Box jellyfish demonstrate learning ability | Hacker News Reader

Box jellyfish demonstrate learning ability

nytimes.com

88 pointsmhb75 comments

http://web.archive.org/web/20230923200015/https://www.nytime...

I think centralizing learning into a brain is smarter, if you pardon the pun. Retrieval becomes faster, which is adaptive, and the body can protect it more easily if it is concentrated in one area, which enables internal specialization. If you can't protect the organ, the adaptive response is to make it redundant, rather than specialized.

>I think centralizing learning into a brain is smarter, if you pardon the pun.

even we humans don't really do that entirely. the gestalt of different nervous systems that interconnect between themselves and our brain appear to be responsible for a significant portion of the constellation of traits associated with learning.

Evolution doesn't care what seems smart or efficient; it just cares about what works.

That’s true, but it’s also true that what’s efficient generally works _better_ than what isn’t.

OTOH, a redundant brain is more resilient. These things have been around since well before we decided to have bones (and brains) and that alone proves the worth of this approach.

I doubt we'll ever find a very smart jellyfish, but I hope no box jellyfish decides I am a threat. Or a treat.

It also doesn't care too much about what's better when good enough is sufficient.

If all the neurons of a brain would be distributed uniformly in the entire body, keeping the same connectivity, there would be no difference in the behavioral abilities of any animal (except the reaction speed), even if it would not have any recognizable brain.

There are two reasons why all animals with a great number of neurons concentrate them into ganglia, and eventually into a single big brain.

The first is that when moving most of the neurons close together the total length of the interconnections between them becomes much smaller, so a much less quantity of materials is needed to build them.

The second is that the speed of signal transmission on the interconnections between the neurons is quite low, so when the interconnections become shorter, i.e. the nervous system becomes more concentrated, the propagation times decrease and the speed of reaction of the animal increases.

The very low speed of signal transmission is why all animals have the main brain in the head, close to the most important sensors.

For a humanoid robot, the speed of light for signal transmission makes irrelevant where the brain is located. Instead of having the brain in the head, a better position is in the chest, where it can be better protected. So what is seen in many SciFi movies, where humans completely disable some killer robots by cutting or shooting their head, would be unlikely to work against a correctly designed robot.

I disagree with you but more importantly, so what if you're right. Are you going to design a proper brain for these jellyfish, or advise them to evolve one? What you say is just so totally non-actionable and uninformative as to be entirely useless.

That's a good point. But there are throughput considerations at the boundaries to the things that the brain wants to control. It makes most sense to centralize logic, and to delegate specialized control out to subsystems only where those specializations are relevant. Essentially making interfaces for control of appendages into APIs where the finer control logic is hidden from the main orchestrator. Exceptions exist where extra control at high levels is warranted (e.g. digits).

"The box jellyfish actively hunts its prey (small fish), rather than drifting as do true jellyfish. They are strong swimmers, capable of achieving speeds of up to 1.5 to 2 metres per second or about 4 knots (7.4 km/h; 4.6 mph).[19] and rapidly turning up to 180° in a few bell contractions.[4] Some species are capable to avoid obstacles.[4]" [0]

So wasn't this already, sorta known? If they can do that, then surely they've some notion of what is worth chasing after, and have demonstrated the ability to differentiate between something nourishing (like small fish) from something like an iPhone someone accidentally dropped off of the side of a boat.

[0] https://en.wikipedia.org/wiki/Box_jellyfish#Australia

These need not be learned behaviours. You don't learn to breathe or to not-breathe water, for instance.

Not necessarily. Their nervous system could have simply evolved in such a way that it is hardwired to go after anything that seems like a small fish.

There was a classic study that showed what we humans might intuitively think how brains and vision should work was wrong, at least with frogs - TLDR - their eyes do processing/filtering (one channel - is this thing in field of vision possible prey?) on input before sending some higher level signal to the brain

https://phdcns.sissa.it/publication/what-frogs-eye-tells-fro...

> Going further, the researchers removed visual neurons from the box jellyfish and studied them in a dish. The cells were shown striped images while receiving a small electrical pulse to represent collision. Within about five minutes, the cells started sending the signal that would cause a whole box jellyfish to turn around.

> “It’s amazing to see how fast they learn,” said Jan Bielecki a postdoctoral researcher at the Institute of Physiology at Kiel University in Germany, also an author of the paper.

Could someone please explain this bombshell?

Did visual neurons learn all by themselves in a dish? And how did the researchers know that the visual neurons would interpret "a small electrical pulse" as a collision? (I'm surprised visual neurons "know" what a collision is.)

> interpret "a small electrical pulse" as a collision

They don't interpret as a collision. IIUC current models of instinctive behaviors and learning are built up from the assumptions that

* if a source of event is "beneficial" (from a chemical, physical perspective, eg "food" / energy packet is there) then those cells, neurons, microorganisms are naturally selected which tend to approach / follow / capture that source

* if a source of event is non beneficial (threat / chemical incompatibility etc) then those survive which avoid that source.

Here it doesn't matter that the negative event is a "collision", eventually the organism('s building blocks) learn to avoid it.

I wonder how the mechanisms of learn and infer in box jellys can inform designs of low energy narrow focus ML, like a toaster that associates a specific person with a specific toastiness rather than operating by time.

Honestly, it would be enough if toasters could operate by "toastiness", instead of time, at all. We had that decades ago and then just stopped, I guess?

I have watched too much Technology Connections...

/rant

Just get a Balmuda.

That thing is almost as big as my microwave and could replace halve of an oven. But I guess it does toastiness.

I wish someone would just build a new toaster that works like the sunbeam radiant control ones, which would be useable on 230V (the issue I have with getting an old one).

FYI if you're not Australian ..... box jellyfish is one of the most deadly creatures on earth.

Because of box jellyfish you cannot swim at any beach in northern Australia - you risk death if you swim at any beach anywhere from Gladstone in Queensland upwards.

Same thing with saltwater crocodiles. Anywhere north of Gladstone you take your life in your hands if you swim cause a salty might eat you, literally.

Worst case you get stung by a box jellyfish then eaten by a saltwater crocodile. No swimming!

Stay aground in Oz and all you have to fear is getting talked into drinking goon

On land you'll very likely be bitten by a redback spider, fall over onto the nest of a funnelweb spider which will also bite you and as you roll away in agony you'll roll onto a nest of tiger snakes, death adders, taipans and brown snakes. Leaping up to run away you'll very likely run in a Gympie Gympie tree whose leaves are covered in fine, silica-tipped hairs that release a potent neurotoxin when touched. Australia is a dangerous place.

Best to stay inside programming.

We should demo a mars colony there.

Don't forget to keep an eye out for drop bears too! They're a particularly aggressive species of koala that are known to jump out of trees and attack unsuspecting humans.

I thought they were polar?

bipolar :P

Or the emus. We've gone to war against them and lost.

If you think a tree can't be that bad, the story with Gympie Gympie goes some officer during WW2 did wipe his butt with a leaf, the pain was such that he killed himself.

What the parent doesn't mention in the water alongside the croc and box jelly is the irukandji which is extremely common and extremly dangerous as well while being as small as a fingernail.

With all those things around, Australia is an amazing place, the moral of the story is ask the locals before you do something you'll regret, they are the ones who knows and if you ask before you will be fine

Irukandji are box jellyfish, just tiny small ones.

From wikipedia:

"The first of these jellyfish, Carukia barnesi, was identified in 1964 by Jack Barnes; to prove it was the cause of Irukandji syndrome, he captured the tiny jellyfish and allowed it to sting him, his nine-year-old son and a robust young lifeguard. They all became seriously ill, but survived".

That's some dedication to science.

Australia is where nature makes "NO TRESPASSING" signs.

thought you can wear woman’s stockings and can be immune or was that a myth?

Off the top of my head, it's possibly a partial myth.

Although they do have suits you can wear to protect against Jelly Fish stings. Clearly barrier protection is effective.

So I suppose the myth portion comes from the efficacy of different types of stockings.

Fishnets probably dont do much good.... Super thick ones of a quality material probably do work!

I'm not sure of the sting mechanism of jellyfish and what level of barrier protection would work.

Stinger suits, basically like a full body wetsuit with a hoodie (and even gloves). You can comfortably wear one all day, with the added bonus of full sun protection.

Wearing womens stockings may not protect you from box jellyfish but will certainly make you look nice.

Not a myth. The stinging cells work not by mechanical touch but in reaction to chemicals on skin. Pantyhose apparently prevent the contact with skin which would trigger the stinging cells.

I didn't know crocodiles were so big! https://maps.app.goo.gl/rnNRoK9gCwVBN3LUA

> FYI if you're not Australian ..... box jellyfish is one of the most deadly creatures on earth.

..And sounding even worse now.

You can swim fairly safely outside of Summer/Autumn though advisable to still wear a stinger suit to minimise the risk. And salties are pretty rare even as far north as Cairns.

I spent a few fantastic days snorkling off the beach at Fitzroy Island one December and turns out I should have been more concerned about sharks after a marine researcher was bitten recently at the spot I was casually swimming around.

While box jellyfish are fairly simple creatures when compared to mammals, I’m waiting for the inevitable study “announcing” that animals have their own thoughts, dreams and emotions.

Something people who’ve been around animals will respond with “duh”.

Edit: I clearly wasn't clear, this comment is focussing on the question as applied specifically to box jellyfish, not animals in general.

Emotions (at least the physiological part, not the qualia) wouldn't surprise me at all even for the simplest of neural structures. Might even be a pre-neural thing, given plants have a transmissible chemical signal to make other nearby plants more bitter in response to being eaten.

Dreams and thoughts?

I won't say "no", but I will say that if you can prove it, then we probably have to start caring about the welfare of even very small AI such as OCR systems made by second year university students encountering MNIST and python for the first time.

Not commenting on the ethical implications, but it's fairly obvious that dogs dream. We could tell our dog was chasing squirrels in his sleep by the running and yelping.

Incredibly obvious. One could say we can't prove it, but it's a fact and I don't need it proven scientifically.

Anyone who had been around my dog would know it. I have no idea if all dogs are the same, in that they act out their dreams--but mine does, and it's unmistakable.

I've seen my cat wake up startled, and hissing and fur straight up, because of a nightmare.

She saw us, and knew she was alright.

Most mammals clearly dream, they even do the twitchy thing in their sleep. Has no one done eeg studies or is that not considered proof?

Just looked this up, all mammals and some birds experience REM sleep

They have:

> "What we've basically found is that dogs dream doggy things," Coren said. "So, pointers will point at dream birds, and Dobermans will chase dream burglars. The dream pattern in dogs seems to be very similar to the dream pattern in humans."

https://www.livescience.com/53743-dog-dreams.html

An actual neuron is far more complex than those in an ANN. I really don't think it follows that if an animal with a similar number of neurons or synapses has some quality related to their brain, so does the ANN.

While true, at present I have no reason to think the extra complexity is important in any way that matters.

Why would you think it doesn't matter? It seems rather unlikely that a major organ would consistently retain a whole lot of useless complexity and functionality across a huge range of animals.

The brain is organic and needs to keep itself alive as well as processing data, and it processes and conveys data by alternating between electrical and chemical signals; chemistry in general is a high-noise mess.

As I don't expect conscious experience to be the result of that mess[0], I expect a digital silicon network to be able to have qualia when it's replicating both the structure of a conscious brain and the simplest parts of the behaviour of the neurones in that structure.

[0] but as nobody knows what consciousness is, I can't rule that out.

Not all animals are equal. While some have their own thoughts and dreams, most don't.

Do we have solid evidence or proof of that?

Even fish appear to have excellent memories in some cases, ability to recognize individuals of other aquatic species, and they have exceptional environmental awareness. With I was a kid, just about everyone was certain that fish were no different from worms or fruit flies.

Do we have solid evidence and proof that not all animals are equal? Yeah.

Memory can be separate from self-awareness, and that's the thing we need to try and be aware of.

It's feasible to have thoughts and dreams and also no sense of "I" within them.

Not only feasible but equally valuable. It’s simply how humans orient their thoughts, and we have no evidence that this has intrinsic value over a being which doesn’t do this.

You're making an awful lot of assumptions that contradict our scientific understanding of animals. A being without self-awareness would only have thoughts that are basically instinct, i.e. 'need food', and those thoughts are in no way equally valuable to more complex thoughts.

I’m not positive that’s true though. There’s so much we think about the world that seems to only be an extension of how we experience the world, but can’t be verified at all. Such as the idea that without an “I” as we experience it, there is no meaningful awareness. I’m open to animals having experiences which are different from mine yet still worth something, and generally worth preserving in the way we typically believe we should with, say, a cat or a dog.

> I’m not positive that’s true though.

Is this possible because you are just speculating out of interest, and not as familiar with current research as you could be?

> Such as the idea that without an “I” as we experience it, there is no meaningful awareness.

This is akin to Russel's teapot IMO.

Yes and no. Without self-awareness, those thoughts are just instinctual desires, e.g. 'need food'. Without self-awareness there won't be much going on in any dreams either.

anyone who has ever had a pet goldfish knows that they're more aware than people give them credit for. they recognise their food containers and get visibly excited when it's feeding time.

I used to spearfish and stopped, then began recording videos of fish. The difference in their behaviour was dramatic.

A genus of fish where I live called the Greenling (I’ve seen kelp, rock, and painted greenling) is very curious in my experience, which spans around 5 years of shooting them and 2 years of recording them.

With the speargun out and hunting mode on, they were unfortunately still curious, but far more apprehensive. They’d leave their perches on the rocks earlier to get away or try to hide in camouflage, not moving a muscle. I’d often pick them off of their rocks when they stayed still. However, with a camera, hunting mode off, they will swim circles around me, leave their perches and move around more, rarely staying completely still. They don’t perceive me as an immediate threat, and some (especially kelp greenling) really want to get close and check me out. They will follow me up after dives, check out the sand I’m tossing up, or just reorient casually to watch me.

When a greenling wants to escape, it knows its environment extremely well and it will race strategically to a spot where it doesn’t think you can access it. It doesn’t go for the nearest spot; it goes for a hidey hole that contextually makes sense. They almost never wind up under a rock where I can get them. The same goes for other fish. They go to specific deep hides they’ve already charted out, even if it means going a little further.

I believe this because I hunted the same spots over and over, sometimes camped on specific spots for a week and diving for hours per day. I came to discover that certain hotspots had very particular safe zones that the fish were well aware of. If a fish took off towards certain features, I’d essentially give up the pursuit; they’d wind up way too deep in a recess to shoot or retrieve. I wasted so much time trying to find them before I figured out that they’re way better at hiding than I imagined.

I don’t think it’s possible for fish to be this good at knowing their environment and how to hide without being somewhat clever. They have mental maps, they know what’s trying to get them (not necessarily a speargun specifically, but an animal that requires a deep hide to escape from), and they evaluate when and how to escape contextually.

I’ve watched seals go after greenling and their strategy differs quite a bit. They will simply squeeze under a rock and wait. The seals will rub their noses into the cracks and crevasses, check stuff out, then you can swim around shortly after and the greenling are still out and about. They understood the danger and hid accordingly, then came back out once the seals left.

Learning this is why I stopped shooting them. Fish aren’t just meat torpedos with special sensors in their heads. They seem sentient to me.

Great, as if box jellyfish aren’t bad enough, now it turns out they can think, too.

It will be another 300 million years before it evolves to be like how moneys were

[stub for offtopicness]

Brainless != distributed brain

Ok, we've made the title less brainless above.

This has great possibilities for new and creative insults but surely something is doing the learning and that can be classified as a brain.

Oh, the puns that are to come. Its like God out of the box jellyfish.

Why is this shocking? Evolution itself is a learning system

Evolution is a process that blindly accumulates beneficial mutations, that seems somewhat different from learning.

It’s easy to prove that just because a process is driven by a characteristic doesn’t necessarily mean the output of that process would also. Example: when I grocery shop while hungry, that hunger may cause me to make irrational decisions, but the collection of items I buy aren’t themselves hungry.

Learning is the mostly blind accumulation of knowledge that will hopefully be beneficial (and not detrimental to the excess of that benefit, which is not a given) in the future. Targeted study is the exception, even for humans.

This is also a somewhat better description of evolution than yours because benefit in excess of detriment is not a guarantee and the process is not blind (because animals breed selectively).

Yes and what’s more is that natural selection is a loss minimization function, similar to gradient descent

Evolution is alive? Learning? What exactly is evolution anymore?

I am honestly surprised to see this downvoted on HN. Everything is a learning system in the thermodynamic sense.

Hint: ML is just thermodynamics re-implemented in software.

Reply on news.ycombinator.com