Echopraxia is a loose sequel to it (in fact, I found it disappointing similar; not that it's in any way a bad book, it's just that it's about the same themes).
His Rifters trilogy is online as well, but Blindsight is a lot better --- the first Rifters book is excellent, the middle one is so-so, but I found the last one to be a dead loss.
Will we ever be able to scan the wiring of this spider's neurons and "run it" on a computer? That would be amazing, I think. Then we could create virtual worlds for these virtual spiders and let them evolve, supposedly much faster than "normal" evolution.
This was a very interesting article. Thanks for posting! I have read about the web-plucking behavior before, but had no idea about its path-finding abilities.
Is truly humbling that so much "intelligence" can fit in a raisin-sized animal. I once thought that these tiny brains would soon reveal their secrets to us. I briefly contributed to the effort - poking microelectrodes into crayfish neurons and creating 3D EM models. Now I am pretty sure that I won't live to see much progress. We just don't have the tools to reverse-engineer a supercomputer the size of a pinhead.
600,000 neurons seems small enough to simulate - and microscopes are easily good enough to discern cells (neurons) and their connections. Though the weightings between them probably can't be detected.
At least, that's my layman's view... what specific aspects of tools are actually deficient?
It could be a useful starting point for eventually understanding ourselves. I expect we'll need to develop hierarchical models for how neurons operate, to cope with the complexity... perhaps even new pure mathematics.
The problem isn't the number of neurons and their connections, but the complexity and diversity of them.
There are hundreds of signalling molecules and receptors that haven't been characterised. Some molecules diffuse over long distances, others are taken up and recycled depending on the local environment. Surrounding non-neuronal cells alter the ionic and metabolic states of networks, and can physically block (or sometimes actively transport) neurotransmitters.
Each neuron type has its own response characteristics (activation function) that can completely change depending on the context. Even different locations on the same neuron have different receptors and transmitters, and can alternate between inhibitory or excitatory functions.
To give an example of the difficulty, we've had a complete map of the neural connectivity of C. elegans for 30 years, which consists of only 302 neurons, and we still don't know how many of the cells and circuits function.
I think the only solution will be high-throughput phenotyping. Test cell types and mutants under different circumstances until we have a deep-enough understand of these systems and their interactions to model them.
It takes humans far too long to do experiments. We need more lab automation.
Hey, I'd like to talk with you about automating some of that stuff, but there's no contact info in your profile. My email and IRC contact are in mine if you'd be willing to get in touch.
But knowing that still doesn't bring you that much closer to simulating it. There's a couple a of great papers where they got biologists to attempt to figure out how relatively simple, very well known things worked - one was a radio[1], one was an old Atari chip[2]. So they used biological techniques used to try to figure out how brains worked. For example lesions - they took transistors out of the arcade machine and tested if the game still started up. And if, say, they found that taking a and b out stopped Donkey Kong from working, and x and y stopped space invaders working, then maybe they could make the assumption that a and b were the Donkey Kong transistors, etc. Or if they destroyed tiny parts of the radio circuit board and it still appeared to work, then those parts were junk, unnecessary. But of course that's completely misleading. It doesn't matter how much data they had either.
It's not to say the data that is currently held isn't important; it is, but it doesn't then translate to 'we can model this insanely complex system', because there's so, so much missing from the understanding of how it works
This, and our failure to simulate something as well understood as the c. elegans' 302 neurons, makes me think that biologists need to push a new approach.
When it comes to understanding a radio or a microchip, my first instinct would be to try to isolate components which are as small as possible, and actually try to map inputs to outputs... And if that's not working, either the component isn't small/basic enough, or I haven't really captured all the inputs. I'd hopefully start getting an idea about logic gates and voltage thresholds and eventually building up to larger and more complex processes and parts.
1) Would such an approach work in theory for a radio or chip?
2) I'm assuming I'm not the first person to think of this - is that approach basically what's already being done? (edit: a lot of stuff in this seem to be along those line: https://en.wikipedia.org/wiki/Electrophysiology )
3) Is it just much harder than I imagine to isolate a single neuron (or small set of neurons) and try some inputs to come up with the algorithm to properly simulate it?
4) Or are neurons too complex for this approach to be effective, and we need to dig down to more basic components or get a better understanding of inputs?
It's 4. Biological neurons are much more complex than they appear: they're sensitive to many, many different chemical signals, they have internal structure that's non-trivial, and they can have not just excitatory/inhibitory outputs, but modulatory as well. It's hard to know how to even start extracting subcomponents.
Biological systems are a tangled mess where everything interacts with everything else, and anything that goes in also goes out, and the other way around.
> The same species of Portia trapped a few hundred miles away doesn’t show any evidence of seeing the egg sac.
Spiders aren't social... but this sounds like cultural knowledge, passed on from parents or peers. Or perhaps acquired over a number of encounters (especially the given trial and error Portia exhibits).
mmh... they are social in one moment of its lives that could explain this. Mother and young have a strong bound for a while. I'm just wildly speculating, but a nymph spider sit in the abdomen of their mother could learn what to chase just seeing its mother chasing. "They born with that skill" is a strong candidate, but not the only possibility here. They need to repeat the experiment from spiders raised since eggs if not done before.
Do you know how long Portia nymphs are connected with their mothers? Seems like a lot of knowledge to gather before they have even had their own feet on the ground, let alone had their first first solo hunt.
Is there any research on whether this local adaptations have a genetic or epigenetic origin? One would expect it to be purely genetic, but I'd be super-cool if the Portia had developed some way to transmit learned adaptations to their descendants.
If a Portia spider has only 600K of neurons it is within range of simulating and testing it with current technology, for example the TrueNorth chip has 1 million neuronlike structures (Spaun, Blue Brain project or the TrueNorth chip) https://en.wikipedia.org/wiki/TrueNorth
It all reminds me a bit of the Vernor Vinge SF novel about intelligent spiders:
"The planet's inhabitants, called "Spiders" by the humans for their resemblance to arachnids, have reached a stage of technological development very similar to that of Earth's humans in the early 20th century, although humans believe that they may once have been capable of space travel."
Is it a coincidence that Spiders and Octopi have 2^3 legs and are more intelligent then we expected them to be?
It's not just the number of neurons that makes brains so complex, it is also the far larger number of connections. The average brain cell is connected to 1000 up to 10000 other cells. So for 600K neurons the simulation needs to simulate something between 600 and 6000 million connections.
moreover every connection is not a simple fixed function * transfer weight either, they have varying activation levels, can be inhibited or self stimulate and have all sort of behaviors related to fatiguing and the presence or absence of certain chemicals around them
Yes, I've yet to read the third installment of the series, the "Slow Zone of the Zones of Thought" barrier of space travel provides an interesting setting.
To me (as a complete layman), this is the most interesting topic in biology, the 21st century's origin of species topic. Intelligence in biology.
For species like this spider or octupi where our last common ancestor is so early, it begs the question is how much of the evolution of these minds is a product of convergent evolution. That would be a partial answer to one of the Drake Equation's components.
Biological Intelligence is such a big question and it goes to the core of discovering what we are.
Some sort of middling-high degree of intelligence seems very likely, because we have it in multiple lines on this planet (octopus, multiple bird lines, whales and dolphins, etc.). It is less clear that the next step to a technological intelligence is very popular; I can't prove it but I seriously doubt all this middling-high intelligence developed in the last ~2 million years during humanity's ascent. Very likely many of those middling-high intelligences were around for millions of years before hominids began their intelligence explosion (with the hominids themselves being another "middling-high" intelligence for who knows how long before then). It's not out of the question that many dinosaurs or other things extinct for millions of years were middling-high intelligence too.
We can't get great data on this, because it is also likely that the first technological species precludes the natural development of any future one at least concurrently, and possibly for extended periods of time. (i.e., even if humanity accidentally designs its replacement, whatever form that may take, that won't be relevant data for the Drake equation, or we may "pollute" the genetics of the planet with a lot more intelligence than could ever naturally develop, causing intelligent species to start popping up every few hundred thousand years all over the place even if they all kill themselves off if our most cynical commentators are to be believed).
We do have good reason to believe intelligence is really quite expensive. We humans pay a lot for it, if you really get into the biology of it. It may be the case there's a quite significant gap between "middling high" and the technological intelligence the Drake equation is about, and having only one sample that said "somebody made it" tells us not much, since we are the observer observing that fact.
> And then there is the realisation that this is a population-specific, not species-specific, trait! It is a bit of locally acquired genetic knowledge.
Some human populations "know" how to digest milk, and some don't. Non species-specific. It's a bit of locally acquired "genetic knowledge" as well.
Because humans that can consume milk are actively expressing genes that process lactose. Those genes are expressed in mammalian young, and adult humans who are lactose tolerant.
To say that adult lactose digestion is similar to Portia's egg sac spotting abilities is to say that all spider's have an egg sac spotting gene and it's only expressed in the Portia species in a specific geographical area.
Maybe, but then we have an issue with the rest of spiders. If spiders have an egg sac spotting gene, is it inactive during certain times?
Actually you may be onto something, being able to spot an egg sac would be useful for a male to determine the receptiveness of a female to mating. Maybe the Portia spiders in the specific area have a mutation so they are always looking for egg secs, instead of just when a male is looking for a mate.
You're missing my point, maybe because I veered off into musing on how an egg sac detection trait could be a gene.
I'm trying to say that the development of what almost looks like a learned behavior but is possibly a geographically isolated gene is more surprising than the modification of a gene so that it is expressed past childhood.
We see the latter all the time in other forms if you look at domesticated animals as we have selected for the cute and cuddly traits for adult animals that are normally present in infant forms.
In fact is more complex than the article says, because Portia do not only eat other spiders. They also steal preys of other spiders and also eat an appreciable amount of spider eggs; egg sacs are a resource. Therefore "spiders with egg sacs can't attack me" is only one of the explanations. Another could be: "spiders with egg sacs worth the extra effort because is two meals for the prize of one", "spiders with egg sacs are starving and are easier to attack", and even "that spider just have some kind of round prey in its mouth currently, so should be well feed and have enough fat to create my own egg sac".
To claim that "that animal is thinking X" you need to discard or at least discuss all possible alternatives. Is very "annoying" when you design a experiment, because maybe you just do not have the money to cover other interpretations, and some kind of experiments have more possibilities to be funded than other.
> However being so small, there is a trade-off in that Portia can only focus its eyes on a tiny spot. It has to build up a picture of the world by scanning almost pixel by pixel across the visual scene.
This is fascinating. I wonder if the limited visual focus plays a similar role as consciousness in higher animals. It enables the brain to work on larger problems by focusing on one small Problem at a time.
I think that there's a common thread that underlies the development of intelligence in both humans and Portia spiders. We both evolved in an environment where we were at a sensory and physical disadvantage to our prey. We both relied more on understanding the behavior of our prey in order to hunt. In the case of early humans, we adopted a style of hunting known as persistence hunting.
Persistence hunting ( https://en.wikipedia.org/wiki/Persistence_hunting )is a general hunting strategy in which a hunter chases their prey over a very long distance(15-30 miles). Eventually the prey becomes weak and succumbs to exhaustion. On its own, that's not very special -- both dogs and hyenas adopt a very similar hunting strategy. The key difference is that humans lack the extremely sensitive sensory abilities that are used by other persistence hunters.
In the absence of such senses, early hominids had to predict where prey would go and accurately pursue them over very long distances based on very small amounts of visual evidence. In short, early hominids hunted by simulating the minds of their prey. If you look at https://www.youtube.com/watch?v=826HMLoiE_o (a documentary on persistence hunting), you can watch some Kudu tribesmen literally simulating where an antelope will go.
It certainly seems to me like lucking into an evolutionary niche where you get caloric benefits that are directly linked to how well you can simulate the fairly-complicated minds of your prey is pretty much a recipe for extreme selective pressure in favor of general intelligence.
The story is a little different in the case of Portia, given that they are apparently at a visual advantage to their prey. However, I think the examples mentioned in the article make a strong case that Portia's comparative hunting advantage is in planning the best attack method based on the behavior of a given type of prey.
What tools do we currently have to map neural signaling and structure?
Are there ways to automate the measurements on a large scale?
Are we able to feed digitally modeled neural signals mimicking real signals to live neural structures - ie. feed complex input signals and record the outputs?
56 comments
[ 370 ms ] story [ 1758 ms ] thread1. http://rifters.com/real/2009/01/iterating-towards-bethlehem....
http://www.rifters.com/real/Blindsight.htm
Echopraxia is a loose sequel to it (in fact, I found it disappointing similar; not that it's in any way a bad book, it's just that it's about the same themes).
His Rifters trilogy is online as well, but Blindsight is a lot better --- the first Rifters book is excellent, the middle one is so-so, but I found the last one to be a dead loss.
http://www.artificialbrains.com/openworm
https://neurokernel.github.io/
At least, that's my layman's view... what specific aspects of tools are actually deficient?
It could be a useful starting point for eventually understanding ourselves. I expect we'll need to develop hierarchical models for how neurons operate, to cope with the complexity... perhaps even new pure mathematics.
But it isn't a far cry from getting to 600,000.
There are hundreds of signalling molecules and receptors that haven't been characterised. Some molecules diffuse over long distances, others are taken up and recycled depending on the local environment. Surrounding non-neuronal cells alter the ionic and metabolic states of networks, and can physically block (or sometimes actively transport) neurotransmitters.
Each neuron type has its own response characteristics (activation function) that can completely change depending on the context. Even different locations on the same neuron have different receptors and transmitters, and can alternate between inhibitory or excitatory functions.
To give an example of the difficulty, we've had a complete map of the neural connectivity of C. elegans for 30 years, which consists of only 302 neurons, and we still don't know how many of the cells and circuits function.
I think the only solution will be high-throughput phenotyping. Test cell types and mutants under different circumstances until we have a deep-enough understand of these systems and their interactions to model them.
It takes humans far too long to do experiments. We need more lab automation.
One of the most promising things I've seen is the Emerald Cloud Lab, but it's still a long way off from what's needed:
http://emeraldcloudlab.com/
It's not to say the data that is currently held isn't important; it is, but it doesn't then translate to 'we can model this insanely complex system', because there's so, so much missing from the understanding of how it works
[1] https://www.cmu.edu/biolphys/deserno/pdf/can_a_biologist_fix... [2] http://biorxiv.org/content/biorxiv/early/2016/05/26/055624.f...
It's not.
When it comes to understanding a radio or a microchip, my first instinct would be to try to isolate components which are as small as possible, and actually try to map inputs to outputs... And if that's not working, either the component isn't small/basic enough, or I haven't really captured all the inputs. I'd hopefully start getting an idea about logic gates and voltage thresholds and eventually building up to larger and more complex processes and parts.
1) Would such an approach work in theory for a radio or chip?
2) I'm assuming I'm not the first person to think of this - is that approach basically what's already being done? (edit: a lot of stuff in this seem to be along those line: https://en.wikipedia.org/wiki/Electrophysiology )
3) Is it just much harder than I imagine to isolate a single neuron (or small set of neurons) and try some inputs to come up with the algorithm to properly simulate it?
4) Or are neurons too complex for this approach to be effective, and we need to dig down to more basic components or get a better understanding of inputs?
Spiders aren't social... but this sounds like cultural knowledge, passed on from parents or peers. Or perhaps acquired over a number of encounters (especially the given trial and error Portia exhibits).
I'd agree. A very generic visual learning apparatus seems less likely than a specialized visual skill to fit into 600k neurons.
The reason it's at least plausible is that epigenetics seem to be involved in the memory formation process (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3549063/)
Perhaps it can already be done with off the shelf hardware and software: https://www.neuron.yale.edu/phpBB/ http://www.nengo.ca/ http://www.nest-simulator.org/
It all reminds me a bit of the Vernor Vinge SF novel about intelligent spiders:
"The planet's inhabitants, called "Spiders" by the humans for their resemblance to arachnids, have reached a stage of technological development very similar to that of Earth's humans in the early 20th century, although humans believe that they may once have been capable of space travel."
Is it a coincidence that Spiders and Octopi have 2^3 legs and are more intelligent then we expected them to be?
For species like this spider or octupi where our last common ancestor is so early, it begs the question is how much of the evolution of these minds is a product of convergent evolution. That would be a partial answer to one of the Drake Equation's components.
Biological Intelligence is such a big question and it goes to the core of discovering what we are.
We can't get great data on this, because it is also likely that the first technological species precludes the natural development of any future one at least concurrently, and possibly for extended periods of time. (i.e., even if humanity accidentally designs its replacement, whatever form that may take, that won't be relevant data for the Drake equation, or we may "pollute" the genetics of the planet with a lot more intelligence than could ever naturally develop, causing intelligent species to start popping up every few hundred thousand years all over the place even if they all kill themselves off if our most cynical commentators are to be believed).
We do have good reason to believe intelligence is really quite expensive. We humans pay a lot for it, if you really get into the biology of it. It may be the case there's a quite significant gap between "middling high" and the technological intelligence the Drake equation is about, and having only one sample that said "somebody made it" tells us not much, since we are the observer observing that fact.
Some human populations "know" how to digest milk, and some don't. Non species-specific. It's a bit of locally acquired "genetic knowledge" as well.
What's surprising here?
What differs is whether we stop digesting milk, not whether we know how in the first place.
All spiders can see eggs in the fangs of their opponents. What differs is whether they react to it.
To say that adult lactose digestion is similar to Portia's egg sac spotting abilities is to say that all spider's have an egg sac spotting gene and it's only expressed in the Portia species in a specific geographical area.
Maybe, but then we have an issue with the rest of spiders. If spiders have an egg sac spotting gene, is it inactive during certain times?
Actually you may be onto something, being able to spot an egg sac would be useful for a male to determine the receptiveness of a female to mating. Maybe the Portia spiders in the specific area have a mutation so they are always looking for egg secs, instead of just when a male is looking for a mate.
So, different populations of same species are expected to have evolved different subtle traits.
I don't think that Portia too carry eggs in their fangs. Nothing points to that as I recall.
I'm trying to say that the development of what almost looks like a learned behavior but is possibly a geographically isolated gene is more surprising than the modification of a gene so that it is expressed past childhood.
We see the latter all the time in other forms if you look at domesticated animals as we have selected for the cute and cuddly traits for adult animals that are normally present in infant forms.
To claim that "that animal is thinking X" you need to discard or at least discuss all possible alternatives. Is very "annoying" when you design a experiment, because maybe you just do not have the money to cover other interpretations, and some kind of experiments have more possibilities to be funded than other.
This is fascinating. I wonder if the limited visual focus plays a similar role as consciousness in higher animals. It enables the brain to work on larger problems by focusing on one small Problem at a time.
(Of course, consciousness is more than that...)
Persistence hunting ( https://en.wikipedia.org/wiki/Persistence_hunting )is a general hunting strategy in which a hunter chases their prey over a very long distance(15-30 miles). Eventually the prey becomes weak and succumbs to exhaustion. On its own, that's not very special -- both dogs and hyenas adopt a very similar hunting strategy. The key difference is that humans lack the extremely sensitive sensory abilities that are used by other persistence hunters.
In the absence of such senses, early hominids had to predict where prey would go and accurately pursue them over very long distances based on very small amounts of visual evidence. In short, early hominids hunted by simulating the minds of their prey. If you look at https://www.youtube.com/watch?v=826HMLoiE_o (a documentary on persistence hunting), you can watch some Kudu tribesmen literally simulating where an antelope will go.
It certainly seems to me like lucking into an evolutionary niche where you get caloric benefits that are directly linked to how well you can simulate the fairly-complicated minds of your prey is pretty much a recipe for extreme selective pressure in favor of general intelligence. The story is a little different in the case of Portia, given that they are apparently at a visual advantage to their prey. However, I think the examples mentioned in the article make a strong case that Portia's comparative hunting advantage is in planning the best attack method based on the behavior of a given type of prey.
"Jumping spiders already have excellent vision and Portia’s is ten times as good, making it sharper than most mammals."
"The story is a little different in the case of Portia, given that they are apparently at a visual advantage to their prey. "
What tools do we currently have to map neural signaling and structure?
Are there ways to automate the measurements on a large scale?
Are we able to feed digitally modeled neural signals mimicking real signals to live neural structures - ie. feed complex input signals and record the outputs?
There's a project to simulate a nematode worm - http://www.openworm.org/
A bit more speculative but super interesting - http://www.fhi.ox.ac.uk/brain-emulation-roadmap-report.pdf
https://www.sciencedaily.com/releases/2013/03/130319091256.h...