BBC Atricle: 'Two simple rules' explain sheepdog behaviour

[GentleLake]

http://www.bbc.com/news/science-environment-28936251

[Maxi]

Well duh!

 

Maybe the article doesn't really reflect the actual research but IMO it does seem to be saying the equivalent of 'scientists have worked out that in order to boil an egg you need to put it in boiling water for 4 minutes'

 

I'm not surprised that the reporter managed to find 'sceptical shepherds.' ..

[mum24dog]

I don't think we can draw conclusions about the purpose of a piece of research and its results from a tiny and highly edited news snippet usually written by an amateur.

 

It's always the same.

[Donald McCaig]

Dear Doggers,

It's given us scientific terminology "selfish herd theory" for what has been known since David's time. One of these days perhaps they'll teach us something.

 

Donald McCaig

[mum24dog]

One day you might be grateful for those with enquiring minds. History is littered with people who thought they "knew".

Anyone up for human sacrifice to placate the Icelandic fire gods?

 

Seriously though, do you really believe that what you read in the papers or on line is all there is to know?

 

And do you never have the slightest interest in finding out why what you believe to be true does appear to be the case on the basis of available knowledge?

[Mark Billadeau]

Don't read the spin a popular press writer puts on published science; their end goal is to increase readership in order to sell more advertising not to boil down into layman's terms what the published science said. Read the published science for yourself.

 

The end goal of the research wasn't on figuring out how dogs move sheep but how to mathmatically define how it moves to enable writing computer instuctions for robots. This is from the discussion section of the published research article:

 

Our approach should support efficient designs for herding autonomous, interacting agents in a variety of contexts. Obvious cases are robot-assisted herding of livestock [20], and keeping animals away from sensitive areas [21], but applications range from control of flocking robots to cleaning up of environments and human crowd control.

 

My question is what will they do to the robots to instill a fear of the robots in those to be controlled?

What will they give to robots in place of a grip?

What will they give to robots that human crowds will fear?

[KnottyClarence]

or cheer

[CMP]

or cheer

 

This.

 

 

A stockdog cannot be replaced. It is not possible, even with all the amazing things they can do with neural nets an what not, to replicate what is hard coded on the DNA.

 

Stockdogs, the moment they are born, know more about the business of herding sheep than any one person will learn in a lifetime.

 

And as far as I know, the dogs, so far, ain't telling.

 

Observation will only get you so far - it will only define actions and action sequences - there is no way of understanding WHY and until dogs learn to talk, this will remain true.

 

But we're human and driven to understand that which is before us. Thank the gods. Also, thank the gods that we will never fully understand everything. Imagine what a total mess we could make of things then!

[juliepoudrier]

I agree with Mark. My first thought on reading about crowd control was what would the robot have that would incite enough fear/concern in the humans that they would move away from it. While the algorithms/programming could recreate the movements of a sheepdog, they can't recreate the predator-prey relationship (the intangible?) that is also a big part of the equation.

 

I can, however, picture that back-and-forth movement being used to push an oil/chemical spill into a more central location for clean up, but in that case, of course, what's being "herded" by the robot is inanimate, so replicating the motions without taking into account the ability to recognize and react to the *intention* of an animal would make sense (to me anyway).

 

Now I guess I'll have to read the original article, if it's open access.

 

J.

[Pam Wolf]

Cylon sounds in the background

[Shetlander]

Cylon sounds in the background

Now I'm scared. Those metal dudes would be terrifying. :-)

[juliepoudrier]

This is from the discussion of that article:

 

The plausibility of our model relies upon two assumptions. The first is that the dog can estimate the space between the sheep, irrespective of their metric distance. This seems reasonable given the border collie, a classic sheepdog breed, is said to use a direct stare to herd the flock [29], and similar heuristics-based models have proved useful in understanding the behaviour of pedestrians in crowds [33]. Nevertheless, it would be useful to gather further evidence using, for example, eye-tracking systems to determine shifts in the dog's visual attention. Also, in our field experiments, we used an experienced dog which was given minimal direction. It would be interesting to conduct experiments with multiple dogs and owners of dogs of varied abilities; this way, we would be begin to investigate the role of task familiarity and learning in herding performance.

 

It would be interesting to see how different dogs with different skill levels might change their assumptions about how a dog gathers and moves a group. I would be especially interested in whether a loose-eyed dog would change the data at all, and if so, how that would affect the study's conclusions (or development of programming for robots).

 

In this study, the dog used was a working kelpie. As I was looking at the figure of the tracking information of the dog and sheep what struck me was that the dog didn't appear to be moving in an arc around the sheep but rather in a straighter line toward them, presumably because as the dog approached, the sheep moved toward the flock and so going out and around wasn't actually necessary. Unfortunately I can't view videos at work, so I couldn't see what the actual movements looked like.

 

It really is rather fascinating stuff, if you can get past all the math (ugh!). They did mention predator-prey interactions vs. not splitting one off but gathering instead.

 

It's worth reading that article, even if you skim over the deeper mathematical discussions.

 

J.

[Pippin's person]

I also think it's really interesting to have asked the (sub)question "what does the dog see" (eta: which seems to mean what does it focus on) and to have suggested an answer that the dog sees gaps rather than sheep. They didn't ask questions about how the dog does things other than move a mob of sheep (the study only focuses on "collecting" and "driving", so not "shedding") and as far as I can tell they infer what the dog sees from the GPS data gleaned from the sheep and dog movement, which is an indirect measure (though note in the discussion that an eye-tracking experiment would be necessary to confirm this). It's also pretty interesting that their models don't match behavior with either smaller or much larger groups of "autonomous interacting agents"

 

I don't think this is actually an answer that has been known "since David's time" and that makes it both interesting and likely to spark further study. The question of actual interaction among these three species and the cognitive work going on is also fully open (like Julie says, what of dogs with different skill levels or handlers of different skill levels?). Having worked a mob of crazy lambs and then a mob of dog-familiar adults last night, I'd imagine you'd need different models to capture the properties of different types of "autonomous interacting agents"

[Root Beer]

 

"At the beginning we had lots of different ideas. We started out looking from a birds eye view, but then we realised we needed to see what the dog sees. It sees white, fluffy things. If there are gaps between them or the gaps get bigger, the dogs needs to bring them together.

 

That is an intriguing idea - to be able to see what the dog sees while working stock. I think it would be interesting from a purely observational standpoint.

 

I remembers seeing a video once of a Crufts Agility dog who ran with a camera strapped to his or her head. The perspective of the dog was fascinating. As the handler I had never considered what the dog was experiencing during a run.

 

And yes, there are limits as to what we can know about what is going on in the mind of the dog from such a perspective, but the Agility video was an eye opener to me.

 

Personally I'd love to see stockwork from this particular vantage point (from the visual perspective of the dog).

[Mark Billadeau]

So who would like to pitch to Discovery "Stock Dog Week"?

 

The folks that did shark cam can do stockdog cam. GPS tagging will be easier and they won't have to worry about being eaten by the study subjects.

[GentleLake]

they won't have to worry about being eaten by the study subjects.

 

Unless maybe one of the subjects is Donald's Fly? (JK)

[airbear]

Here's a different perspective, courtesy of a drone camera.

 

https://www.youtube.com/watch?v=RhdXHN7Laak&feature=youtu.be

[Pippin's person]

Oh my, that was a treat. Thanks for posting that, Kristi!

[CMP]

 

I don't think this is actually an answer that has been known "since David's time" ...

In the interest of clarity and claiming geezer rights: If I am not mistaken, Mr. McCraig's reference was to the specific phrase "selfish herd theory" (which is essentially that sheep will drive to the middle of the bunch to save themselves - thus creating an ever tightening herd as the remnants on the edges are pressured to act on this instinct) and not to the study as a whole. I would venture to say that yes, indeed, this particular behaviour of sheep has been known since David's time.

 

Edited because my ability to generate matching tense in a single sentence leaves a great deal to be desired.

[CMP]

Here's a different perspective, courtesy of a drone camera.

 

 

That was a great watch. Thanks for the post!

[Pippin's person]

In the interest of clarity and claiming geezer rights: If I am not mistaken, Mr. McCraig's reference was to the specific phrase "selfish herd theory" (which is essentially that sheep will drive to the middle of the bunch to save themselves - thus creating an ever tightening herd as the remnants on the edges are pressured to act on this instinct) and not to the study as a whole. I would venture to say that yes, indeed, this particular behaviour of sheep has been known since David's time.

 

Edited because my ability to generate matching tense in a single sentence leaves a great deal to be desired.

 

O.k. if we're going there.....the article didn't propose the "selfish herd theory" in the first place so also didn't "give us scientific terminology for what has been known since David's time." That terminology (and the larger biology literature of which it's a part) is used to set up the overall question of the study, which is stated in the article as "the fact that the flock tightens does not tell us how the dog is able to manoeuvre this aggregation and herd the flock towards a specific destination." Just because a behavior has been observed forever doesn't mean that its properties are understood or that there aren't questions worth asking about it, and if you're going to dismiss a study out of hand, it's often a good idea to know first what the study says.

 

It's my opinion from having read the article (albeit quickly) that this is one of those days where "they've taught us something" and perhaps opened up some intriguing further avenues of inquiry.

[CMP]

 

O.k. if we're going there.....the article didn't propose the "selfish herd theory" in the first place so also didn't "give us scientific terminology for what has been known since David's time." That terminology (and the larger biology literature of which it's a part) is used to set up the overall question of the study, which is stated in the article as "the fact that the flock tightens does not tell us how the dog is able to manoeuvre this aggregation and herd the flock towards a specific destination." Just because a behavior has been observed forever doesn't mean that its properties are understood or that there aren't questions worth asking about it, and if you're going to dismiss a study out of hand, it's often a good idea to know first what the study says.

 

It's my opinion from having read the article (albeit quickly) that this is one of those days where "they've taught us something" and perhaps opened up some intriguing further avenues of inquiry.

Fair enough You're certainly entitled to disagree, I just thought there had been a mis-read, but I was mistaken. Mea culpa.

[Maxi]

In the dim and distant past, when I was taught to read a scientific research paper, I was always told to focus on the methodology (what the researchers did) and results (what the researchers found out) . The introduction was the authors opportunity to set the scene and explain why the question is important, relevant and timely. The discussion gives the authors the opportunity to put their results into context with what is already known and also allows them some freedom to extrapolate of where their results may go (including any future use).

Now I’m not a mathematician, but I do know that these kind of models rely on the quality of the data sets that are used to devise & then test the model. (They also rely heavily on the assumptions made).

 

So when reading this research article that describes a mathematical model, the first thing I checked was the ‘biological basis’ of the model i.e.what experiments ("sheep gathers & drives") they did prior to devising the model & then how did they test their model.

 

So firstly, they did not seem to gather any data before devising the model itself- to me this is flaw one.

 

When they tested their model as Julie states in an earlier post (#12) they used a single trained Kelpie, 46 merino sheep and they repeated the exercise (gather & drive) 3 times. To me this equates to a single data point (1 dog & same flock of sheep) repeated 3 times...Although they talk about multiple “N/n” numbers, these seem to relate to the numbers they put into their model rather than to any “real-life” scenario.

 

Not only is there only a single data point, the variability between the 3 repetitions when a real dog actually gathered & then drove the flock seems very large (their figure 5c).

 

So for me this is the second and major flaw in this paper because with such variability in their single biological data point, it must have been relatively easy to fit it into their model.

Although the authors speculate and expand on their model in the discussion, I was always told to be very cautious of the claims made in this part of a scientific paper. In some ways the discussion is the researcher’s equivalent of political spin- based on fact, but very much their own positive interpretation of what may be happening. I am not judging scientists about this, they have to publish in order to survive in academia. Publications are used to help career progression and also get the next pot of funding so it is not surprising that many discussions in research papers sometimes get extrapolated towards ‘blue skies’ thinking

 

Yes, this paper may have "taught us something".,.and perhaps it may lead in the future to something as different from sheepdogs as robotics or managing oil spills. However, personally I just wish the model was based on better quality data because then I could believe that it was saying something valuable and providing some new insight.

Obviously this is just my opinion.

YMMV

[Donald McCaig]

Dear Doggers,

 

Lots of good info. Thanks.

 

Mark cites from the study: "Our approach should support efficient designs for herding autonomous, interacting agents in a variety of contexts. Obvious cases are robot-assisted herding of livestock . . ."

 

Science is one among many human cultural activities and scientists behave from common human motivations and cultural understanding.

 

Question #1: If the dog is herding livestock efficiently - and reproduces itself - why seek a robot?

 

Sometime last year I read a report of Japanese scientists who learned that some dogs could detect forms of bowel cancer with 80% accuracy, long before other diagnostic tools - important in this case because late detection was almost always fatal. The scientists hoped a machine could be designed to replicate what the dogs did.

 

Question #2: Why not bring a few more cancer dogs and handlers into the hospitals?

 

Scienticists, like us, belong to a technophiliac culture and rather than seeking zoological solutions, they seek engineering ones.

 

I find the interaction between sheep, sheepdog and shepherd astonishingly complex. Some humans understand this interaction brilliantly though few are able to articulate their knowledge. That's a loss.

 

Vicki Hearne wrote the seminal Adam's Task to decrease the gap between what top dog trainers knew about dogs and how most people (including most scientists) thought about them. She tried to translate anecdotal knowledge into philosophical knowledge.

 

Patricia McConnell studied how dogs reacted to a range of sounds by going to sheepdog handlers and expanding from her (and their) observations. Hers was a basic study but because it began with and respected what was already known and didn't intend to replace the original with something Samsung might design, it was informative and useful.

 

Only fools and frightened conservatives are anti-science. Scientific inquiry has given us control and luxuries we didn't have. But, like any human endeavor (including sheepdogging) it is constrained by the general culture and the inclinations and (non-scientific) beliefs of the scientific sub-culture.

 

In the early 70's, physicists I knew were delighted by enormous defense department funding for laser research. "They're looking for a death ray," they snickered. They were happy to take the money because they knew lasers couldn't kill people.

 

Today: they look at sheepdogs and dream of robots.

 

Donald McCaig

[Mark Billadeau]

Question 1: Not why seek a robot to replace the dog but what can a dog teach us about building smarter robots.

 

Some statements in research articles are designed to "sell" the potential uses of the research to funding organizations. Few people will fund "I want to mathematically how crowds move"; however, someone may fund "using a mathematical model on how crowds move one can design more efficient transportation systems".

 

 

Question 2: 80% accuracy does not really meet current clinical diagnostic methods (better than nothing for now).

a. I'm sorry for making you go through surgery (or unnecessary invasive testing) on your bowl; there was no cancer. You are one of the 20% of the patients our dog does not diagnose correctly.

b. I'm sorry your cancer is now late stage; I guess our dog missed your cancer during early screening.

 

 

 

If dogs can smell odors produced by cancer than there are chemical markers unique to cancer for which a more accuract test can be developed. Dogs can be used to help guide researchers to those chemicals and a more accurate test.