Dyer: Actually, it is really interesting. So, the experiments happen outdoors, so we’re not in a lab environment. They conducted on university grounds and we have beehive or beehives set up and maintained by professional beekeeper. So, those bees are just foraging all over campus and then what we will set up is a small gravity feeder in close proximity to the hive, and some bees will come and collect about five or ten percent sucrose concentration and just take it back to the hive. And, what we can do is go along with a spoon, it’s actually a special bee spoon but it’s a bit like a spoon, and put it under the nose of one of the bees, and mister kept their proboscis and lick this and liked it very much. And then, we can take this bee across to a separate testing site, which is outside, and train the bee to come back to that testing site and we put a little color mark on her back, so we know which bee is. So, once she comes back, that takes about an hour to train the bee to do that. We can start training him to use the apparatus and then introduce visual stimuli and get them to start solving the problem. The nice thing is, because of bee we’ll come back and do this all day long, we can give fairly long training programs. So, we can get good insights into their visual learning. What I like to think about is we have an animal or an insect, we have a small brain, less than a million neurons, and she wants to come back and participate in our experiment all day long, which means we can train her for a long time; but also conduct lots of control experiments which is very important in science. And, she just keeps coming back and participating in the experiment and she’s highly motivated. So, our experiments typically designed so that all the data is collected in one day, but the bees will come back on subsequent days and some of the work we do we’re interested in their long-term memory and how memory might be robust over several days or might decay depending the type of visual problem the bees have been encountered with. These links back to us as humans ourselves. We’re interested how we remember certain things and why we forget certain things, are these general principles in how brains operate, and what are some of the rules governing that. So, it’s some of the other work we do.

Read More

Leigh: This made us wonder when it was that people first began researching sensory perception in bees, as well as what’s been learned about their cognitive abilities in the time since.

Dyer: Calvin Frisch sort of established the principles behind this in the early part of the 20th century, and what he found was if you put out a visual stimulus, let’s just think of a piece of blue card and you associate that with a sugar reward, the bees will come and drink that and like it very much. They collect that nutrition, fly it back to their hive, and then the same bee will come back and collect more. We’ll actually do this all day long. So, you can put a little color mark on the 4x or the back of the bee and know what the bee is doing. So, he was doing classical conditioning a bit like a Pavlovian type conditioning and showed that bees learn and you can collect data from them all day because when they fill up their crop, or honey stomach, they take that back to the hive and give it to the other bees to make it to honey for our breakfast. And then, I want to come back and do the experiment again. So, though you can do an experiment for 6, 7, 8 hours a day with the same subject, which is a very rare animal model, but the collecting of data takes a very very long time and that’s why you might see between the study in the 1990s, and then a couple studies 2008 – 2009, then our study which is now 2018. There’s big gaps in between because the study, which just published, probably took 3 or 4 years to get all the data together.

Watkins: Since modern bee research dates back just 100 years, Doug and I were interested in knowing what Adrian thinks might be some potential applications of the discovery that bees can understand the concept of zero.

Dyer: So, on the social side, a very big question is why did humans in some communities evolve the ability to think and understand the concept of zero, where there seems to be other civilizations where that process didn’t unfold. So, what is novel thinking what is our ability to be able to come up with new concepts. Is this a limitation of our cultural setting, Is it a limitation of our brain size, is our brain changing with time to enable this when we step sideways and see in comparative animal models like for parrot and now the bees that they can understand and solve these types of apparatus. Then we can pretty much make the conclusion that ancient human societies which didn’t understand the concept of zero just had to be because there wasn’t the need for their culture to use it. Because if they were had a complex environment where they did need to process zero their brain would have been able to learn it, or if for example you had a child from a community which didn’t press a zero but you gave him a modern education they would probably pick this up pretty quickly. Because what we did was just take a bee with a fairly simple brain compared to the primate brain teach them some rules and very quieted information very quickly actually. Now, once we understand that about brains it gives us some insights into how we might think about artificial intelligence and here when we look at comparative brain sizes, like primates and bees, we see similar or different strategies we can start to understand what size of brain enables very complex cognitive-like processing and what might be the neural structures which support that. So, some of the work we are hoping to do in the near future, in collaboration with a few labs that have made some inroads to this already, is having a look inside the bees’ brain and understand what changes we’ve experienced, what areas of a brain are involved in processing complex stimuli, how we want to do numerosity processing, and when we understand how that network operates, how might we be able to work with computer scientists working artificial intelligence to come up with better solutions for robotic vision.

Leigh: Adrian and his team used a couple of different devices to teach and test their bees, so Ryan and I wanted to find out more about how they work.

Dyer: We use two types of apparatus to test the bees. One is rotating screens. So, this is a vertically presented screen of about 50 centimeters in diameter, and it enables us to present a number of different stimuli to the bees. Because it’s rotating, we can throughout the experiment keep moving its position to randomize for spatial positions of the stimuli and there’s also multiple pegs where these are presented. And, that enables the bee to see stimuli and learn over a number of trials, how rules might govern the differences between stimuli. Bee, in the field of working bee vision, there’s another type of apparatus, coral-y mace, and that’s where a bee flies into a tunnel and on the left or right hand side the stimuli presented and the bee must choose which one she thinks is correct. That gives very good control over the visual angle at which the bee makes decision. It’s a very well accepted technique for understanding bee cognition. Because our data of the concept of zero was so important to understand, we just repeated the experiments with both types of apparatuses, which give different perspectives or potentially might give different answers, but both ways we tested to get very consistent results so we have a very robust finding.

Watkins: The Russian psychologist Ivan Pavlov famously discovered classical conditioning. In his research with dogs, he repeatedly paired a potent natural stimulus – food – with a previously neutral stimulus – sound. In doing so, he was able to cause the dogs to salivate in anticipation of the food simply by making the sound. Doug and I asked Adrian how many times a bee has to have a behavior reinforced in a similar way for it to reliably reproduce the behavior.

Dyer: I was discussing this with my children, we have a day actually. So, in our previous work, we showed that honeybees can very reliably learn to recognize and encode faces, including human faces and the period of time it takes them to acquire any information is about somewhere between eighty and a hundred decisions, which seems to be faster than what we expect from vertebrates. This is well reproduced by a number of groups now in different labs and in different models actually. So, we showed that honeybees could recognize faces very reliably and then Elizabeth Tibbets, in Arizona in the USA, showed that wasp could learn to code faces very reliably, and she had a major paper in science on that. So, it it’s interesting that insects appear to be able to learn some visual concepts faster than vertebrates do, and it might just come back to what is their lifespan. So, a foraging bee may be only foraging for a few days in their life. I need to be able to learn very complex information very quickly, and if I have a brain which seems to be very good at doing this.

Leigh: Flower colors are believed to have evolved over 100 million years to in response to their bee pollinators’ color vision. So, Ryan and I were interested in learning what the quality of bees’ vision is like.

Dyer: So, the honeybee has a very different color visual system. I also try chimeric but they have an ultraviolet sensitive photoreceptor, a blue and a green sensitive photoreceptor. And if you go out into the garden and look at many flowers, about 20 to 30 percent of flowers actually have reflectance properties in the ultraviolet. So, they have evolved to suit the visual system, the most important pollinators which are typically bees. So, it’s a very different visual system to how we see at summer color side. And then, on the spatial side, their spatial resolution is much poorer than ours. And what spatial resolution relates to everyone can probably think of going to the optometrist, or the doctor, and having to read a funny-looking chart of Al TCN and you’re reading progressively down the lines. We’re not actually very interested in your ability to recognize four letters. They’re testing the resolution of your eye, our resolution, because we have a lens eye which is actually very very good. But a honeybee, because they have a very small body, I don’t use lens eyes because diffraction would really reclamation. I have a compound eye, and the compound eye is not very accurate at seeing detail at a distance. But when you get very close to an item like a flower, then I see we used to be sharply the image.

Watkins: Bees are well known for their intelligence and industriousness, as well as their ability to communicate the presence of food to others in their hive. Given this, we asked Adrian how he and his team prevented the trained bees from teaching their nestmates about the sugar water in the study’s experiments.

Dyer: In the experiment we published, we just had control mechanisms to exclude that and that’s basically having the gravity fader, which captures the attention of the other base. So, if you haven’t actually trained them to use the apparatus but our base don’t get attracted it’s it’s a complex scenario because if you control with distances between the hive and the testing apparatus and you keep that at about 20 or 25 meters, that distance of bees are only doing what’s called round dance. They have seen all that food rewards out there but we’re not actually giving specific vector information about where it is. So, our test bee might be going back to the hive and doing a round dancing, going: hey is great food out there, any other bees go: yeah we know we know and let’s go to the gravity feeder. If the distances get further, then the bees do their waggle dance and you would have a problem. But a bit like a bike working, you just have sort of half of this whole situation in control and just chill nicely. For some other experimental work, people are very interested in social learning of bees and this work is being driven by [inaudible] in Queen Mary University of London, and there, they do actively allow a bee to recruit another bee or to let him never be watched. They are interested in how the interactions improve learning. But we actually, in our experiment, went to a lot of lengths to totally exclude that, because we wanted to look at individuals.

Leigh: Next, Ryan and I were interested in finding out more about how the bees in the study were specifically trained to understand the concept of zero.

Dyer: To understand whether an animal might perceive a concept of zero is actually a complex set of experiments. Because you need to unpack data on several different levels to see if it might be confound like low-level explanations, and bee to understand, is zero being processed relative to higher numbers. So, the first thing we did was train bees with stimuli with elements of 1, 2, 3, or 4. So, if you think of a 6 by 6 centimeter piece of white paper, and it’s containing some different elements on it and this might be squares or triangles or circles for example, and then a bee comes along and she sees maybe two elements and four elements. So, if we’re teaching her a rule of choose less fan she has to choose two as the correct answer. Once she’s made some choices and filled her crop, she flies back to the hive to give to other bees and gives us a chance to clean everything and she comes back. But now, she sees one and two almonds displayed she might first go to all that worked last time, but she finds that it’s not the correct answer this time and she actually gets a bit of tasting clean Hemi sulfate, which I don’t like the taste of. So, it promotes learning. So, on this experiment, number one is the correct answer and this goes on for several hours, until she learns the way to solve this problem is to always choose the number less fan. And then, we present the bee with a novel problem of some of the ambiguous numbers to offer. So, these are ambiguous because about half a time while rewarding and half a time are not depending if the alternative what the alternative number was. And then we present this number against a blank sheet, and if we’re using a simple associative type mechanism they should just go to the number which is presented. But, if they’re using a rule of less fan, then I should choose the sheet with no items on it and that’s what the bees did, or statistically speaking, they did that between 60 and 70 percent of the time. Now, that was a little bit interesting. So, that’s suggestive at though understanding the the concept of zero.

Watkins: The team didn’t stop there, however, as we’ll hear after this short break.

Ad: We Share Science

Watkins: As Adrian explains next, after teaching bees the “less than” rule, the team then tested whether the bees could apply that learning to a novel task: namely, choosing a blank card representing one less than 1, or 0. Here’s Adrian Dyer.

Dyer: As is often the case in science, you have many potential confounds. So, we had to do preference experiments. So, we trained bees on just some numbers but not actually teaching them a specific rule. So, there’s an associative mechanism just go to numbers and then we gave from a presentation of novel numbers of either 0, a blank sheet, or 1, but never seen as before, and here they went to the number 1. So, it had an element and so if I hadn’t specifically being taught the less van rule when the bees didn’t just by preference go to a blank sheet so that excluded that factor and then one of the most important pieces of information or evidence for animals, and this includes non-human primates or even young human children, of their ability to understand 0 is fitness a number ordering effect. So, 0 versus 1 is a difficult task, 0 versus 2 is a bit easier, and as you get to larger numbers, 0 versus 5 or 6 becomes much easier for the visual system to processes. And it means that the numbers are perceived as being an order and these experiments  are always done with independent bees which is important for the statistical analyses. So, in a another separate experiment we train these two different pairs, so we had to choose less than again but when they get tested with 0 versus 1, or 0 versus 2, 0 versus 5, 0 versus 6, we saw their crew was significantly higher for the larger number differences which is consistent with the evidence. So, the number sequence experiment is very important because it’s been established previously by other researchers that in young children there is number sequence effects. So, a larger number difference compared to zero is perceived more accurately and this is also been shown in non-human primates. So, it’s taken as one of the key pieces of evidence that zero is processed in relation to a number sequence which fits one of the formal definitions for what zero is.

Leigh: Since the minimum sample size necessary for a study depends on the data analyses that a researcher plans to carry out, we asked Adrian how his team went about determining how many bees to include in their experiments.

Dyer: In honeybee experiments like we do, we have very tight control over the type of stimuli which are presented to the bees, and we can test a number of individuals in a very careful way. Typically, about 10 bees is regarded as fairly robust for pulling apart factors of whether they can detect stimuli or process stimuli. Now to support this, we do very robust statistics and if you get consistent bees, which we observed in our experiments of what was reported, that was the entirety of the bees’ history. So, every bee, who was tested, was able to acquire these rules and you start getting 10 bees which are tested and consistently performing at 70 to 80 percent accuracy. The p-value start approaching much less than 1 in 100, much less than 1 in 1000, which is regarded as fairly robust. But then, what we do is repeat the experiment with different types of apparatus and you get consistent results again, and then you do extra control experiments where you vary some other factors and you see consistent results again. So, you know that results are very very robust.

Watkins: Both in e-letters posted to Science magazine’s website as well as in a vigorous debate on Reddit, commenters have suggested that the bees may have been differentiating the cards used in the study by way of their lightness or darkness rather than arithmetic. However, Adrian and his team designed the cards so that – except, obviously, for the blank zero card – they all had same ratio of black to white, as he explains next.

Dyer: Brightness is what you’d call an associative type mechanism, so is for a low level explanation, which would explain the results. That’s why we did a number of control experiments to show that this is not the case. So, what we did was continually change the stimuli throughout the experiment. So, even though they might be viewing three versus four, those shapes changed and the configuration of those shapes for always changing, but the overall brightness between all the stimuli was equal. Then, when the bees had to solve the testing problems, they had to extrapolate the information, the rules they had been taught. I had to extrapolate that to solve the concept of zero. Now, any associative, and we did lots of computer modeling, I miss any associative mechanism would have given us the opposite answer. So, you have an experiment in balance if they were just using an associative mechanism they should have chosen the opposite to choosing the blank zero stimulus. But, since they were able to apply a rule that showed they comprehended the context of information, that in collaboration with a number order sequence experiment. So, the fact that we’re increasing magnitude their accuracy improved, is taken as a fairly robust set of evidence that they were doing numerical processing rather than a low-level queue. Previous researchers have shown that bees can count. So, you put all these pieces of information together to be able to understand how the bees were solving a problem.

Leigh: Lastly, we couldn’t resist asking Adrian whether he or the study’s lead author, Scarlett Howard, often get stung by bees in their work. And, if so, whether they’ve grown accustomed to it.

Dyer: I’ve been working in a professional sense with bees both bumblebees and honeybees for about 18 years now, and I have never been stung by working the bees and neither has Scarlet to the best of my knowledge. So, we have fairly careful lab protocols for how you work with bees. But also, if you work with bees and you treat them with a lot of respect, you’re very clean with how you use all your stimuli which you need to do for scientific robustness anyhow, it’s a fairly calm relaxed environment. Usually the way people get stung with bees is being careless with food or drink and something gets spilt and there’s some sugary material around, or once I was stolen a foot by a honeybee. But, that was when I was 6 years old and my mama told me not to walk across the grass, because there were flowers and bees, and being a six-year-old boy, I ignored her and the bees taught me an important lesson of always listen to mom. So, that was my fault not the bees. But, in a professional sense, the way we run the lab we don’t get stung. But, I don’t want to give listeners the impression that to be cavalier with bees. In this regard, we have a beekeeper who very carefully manages our hive and the hive has a queen, who has very fresh pheromones. So, she’s in good control of her hive and because it’s well-maintained, the bees are not aggressive at all. So it’s fairly easy to recruit them to do an experiment once they feel easy. It takes several hours, but just a protocol we go through and there’s really no issue we have with bees being aggressive or stung. So, it’s not to mean we should approach any bee or beehive, but in a university setting we’re able to set up a protocol. So, it’s not a problem.

Leigh: That was Adrian Dyer, discussing the article “Numerical ordering of zero in honey bees,” which he published with Scarlett Howard and four other researchers in the June 8, 2018 issue of the journal Science. You’ll find a link to their paper on parsingscience.org/e31, along with bonus content and other material that he discussed during the episode.

Watkins: Though we launched it just a few weeks ago, Parsing Science’s weekly newsletter continues to be a big hit! You can sign up at parsingscience.org/newsletter, or if you’d like to check out our first three issues, go to parsingscience.org/newsarchive.

Leigh: Next time on Parsing Science, we’ll be joined by Mike Vitevitch from University of Kansas. He’ll talk with us about his research into the “Speech-to-Song Illusion”, in which a repeated spoken phrase results in it being perceived as if it is being sung.

Mike Vitevitch: We have this misconception, I think, in perception and in memory too that, you know, what we see is what we get and there’s a lot of construction going on for memories, as well as for perception.

Watkins: We hope that you’ll join us again.

[ Back to topics ]