Honeybees Seem to Understand the Notion of Zero | Hacker News Reader

Honeybees Seem to Understand the Notion of Zero

npr.org

100 pointsjohnny31348 comments

Some images here hint at the methodology: https://phys.org/news/2018-06-scientists-bees-concept.html The number of ink marks (a discrete quantity) was varied independent of the mass of ink; the bees aren't just gravitating to the least inky paper.

I'm curious to see if they tried to keep the ink surface area of each number the same. If not, maybe the bee's optics just average the card and go for the one that's least filled in? Or at any rate, some way to rank visual complexity rather than ruminate empty sets.

Yeah... I don't see anyway for the study that 'more white' (or whatever the card color was) wasn't what the bees learned. Particularly since this is the actual correlation. "Howard trained one group of bees to understand that sugar water would always be located under the card with the least number of symbols."

EDIT: adonovan linked to an article with images and showing it's not as simple as I imagined.

About submissions of NPR's posts: NPR prompts to accept tracking cookies and the like or go to the plain text version. Unfortunately in the latter case it redirects to the index of the site, that never has a link to the submitted post and there is no search box.

Would fellow Hacker News submitters mind to post links to the text only untracked versions? Thanks.

https://text.npr.org/s.php?sId=617863467

When a cat had a litter of babies on our farm we took her in since it was a cold winter and otherwise they probably wouldn't have made it. She had six kittens and we left her alone with them since new mommy cats don't like to be bothered much during early nursing.

Fast forward a bit right after the little ones started crawling around more and she was getting up to move around, letting them out of her sight etc. One of the kittens crawled into a basket and we picked it up and walked into another room to see if she would notice the missing kitten. She came back and checked around and determined she only had 5 babies and was missing one. We felt terrible and quickly returned the kitten but it was clear she could count or something very similar to counting.

Or she just recognized each one and came to a conclusion like "John is missing"

Yes, it seems very plausible that a cat might recognise individual kittens.

I think there are also documented cases of birds noticing if one of their newly hatched young is missing. In this case, because they're newly hatched, one might guess that the parent is not recognising each one. An experiment is called for: swap the young of two parents and see if both parents are happy with the right number. I suspect someone has already tried this, but I can't immediately find any reports.

Yeah I imagine it’s much easier for animals to recognize individuals of their own species than individuals of another. e.g. Humans have specialized brain circuitry for recognizing human faces, but all horse or bird faces basically look the same to us. A bird might easily think all humans look basically the same but obviously each bird is totally different.

With cats, like with dogs, recognizing conspecifics is most likely highly dependent on scent, a sensory modality whose richness in those and many other species we humans can hardly imagine.

It could be based on scent as well. Swapping young would introduce another set of pheromones from the other mother.

> I think there are also documented cases of birds noticing if one of their newly hatched young is missing.

Cuckoo success would seem to suggest that as long as there is something in the nest demonstrating the correct 'feed me' cues, the adults will feed it

This would be easy to test by replacing John with another kitten, right?

This is claiming more than I think is really true. They're just training the bees to avoid circles. The more circles, the less sugar water.

They're talking about zero like some highly advanced concept, probably because it took humans a while to grasp. But like others are saying, they're was always the concept of Nothing or None, and this is only how we interpreted it. Of course many animals can "comprehend" zero, but tell them to represent it and that gets a little different. Might take a while before they group the terms that they think of as 2 or None

Yeah, I'm pretty sure humans never really struggled with the idea of Nothing or None. From what I can tell, the "discovery of zero" refers to treating it philosophically as a number, developing a specific symbol for it, and using that symbol in positional number systems, which are all things that bees likely do not do.

> using that symbol in positional number systems

Exactly. There is a book about the number zero which is a good read (as are the ones about pi, e, i, etc.).

>"If bees can perceive zero with a brain of less than a million neurons, it suggests there are simple efficient ways to teach AI new tricks."

Why are they just speculating about this? Do they think it would be hard to train an image classifier to say "left" or "right" depending on whether that half of the image had more geometric objects in it, then show it one vs none objects to test it?

I don't think this would be at all difficult to do, the training data could even be procedurally generated.

EDIT:

It looks like the test accuracy was only ~60% but from the supplements it took them only 30-50 training examples over the course of 2-4 hrs to reach the criteria of 80% correct:

>When presented with the unfamiliar numbers of one versus zero, bees chose the lower number of zero (63.0 ± 2.9%; z score = 4.23, P < 0.001; Fig. 1D), showing an understanding that an empty set is lower than one,which is challenging for some other animals (5, 6, 18). http://science.sciencemag.org/content/360/6393/1124

[1] http://science.sciencemag.org/content/suppl/2018/06/06/360.6...

This is very interesting reading and I don't want to discount what was discovered. I would however like to see what other possible explanations are possible and come up with tests to determine one or the other.

It is good to think in terms of why bees have this ability. Identifying patterns in flora is beneficial, how do these differences arise? It could be that certain patterns appear with others in correlated or inversely correlated couplings. As far as the bees are concerned the amount of sugar is correlated with a card shape of paper which hasn't expressed circles or triangles. Perhaps that same blank card if left for a while will have some symbols and less sugar. The 'concept', if we can even call it that, is that the presence of symbols reduces the sugar concentration. I would definitely say they grok an inverse relationship and that a blank card has a 'small quantity' of symbols. I can't think of an experiment to tell if the bees consider a blank card to be a small quantity or zero.

A good thought experiment is if we trained an AI similarly with these same cards. If when presented with a blank card as yet unseen and a card with one or six symbols, it chose the blank one, I don't think many would be arguing (yet) that the AI understands zero, rather that it's pattern matching wiring just classified it one way or the other.

Even AlphaGo when playing Lee Sedol apparently had gaps in its play and the team was uncertain if they would be uncovered or not. Many, not all of these were discovered and 'trained out' earlier working with Fan Hui who developed a sense of where such weaknesses remained in some way to the way Kasparov could sus out Deep Blue's 'thought patterns'. I wish there were more details on this and how much less susceptable AlphaZero is to it.

It's pretty amazing to me that insect brains, as big as a few grains of sand, still outperform the best we can do in autonomous driving and navigation. They can fly around without hitting each other or other objects, they can leave their nest and find their way back, etc.

> They can fly around without hitting each other or other objects

They actually can't. They just harmlessly bounce off things when they hit them due to their small size and low speed.

Also our creations lack a self-healing exoskeleton.

I'm going to need about 4 years a quite a bit of VC money and will I have the autonomous driving solution for you!

It's interesting though that a fly will bounce against the window all day (barring statistical luck) while a wasp or bee will bounce once, understand that that place cannot be passed, bounce the next window once and then find the open one very soon.

My milage varied. After 30 mins of banging on windows, half dead bees fall down and sting you if you stand on them. Don’t have open honey in your house in autumn, and the problem is much reduced.

This is not true at all. First, many insects can fly very fast. Even honeybees, which are notoriously poor fliers in the insect world, can reach up to 20mph, carrying a load of pollen or nectar. Dragonflies can reach over 60mph, and some flies may be able to reach upwards of 80-90mph. Second, many of these insects have excellent visual collision avoidance mechanisms and the strongest fliers have exquisite control over their flight. There is a lot of literature on this topic.

Well bugs hit windshields and fly against walls all the time, but they can afford to "crash"

I mow the lawn near beehives and they do not steer well. There is a special terror that comes from a mouth hit from an inbound bee. The more loaded up they are and the sunnier it is the worse they get.

The bigger bees steer pretty well considering their size. They just don't react to moving obstacles as well as static ones.

Text-only version: https://text.npr.org/s.php?sId=617863467

Thanks. Did you have to accept cookies to find this link?

I didn't agree to cookies, but to be frank, I don't know how I was supposed to find the link. The "Decline" button sent me to https://text.npr.org/, which doesn't seem to link to the article in question. So I took a random aritcle and substituted the number in the URL for the one in the HN's URL. That worked. :-)

Next step: automation.

https://news.ycombinator.com/item?id=17263446

Are there similar experiments where animals exhibit the behavior they expected? Specifically "that the bees would see a blank paper as something irrelevant that was completely different than what they had been trained on."

It would make this research much more interesting, because it sounds like zero is just a concept that is included with mathematical abilities, albeit with a little more conceptual difficulty.

There are many experiments demonstrating computational abilities of insects. As another example, see very detailed article

Argentine ants solve the Towers of Hanoi http://jeb.biologists.org/content/214/1/50

“Zero” is a very different concept from “none.”

Human civilization had a word for “nothing” / “none” / “not any” for as long as we’ve had language. “None” is something chimps can understand easily. Etc.

But zero—that is a counting number. Numbers themselves are a very abstract technique that required specific situations (shepherds counting sheep, etc.) to invent. And zero came much later still.

Which is to say: sum types are easy. Types with null values are hard ;)

I would bet that these bees aren’t so much reading this data as “zero”, but rather “presence/absence”, where absence is okay and presence is bad. (It’s not like they first had to do a task that trained them to find the card with a three, after all.)

Nah.

From the study[1]:

"After reaching a criterion of ≥ 80% accuracy, bees demonstrated the learned concept of numerically less when presented with the numbers two to five(73.8 ± 1.9%; z score = 10.18, P <0.001). When presented with the unfamiliar numbers of one versus zero, bees chose the lower number of zero (63.0 ± 2.9%; z score = 4.23, P < 0.001; Fig. 1D), showing an understanding that an empty set is lower than one, which is challenging for some other animals."

[1]http://science.sciencemag.org/content/360/6393/1124

They didn't choose the lower number zero. They chose the lower quantity zero. It's not the same.

Bees communicate with each other to find nectar. They can let the other bees in the nest know what direction to travel and how abundant the resources are there, but they can also warn other bees to avoid places they didn't find nectar. It makes sense that bees would have evolved some way to determine a relative numerical value of something i.e. flowers and be able to act on that information at a later time, like when they get back to the nest. We just never quite known how bees figured this out. For example, one of the previous presumptions was bees used visual cues like variations in colors to determine what area had more flowers but what experiments like this imply is that bees actually have an inherent concept of numerical values themselves and can apply it to other situations, which is something we didn't think an insect could do.

It seems like your criteria for understanding is essentially impossible to test for. What's the point of it?

"Which is to say: sum types are easy. Types with null values are hard ;)"

I'd like to present you with the 2018 Hacker News Award for Smoothest Surprise Segue Into A Flamewar Topic.

I know, I know, it's a bit early, but I don't think anyone is going to top this.

What's the flamewar topic?

Type systems. Granted, it may not be US politics, but it can produce heated conversations short on light.

And which programming languages do it right/wrong and how we'd all be better off if we switched.

> I know, I know,

No you don’t.

> it's a bit early

No it isn’t.

> but I don't think anyone is going to top this.

Yes they will.

... Oh, you wanted Abuse? This is Arguments. Abuse is in room 1 [1].

[1] ref: Monty Python, Argument Clinic sketch.

I hate to type this out, but it's all just a zero sum game.

Perhaps the bees figured out that the card with the most white space was the card with sugar water?

A card with fewer symbols presumably has more white space. A card with no symbols has more than a card with one symbol.

Or maybe they smelled it?

I'm not convinced.

Maybe i don't get it, but this feels like not zero. It requires more things to be present so they could pick less as a concept, not specifically none as a concept

How did they hide the smell?

Reply on news.ycombinator.com