids I've seen have the trainer themselves knowing exactly what lies in the course.

There are two problems with this, but both can be solved by having two people.

The first is you'd probably have to get someone else to hide the scent for you. Not so easy when training solo.

The second is that when you want to reinforce a behaviour you generally give a trigger or marker as soon as possible after the event. If you wait too long, the dog may associate the marker with another behaviour they've done in the meantime and overwrite the correct learning. So a 'yes' or clicker as soon as the dog alerts is the best way to go. But if you yourself don't know where it is, you risk reinforcing a failure. This is easy with a helper so you can glance at them and they indicate if you are correct or not.

I don’t think this would be easy. For training, the reward needs to come as quickly after/during the action as possible. Most dogs don’t have a great mental capacity for figuring out that they’re being rewarded for something that isn’t happening right now. If the dog has to wait even 10 seconds for the trainer to figure out whether to reward or not, a lot of dogs are going to get confused.

During training/testing a second person, who is not visible, audible, or smellable to the dog, and who knows the conditions of the course, observes the pair via cameras and microphones.

When the dog hits on something, the observer presses a button, indicating which target or decoy was indicated, or that an unexpected other object was indicated. The observer may also indicate when the dog seems distracted, or especially attentive to the handler.

A computer takes the buttons as inputs, and if its intermittent reinforcement program indicates it, the handler receives an instruction to reward the dog. The program may also reward the dog for misses, or just for paying attention to handler instructions, so the handler is not able to associate rewards with accurate hits. The handler then still has to investigate the hit to determine what the dog found.

Reply on news.ycombinator.com