Clever Hans or Cognitive Canine?

This is a copy of the original article I wrote in June 2018.

I am a big fan of concepts work and discrimination games. These are games where I teach my lurcher, Beanie, how to tell apart colours, match visual and audio patterns, use word cards to indicate whether or not patterns are identical, tell apart quantities by size and to count items, as examples.

There are some cornerstone concepts to begin with that can then be combined to create endless games for dogs to think through, decide and indicate effortlessly. Currently I’m working on teaching Beanie to indicate how many items of a sample item are contained in a larger set of mixed items. I’m teaching him to express preference for how he wants to spend his training time using word cards and verbal cues. I’m also broadening his audio matching skills to be able to indicate whether or not multiple notes played on a piano are identical to each other.

These games are excellent for active dogs with limited mobility, whether that’s due to an injury, puppyhood or old age. They build confidence for shy dogs, but also work very well for independent thinkers and draw upon our dogs’ natural abilities. I constantly scout for inspiration for new games I could teach Beanie. I read up on our dogs’ perceptual and cognitive abilities. Most importantly, I love watching Beanie’s cogs spin when we work together.

Discrimination games have a set pattern. I teach Beanie how to understand my question, draw upon his strengths (sight and smell), make up his mind and show his decision with a very easy paw or nose target. These games are my window into Beanie’s perception and cognition. The decisions he makes, especially the mistakes, show me how he sees and reasons with the world. And Beanie loves it too. Anyone who has seen our videos on The Unlikely Tricksterswon’t fail to notice how engaged he is in every game!

Recognising and working with Clever Hans

Almost every video I post brings up the Clever Hans question. Could Beanie “just” be reading any subtle cues in my body language to make the right choice? The short answer is yes but I do my best to minimise this to the extent that the game is still fun for both of us. Read on for the long answer.

Most of us know the story of Clever Hans, a horse who purportedly answered arithmetic questions. Examined further, it was discovered that he instead read subtle body language cues from his trainer. These findings have had profound consequences on experimental design in animal cognition studies. (One might argue that freestyle dancing or agility routines are just as much a symbol of dog cognition and are visibly cued by the handler on the ring without undermining the dog’s ability to impress but that is a topic for another day!)

Now I want to fess up about Beanie’s Clever Hans clues to demonstrate how simple they can be. Beanie will favour:

  1. the choice that is closer to him, both on the floor or held in my hand
  2. the choice that is closer to me, i.e. if I sit more to his left or right side
  3. the choice that is shown first out of multiple options
  4. the choice that moves
  5. the choice in the direction of my gaze or my face
  6. the choice I touch last when placing down his options

I know these because I have fastidiously recorded videos of every single training session for the past 7 months and re-watched them in slow motion, frame by frame, using software like iMovie.

This list is of great interest to me because it shows how beautifully Beanie observes me. Isn’t the fact that he can pick up on such subtle cues a testament to the strength of our relationship? If he can pick up a cue that nobody watching the video can spot, that in itself blows my mind. Beanie and I are almost psychic in daily life because that’s how much work we’ve both put into building our communication. I am very proud of that. I cannot be his voice if he didn’t know how to communicate with me and I cannot teach him if I didn’t know how to influence him.

But I am a scientist by profession. I understand how double-blind studies are designed. The ideal set-up for our games to truly eliminate confirmation bias would be for two handlers I understand that the most truthworthy cognition studies are ones where the handler is not present at all (unless bias is the topic of the study!).

So I find myself often torn about where to draw the line between my inner dog owner/enthusiast and the scientist. Simply placing the cards on the floor instead of holding them in my hand isn’t enough to eliminate handler bias. Neither is having the choices against a wall nor having him wait at a distance as I set up and leaving the room to video his choice. From a peer-reviewed cognition perspective, there is no game that we could play where I could guarantee that my mere presence in the same room as Beanie will not influence his choice.

Where does that leave us?

I’m not sure. I do my best to minimise my tells. Here’s what I do:

  1. I show my videos on mute to others to look for tells I can’t spot. I publish our videos on our Facebook page for everyone to do the same.
  2. I proof my tells, teaching him that the body language cues I am aware of are not necessarily indicative of the right answer.
  3. There are games we play where I can show opaque cards and ask for visual matches or size discrimination. I first teach Beanie the game by knowing and showing him the right answers. Eventually I move to blind versions where I don’t see the card I hold up until Beanie has made his choice. But that is not feasible for all the games we play.

Extending this line of thought, I also cannot imagine that either of us will enjoy our training time as much as we do if we had to spend it mostly apart! So I draw the line where I am happy with the trade off between strengthening our bond by working together every day and scientifically proving Beanie’s cognitive capabilities, both of which are incredibly valuable to me as his guardian and a scientist. But that means accepting that I will never know where Clever Hans ends and Cognitive Canine begins. I am fine with that.

Anyone venturing into cognition games will have to find their own balance. I believe that Beanie’s abilities are nothing short of remarkable, irrespective of where he is on the Clever Hans vs. Cognitive Canine spectrum. I value us as a team. I value my ability to teach him. I value his powers of observation. I also value his intelligence.

Dog vision and environmental influences on colour discrimination

This post is a copy of my original article on Medium here.

I absolutely love training concepts with animals. My jaw hit the floor when I first saw Ken Ramirez’s video of Coral combining match to sample and quantity recognition on numerous objects in a double-blind study (see videos hereand here).

I will preface this article with a caveat that it simply reflects my reading, my own experience and my endless curiosity. If you have any view on what I’ve written here, I want to hear from you. This isn’t me pontificating at the world. This is me inviting comment from people who know and think about stuff like this.

What are examples of concepts?

  1. Match to sample (123456)
  2. Imitation (our videos: 12)
  3. Colour (cat: 1), shape, quantity (2) and size (34) discrimination
  4. Quantity recognition/counting (12)

I’ve linked to our videos above.

I want to focus on colour discrimination today. Thisis how I have taught it to Beanie. But we cannot discuss it without first understanding how a dog sees the world. Hereis a great rounded article on the subject. Visual acuity is discussed well hereand here.

Dog Vision — my summary relevant to colour discrimination concepts

1. Dogs are red-green colour-blind. Their colour spectrum goes from blue to yellow. Reds appear green. Darker shades appear grey/green. They can also see the colours black and white.

2. Their vision is blurry. If something is 20 feet away from them, it has the same acuity to them as it does to humans if the object is 75 feet away from us.

3. The world is very bright to them, maybe best described as ‘over-exposed’ in our world. Although dogs see better than us when it’s darker, their visual acuity drops in the dark like ours.

4. Different breeds have different visual skills. (I have a lurcher. He can spot a running squirrel a football field away from us before I can.)

Why this article?

Teaching Beanie the cue “blue” was easy. Within a few sessions, Beanie would target any object that had any blue colour in it given two choices where the other object had no blue colour at all. But I simply couldn’t teach “yellow”! I couldn’t fathom it. I wondered if he simply couldn’t see it at all in my lounge, in that lighting. I started reading about a dog’s visual perception. Turns out that I wasn’t too far off. I would like others who work with their dogs like I do to gain from my learning or tell me I’m wrong.

Technology to simulate dog vision — the Dog Vision HD app

There are now a handful of phone apps out there, most of which I have tried out since, but the one I came to like is called Dog Vision HD. I discovered it from a vlog by Susan Garrett on why colours matter in dog agility. There is also a website https://dog-vision.comwhere you can upload images and view the processed outcome.

The rest of this article is about my findings in using the app to explain why Beanie was struggling with colour discrimination and how the environment I set up for him made a huge difference.

Here are some observations of human vs. dog vision. They are according to the Dog Vision HD app, which incrementally toggles colour blindness, acuity and exposure.

Dog vs. human vision: a few photos

First, the environment. My front garden looks like the top photo to me. Each picture in the bottom adds one feature of dog vision at a time (left to right). The photo on the bottom far right is what the dog supposedly sees. Not quite the same at all.

Human vs. dog vision of the environment

Similarly, below is what he would look like to himself compared to how he looks like to me from a couple of metres away on a beautiful sunny day.

(By the way, I am just as amazed as you are that he sat that still whilst I fiddled with every setting on the Dog Vision HD app to take these four individual photos over many minutes. Never underestimate a sighthound’s ability to just chill.)

Notice how blurry and over-exposed he is in the sunshine. Note how washed out the grass looks. Note how the fence blends into the shrubs, although there is still depth in the image. Note how anything black stands out.

Dog vision of a stationary brindle/white dog outdoors on a sunny day.

Let’s move indoors — far left below is what stares back at me in my full mirror when Beanie and I sit awkwardly on the staircase together for rubs every evening (I hope you have spotted the lean). Far right is Beanie’s equivalent image. There’s only sunlight coming through the window on my right.

Note how dog vision is brighter than ours on the darker stairs and how the white walls and the door blend away behind us. Note how my dark blue jacket just about stands out from the black trousers. My face is hardly discernable at this distance.

Human vs. dog vision indoors in natural light

Then I switched on an LED light that is right above us. See what a difference that lighting has suddenly made to Beanie’s vision in terms of the amount of exposure in the room. Note how my blue jacket is still different to the black trousers.

The effect of indoor lighting on dog vision

My key point here is simply that our visual perception is very different to that of a dog. Compared to us, dogs have gained better night vision and motion detection in return for poorer colour vision, over-exposure and blurriness.

The importance of environment — lighting, contrast, lamination and object size in colour discrimination concept training

Now let’s look specifically at colour discrimination tasks. I recently used three identical pairs of objects in blue and yellow for colour discrimination concept work with Beanie. The items are silicone moulds, square plastic lids and oval bottle lids from my kitchen. It is accepted (see hereand hereas examples) that dogs can tell apart blue from yellow, so the colour choice in itself is not an issue.

I placed the items just like I would to Beanie (on either side of his paws) and lowered the camera to an angle where his eyes are likely to be. The items are clearly visible against the grass in the top photo below to a human. However, note how the yellow items, even the one closest to the camera, wash out in the dog’s eyes against the green grass on a sunny day!

Blue/yellow colour discrimination outdoors on a sunny day in the garden

But I tend to do concepts work indoors. I have beige carpets and I used to place items directly on the carpet. The first pair of photos below showed me the reason why Beanie simply never took to “yellow” like he did to “blue”! I sometimes use a turqoise towel underneath (mostly for convenience to keep food off the carpet!). This is the second pair of photos below. Although it makes yellow appear starker, I was surprised by how similar turqoise appeared to blue. Finally the black towel underneath provides most contrast against all blue and yellow objects — third photo pair below.

Beanie learnt the cue “yellow” as a colour label in one session when I put the black towel underneath. He was not being obtuse (he never is), blue objects were simply far easier to pick and Beanie is the kind of gambler would rather make any choice than no choice at all.

So what now?

I wish more people saw the world from their dog’s eyes. Literally. Especially those who do colour discrimination work with their dogs. I see plenty of videos where dogs supposedly tell apart colours that scientists believe dogs simply cannot differentiate. This undermines the credibility of concept work in general. When you do concept work with your dog, understand the limitations (and strengths!) of their perception beyond the bare bones of whether or not a certain colour is visible to your dog.

Think about the environment you’re in, the colour of your walls, the lighting above your training area, the glare from shiny objects you might choose and the background against which dogs are making their choices during training.

Your dog is never stupid. It is up to you to set them up to be the best they can be.

Caveats — other influences and limitations of simulation technology

Dogs do not live in a world of vision alone. Beanie does not need to see the yellow lid on the floor to tell it apart from the blue lid if the two smell any different at all. Also bear in mind that handler bias has a phenomenal influence on how they perform.

The Dog Vision HD app is technology that makes science accessible to the average dog person and makes it possible for us to look through a dog’s eyes. Having used it out and about on our walks, I am not entirely convinced of its accuracy. Nor does the app description inspire confidence, as it states that dogs have three types of cones when it is currently accepted that they only have two. The app doesn’t take into account variations between breeds. But until another app outperforms it in terms of simulating as many dimensions of dog vision as possible, I will continue to use it as a tool in my kit to understand why my dog might make the choices he does. If anyone has a better recommendation, I would love to hear from you.