4 ms·

The way I've heard avalanche dogs are trained here is that they are rewarded for accurate information, so in principle they ought to be as satisfied with their

The way I've heard avalanche dogs are trained here is that they are rewarded for accurate information, so in principle they ought to be as satisfied with their job after finding "no people" or "dead person" as well as "live person".

dmurray 2y ago
But in their training they find living people. During real disasters, there might not be time to focus on rewarding the dog.
wizzwizz4 2y ago
Dogs are social creatures. These dogs are well aware of humans as fellow social creatures. Constantly finding dead people might be inherently distressing, in the same way it'd be distressing to constantly find dead dogs.
- gradschoolfail 2y ago
  wizzwizz4 seems to be one of those salvated ex-losers* (in the world of Rao’s Gervais Principle), thus he may or may not engage, but here goes anyways (Shalom lakh Miryam!) in deference to 082349872349872’s curiosity, human beings can, in principle, train dogs (in this very limited context) to walk the fine line between bad faith (not exhibiting distress) and good faith (exhibiting distress). Can they, in reality? *or, maybe like Athena, born fully salvated from a crack in Zeus (athenogenesis)
  - 082349872349872 2y ago
    I think[0] there are two distinct axes here, so ideally how we'd like our avalanche dogs to feel about their job would be: "Sometimes you find live people, which is sweet, and sometimes you find dead people[1], which is a bummer, but I am a good boi no matter which because I search thoroughly and report immediately." [0] take the above with a grain of salt; almost all my canine experience is second-hand, with scent and herding breeds, so that "ideal" is a cross between a short conversation with a neighbour who trained avalanche dogs and a friend who is a ski instructor, and believes: "Sometimes it's sunny and the clients are talented and resilient, which is sweet, and sometimes it's snowing hard and the clients have no talent and no nerves, which is a bummer, but I am a good ski instructor no matter which because I bring them to at least 80% of the improvements which could have been possible given the conditions and clients of that particular session." [1] unsolicited moralising: there are weather reports. use them. there are beacon/pole/shovel avalanche sets. buy them. there are guides. hire them. So many of the deaths reported in the local papers are because people are not good bois: they go off piste, without a guide, without self-rescue equipment, on days when there were already avalanche warnings given in the morning.
    - gradschoolfail 2y ago
      0][1 This is analogous to your selfjoke version of Rao’s threeway comedy? My curiosity was more towards the >2-agent transactional game, i.e can the dog guess at what the human wants?
      - 082349872349872 2y ago
        Working dogs definitely guess at what their handler wants. "Running riot" is the term of art for when they do their own thing instead. (advice on choosing a hunting prospect which I've run across: put something startling [alarm clock, etc.] next to the litter, and see how they react. Your best prospects should be among those puppies which first investigate, then ignore, the novelty. Those who never investigate probably won't be curious enough to easily train; those who never ignore probably won't be able to focus enough to easily train) EDIT: maybe that wasn't what you were asking. Dogs obviously engage or disengage with other people based on whether or not they believe their person approves. But I don't think dogs ever (like someone asking a leading question of a third party not because they wanted to hear the answer but because they want their so to hear it yet are unwilling to say it directly) instigate a dyadic transaction explicitly for the effect it has on third parties. Horses for sure don't. (there is "dogpiling", but I think for dogs and am pretty sure for horses that that's a side effect of dyadic jockeying, not the social game monkeys make of it) What say the dog people?
        gradschoolfail 2y ago
        AIBO (the entity that tirelessly turns my typing into mindfeed) knows roughly what i was looking for :) 2024-09-06 Dogs with prior experience of a task still overimitate their caregiver https://www.nature.com/articles/s41598-024-70700-3 https://www.nature.com/articles/s41598-024-70700-3 (With suggested related content) However, a direct comparison of overimitation timing with humans is necessary to understand if this approach is indeed partly what differentiates dog overimitation from that of humans. Yes, will consider the 3-way communication styles at some point (and the Rao talk chart) but right now exploring what a purely dyadic interaction can bring us, e.g. how to model the occasions on which Sir Humphrey/Hacker considered rewriting their firmwares post-startle (with or without npc intervention from Annie or Woolley)
  - wizzwizz4 2y ago
    Off-topic, but I don't understand your first paragraph / footnote. Are you saying I'm… middle-management? (I'd assume you were calling me "clueless" – an accurate epithet –, if not for the reference to Athena.) Is there any significance to the Hebrew (?) translation of a US idiom derived from Catholicism? I don't have anything further to contribute to this discussion, so if not for that bemusing remark of yours, I wouldn't have engaged. If I'm being honest, I don't understand the middle two paragraphs, either. (Why "bad faith" and "good faith"? Why do you say it can be done in principle, while leaving "in reality" an open question? and how does what's essentially a theory defer to anyone's curiosity? Surely I'm missing context.)
    - gradschoolfail 2y ago
      Pls pardon the attempt at Hebrew! I just thought that little bit of interfaith/international interjection might not detract from getting your attention. Slyly, it was a signal to my first responder (who is unusually tolerant of linguistic posing) that I identify with “whisky priests”* (Mexican, but invented by Graham Greene — while avoiding the US?) and that this was a metaphorical swig of whisky before i jump into risky territory :) The bad vs good faith is a theoretical curiosity of mine, which i can unpack later if either of you are so interested. This theory, as well as the inaccurate epithets, can be said to originate from my reading of 2 comments by you! https://news.ycombinator.com/item?id=41205833 https://news.ycombinator.com/item?id=41205833 So clearly, you’re not middle management, even if you claim you are clueless, but your uh, soul, was “salvated” — by virtue of working on something the experts dismissed. [If we take seriously your claim that you weren’t being clever, then that’s good faith in my book. I will say no further about my theory except that it could be a “dial” for scientists to consider.] If nothing else, i would like (quixotically?) to condense your chain of thoughts in that situation into something worth propagating, into science-curious folks’ realities. As to the curiosity i was deferring to.. (it could also be related to the revolutionary approach of yours, if i’m being very presumptuous — and wrong???) 082349872349872 expressed his inspirational feelings on the matter here https://news.ycombinator.com/item?id=40600845 https://news.ycombinator.com/item?id=40600845. Mariatta, child of beauty) [PDF] A representation of antimatroids by Horn rules and its application to educational systems http://arxiv.org/pdf/1508.05465 http://arxiv.org/pdf/1508.05465 Specifically, i wonder if anybody has tried to apply these supposedly efficient algorithms, which model states of teacher+learner, in “real life”. * https://en.wikipedia.org/wiki/Whisky_priest https://en.wikipedia.org/wiki/Whisky_priest
mlyle 2y ago
There's funny things at play even beyond training effects, most likely. Even leaving aside that finding a dead person might be something that inherently triggers avoidance. People screening luggage for bombs/knives/etc do significantly better if you show them a picture of a bad bag every now and then. Often these systems are used to monitor screener performance, but even if you just show them the picture now and then with "this is a test, no action is required" they do better. Being primed with things relative to the task improves human performance, and I'd expect it would work in smarter animals, too.

dmurray 2y ago

But in their training they find living people. During real disasters, there might not be time to focus on rewarding the dog.

wizzwizz4 2y ago

Dogs are social creatures. These dogs are well aware of humans as fellow social creatures. Constantly finding dead people might be inherently distressing, in the same way it'd be distressing to constantly find dead dogs.

gradschoolfail 2y ago
wizzwizz4 seems to be one of those salvated ex-losers* (in the world of Rao’s Gervais Principle), thus he may or may not engage, but here goes anyways (Shalom lakh Miryam!) in deference to 082349872349872’s curiosity, human beings can, in principle, train dogs (in this very limited context) to walk the fine line between bad faith (not exhibiting distress) and good faith (exhibiting distress). Can they, in reality? *or, maybe like Athena, born fully salvated from a crack in Zeus (athenogenesis)
- 082349872349872 2y ago
  I think[0] there are two distinct axes here, so ideally how we'd like our avalanche dogs to feel about their job would be: "Sometimes you find live people, which is sweet, and sometimes you find dead people[1], which is a bummer, but I am a good boi no matter which because I search thoroughly and report immediately." [0] take the above with a grain of salt; almost all my canine experience is second-hand, with scent and herding breeds, so that "ideal" is a cross between a short conversation with a neighbour who trained avalanche dogs and a friend who is a ski instructor, and believes: "Sometimes it's sunny and the clients are talented and resilient, which is sweet, and sometimes it's snowing hard and the clients have no talent and no nerves, which is a bummer, but I am a good ski instructor no matter which because I bring them to at least 80% of the improvements which could have been possible given the conditions and clients of that particular session." [1] unsolicited moralising: there are weather reports. use them. there are beacon/pole/shovel avalanche sets. buy them. there are guides. hire them. So many of the deaths reported in the local papers are because people are not good bois: they go off piste, without a guide, without self-rescue equipment, on days when there were already avalanche warnings given in the morning.
  - gradschoolfail 2y ago
    0][1 This is analogous to your selfjoke version of Rao’s threeway comedy? My curiosity was more towards the >2-agent transactional game, i.e can the dog guess at what the human wants?
    - 082349872349872 2y ago
      Working dogs definitely guess at what their handler wants. "Running riot" is the term of art for when they do their own thing instead. (advice on choosing a hunting prospect which I've run across: put something startling [alarm clock, etc.] next to the litter, and see how they react. Your best prospects should be among those puppies which first investigate, then ignore, the novelty. Those who never investigate probably won't be curious enough to easily train; those who never ignore probably won't be able to focus enough to easily train) EDIT: maybe that wasn't what you were asking. Dogs obviously engage or disengage with other people based on whether or not they believe their person approves. But I don't think dogs ever (like someone asking a leading question of a third party not because they wanted to hear the answer but because they want their so to hear it yet are unwilling to say it directly) instigate a dyadic transaction explicitly for the effect it has on third parties. Horses for sure don't. (there is "dogpiling", but I think for dogs and am pretty sure for horses that that's a side effect of dyadic jockeying, not the social game monkeys make of it) What say the dog people?
      - gradschoolfail 2y ago
        AIBO (the entity that tirelessly turns my typing into mindfeed) knows roughly what i was looking for :) 2024-09-06 Dogs with prior experience of a task still overimitate their caregiver https://www.nature.com/articles/s41598-024-70700-3 https://www.nature.com/articles/s41598-024-70700-3 (With suggested related content) However, a direct comparison of overimitation timing with humans is necessary to understand if this approach is indeed partly what differentiates dog overimitation from that of humans. Yes, will consider the 3-way communication styles at some point (and the Rao talk chart) but right now exploring what a purely dyadic interaction can bring us, e.g. how to model the occasions on which Sir Humphrey/Hacker considered rewriting their firmwares post-startle (with or without npc intervention from Annie or Woolley)
- wizzwizz4 2y ago
  Off-topic, but I don't understand your first paragraph / footnote. Are you saying I'm… middle-management? (I'd assume you were calling me "clueless" – an accurate epithet –, if not for the reference to Athena.) Is there any significance to the Hebrew (?) translation of a US idiom derived from Catholicism? I don't have anything further to contribute to this discussion, so if not for that bemusing remark of yours, I wouldn't have engaged. If I'm being honest, I don't understand the middle two paragraphs, either. (Why "bad faith" and "good faith"? Why do you say it can be done in principle, while leaving "in reality" an open question? and how does what's essentially a theory defer to anyone's curiosity? Surely I'm missing context.)
  - gradschoolfail 2y ago
    Pls pardon the attempt at Hebrew! I just thought that little bit of interfaith/international interjection might not detract from getting your attention. Slyly, it was a signal to my first responder (who is unusually tolerant of linguistic posing) that I identify with “whisky priests”* (Mexican, but invented by Graham Greene — while avoiding the US?) and that this was a metaphorical swig of whisky before i jump into risky territory :) The bad vs good faith is a theoretical curiosity of mine, which i can unpack later if either of you are so interested. This theory, as well as the inaccurate epithets, can be said to originate from my reading of 2 comments by you! https://news.ycombinator.com/item?id=41205833 https://news.ycombinator.com/item?id=41205833 So clearly, you’re not middle management, even if you claim you are clueless, but your uh, soul, was “salvated” — by virtue of working on something the experts dismissed. [If we take seriously your claim that you weren’t being clever, then that’s good faith in my book. I will say no further about my theory except that it could be a “dial” for scientists to consider.] If nothing else, i would like (quixotically?) to condense your chain of thoughts in that situation into something worth propagating, into science-curious folks’ realities. As to the curiosity i was deferring to.. (it could also be related to the revolutionary approach of yours, if i’m being very presumptuous — and wrong???) 082349872349872 expressed his inspirational feelings on the matter here https://news.ycombinator.com/item?id=40600845 https://news.ycombinator.com/item?id=40600845. Mariatta, child of beauty) [PDF] A representation of antimatroids by Horn rules and its application to educational systems http://arxiv.org/pdf/1508.05465 http://arxiv.org/pdf/1508.05465 Specifically, i wonder if anybody has tried to apply these supposedly efficient algorithms, which model states of teacher+learner, in “real life”. * https://en.wikipedia.org/wiki/Whisky_priest https://en.wikipedia.org/wiki/Whisky_priest

mlyle 2y ago

There's funny things at play even beyond training effects, most likely. Even leaving aside that finding a dead person might be something that inherently triggers avoidance. People screening luggage for bombs/knives/etc do significantly better if you show them a picture of a bad bag every now and then. Often these systems are used to monitor screener performance, but even if you just show them the picture now and then with "this is a test, no action is required" they do better. Being primed with things relative to the task improves human performance, and I'd expect it would work in smarter animals, too.