My experience with turtles is confined to reef diving.
Turtles are unusual in that they are extremely tolerant of hypoxia. It may be that they have adapted to a relatively shallow network that doesn't parse much, but is pretty good at extracting the most meaningful features (oxygen/energy efficiency favored over information efficiency)
I wonder if this patterns persists for other reptiles.
I think it's unrelated. You could say dolphins and whales are "tolerant of hypoxia", yet from their behavior and the layout of their brains it is evident that evolution has not limited "information efficiency" in exchange for "tolerance to hypoxia".
> I think it's unrelated. You could say dolphins and whales are "tolerant of hypoxia", yet from their behavior and the layout of their brains it is evident that evolution has not limited "information efficiency" in exchange for "tolerance to hypoxia".
Not all species will make the same tradeoffs or converge on the same solutions, though.
Usually though if a tradeoff doesn't exist then it doesn't happen. In this case, most likely it's not a tradeoff that's going on it's just parallel lines of evolution.
It is fascinating. But I always get uncomfortable reading stuff like this... I mean, they are placing sensors in a turtle’s brain, physically immobilizing it, aiming its head at a screen for a while, and you generally don’t release something to the wild after that. The ethical thing is considered to be killing it.
I get that there are a lot of aspects to modern life that sterilize the gruesomeness of reality. Still makes me uncomfortable when the illusion is peeled back and I catch a glimpse.
Agreed. To put into another context if someone were to try this on my Boxer dog , an arguably similar entity, I would protect her with my life if need be. Given that while I am fascinated with such information I will have to come down on the side of the turtle.
It is possible to eat meat and to value and respect animals. It requires the holding of two paradigms simultaneously. These paradigms are not incompatible or necessarily in conflict with each other.
Paradigm 1. Animals are awesome. Wanton killing is wrong.
Paradigm 2. Killing animals for food is OK provided the animal is killed humanely, the food is not wasted, and the animal is revered for its contribution.
Most people are able to do this. Some of them get lazy with the part about reverence, because the supermarket puts animals and the life cycle process out of sight and out of mind, hence there are few triggers for reverence.
Similarly, it is possible to use animals for scientific research and to value and respect animals. It requires the holding of two paradigms simultaneously. These paradigms are more difficult to state than in the case of carnivorous food above.
If that’s your response then what’s the point of trying to rationalize your meat eating? If you like meat that comes from animals that are unnecessarily killed then that’s your prerogative, but don’t try to dress it up in spiritual terms like “reverence”.
Animal rights activists regularly target both. They even go as far as doing sit ins in slaughter facilities to show how gruesome and cruel they are to the animals.
I don't think it's proportional. There should be way more on the animal farming industry and way less on scientific research. But that's just my opinion.
Not quite. Nowhere in those paradigms is the "pretty objects" part implied. But it does put some version of "I'm hungry" as passing the "wanton" threshold for killing something "awesome".
Which in my opinion is the real weakness of this proposal. You won't find me destroying something I consider awesome unless I'm actually starving.
Pitts is probably known to HN readership for his early work on neural networks which put him in dialogue with people like Claude Shannon. It's fascinating to see how much that early world of 1940s/50s era computer researchers overlapped with people literally cutting into animal brains (not to mention experimenting on their own via psychedelics and the like).
Colleague of mine was working on research of vision system of a spitting fish. The fish hunts by spitting water at insects sitting on the leaves above the water surface. What they found out that "firing solution" is formed in the retina and passed to the brain to initiate the spit.
I'd be interested to see if the LGN of the turtle is performing distinct functions to those of mammals. LGN is a super interesting component of the visual system - e.g. iirc people with dyslexia have morphologically distinct LGNs.
hey, i dont know much about turtology/tortology, nor do i know much about the eye other than i need glasses because my eyes are a weird shape. could somebody ELI5 please. for us plebs.
I'm also a layperson but my take was that the turtle visual cortex doesn't carry individual 'pixel' signals (and low-level features) like ours does. Instead it immediately turns the pixel signals into a high-level holistic flow of perceptions.
Disclaimer: this is all _very_ rough and approximate - my only qualifications are that I hung out with neuroscientists in grad school.
You can largely think of mammalian eyeballs as camera sensors, which see the world and pass lightly pre-processed pixel data through cables (your optic nerves) into the back part of your brain where your visual cortex "starts". That pixel-level data flows through a kind of pipeline of filters which detect increasingly complex features the further forward they are in the brain.
The first few layers are detecting fairly understandable things - like little lines bits, and corners that turn left, or turn right, etc. The downstream layers take those detections as inputs to detect more complex features, e.g. helping to detect larger closed shapes and determine which pixels are inside vs. outside those shapes. This data then feeds into higher level object detection, etc etc.
What's really cool is that the first few layers of this are more or less laid out like an image in RAM. A star pattern of shapes in the real world would cause a star pattern of neurons in your head to start firing. This sensible layout has really helped neuroscientists grasp what the brain is doing, and has IMO been one of the big reasons we've been able to take such inspiration from neuroscience into machine vision.
Apparently though, the turtle brain doesn't work like this at all. As far as I can tell, the turtle visual processing system is just a big weird soup of neurons with no familiar structure to it. It works somehow, and I'm excited to follow the progress of decoding what it's actually doing.
By the way, if any of this interests you I recommend grabbing a copy of David Marr's book "Vision". It's probably pretty out-dated at this point, but it's wonderfully written and I think gets the basic points across very well: https://mitpress.mit.edu/books/vision
37 comments
[ 3.1 ms ] story [ 80.1 ms ] threadTurtles are unusual in that they are extremely tolerant of hypoxia. It may be that they have adapted to a relatively shallow network that doesn't parse much, but is pretty good at extracting the most meaningful features (oxygen/energy efficiency favored over information efficiency)
I wonder if this patterns persists for other reptiles.
Not all species will make the same tradeoffs or converge on the same solutions, though.
I get that there are a lot of aspects to modern life that sterilize the gruesomeness of reality. Still makes me uncomfortable when the illusion is peeled back and I catch a glimpse.
Paradigm 1. Animals are awesome. Wanton killing is wrong.
Paradigm 2. Killing animals for food is OK provided the animal is killed humanely, the food is not wasted, and the animal is revered for its contribution.
Most people are able to do this. Some of them get lazy with the part about reverence, because the supermarket puts animals and the life cycle process out of sight and out of mind, hence there are few triggers for reverence.
Similarly, it is possible to use animals for scientific research and to value and respect animals. It requires the holding of two paradigms simultaneously. These paradigms are more difficult to state than in the case of carnivorous food above.
Those two paradigms can be boiled down to “Animals are pretty objects that I can kill and eat if I so desire.” It’s the polar opposite of reverence.
What you say I think and feel is not an accurate representation of what I think and feel. It is an indication of what you think and feel.
And the majority of animal research is on pre-clinical treatments for human diseases.
I've always been amazed at how animal rights activists target scientists over the animal farming industry.
Which in my opinion is the real weakness of this proposal. You won't find me destroying something I consider awesome unless I'm actually starving.
Pitts is probably known to HN readership for his early work on neural networks which put him in dialogue with people like Claude Shannon. It's fascinating to see how much that early world of 1940s/50s era computer researchers overlapped with people literally cutting into animal brains (not to mention experimenting on their own via psychedelics and the like).
Since they hunt in a school they need to predict where the prey will fall due to competition.
The fish will initiate a swim maneuver after observing the first 75ms of the ballistic path of the prey as it falls.
https://www.cs.bgu.ac.il/~ben-shahar/Publications/2018-Ben_T...
You can largely think of mammalian eyeballs as camera sensors, which see the world and pass lightly pre-processed pixel data through cables (your optic nerves) into the back part of your brain where your visual cortex "starts". That pixel-level data flows through a kind of pipeline of filters which detect increasingly complex features the further forward they are in the brain.
The first few layers are detecting fairly understandable things - like little lines bits, and corners that turn left, or turn right, etc. The downstream layers take those detections as inputs to detect more complex features, e.g. helping to detect larger closed shapes and determine which pixels are inside vs. outside those shapes. This data then feeds into higher level object detection, etc etc.
What's really cool is that the first few layers of this are more or less laid out like an image in RAM. A star pattern of shapes in the real world would cause a star pattern of neurons in your head to start firing. This sensible layout has really helped neuroscientists grasp what the brain is doing, and has IMO been one of the big reasons we've been able to take such inspiration from neuroscience into machine vision.
Apparently though, the turtle brain doesn't work like this at all. As far as I can tell, the turtle visual processing system is just a big weird soup of neurons with no familiar structure to it. It works somehow, and I'm excited to follow the progress of decoding what it's actually doing.
By the way, if any of this interests you I recommend grabbing a copy of David Marr's book "Vision". It's probably pretty out-dated at this point, but it's wonderfully written and I think gets the basic points across very well: https://mitpress.mit.edu/books/vision