Listeners:

Top listeners:

00:00 00:00

Go to album

Paul Verschure on consciousness and distributed adaptive control CSN Podcasts
Edvard Moser on grid cells and entorhinal cortex CSN Podcasts
Giacomo Rizzolatti on mirror neurons and action understanding CSN Podcasts
Robert Axelrod on game theory and prisoner's dilemma CSN Podcasts
Adrian Owen on disorders of consciousness and vegetative state CSN Podcasts
Jonathan Whitlock on markerless motion capture and posterior parietal cortex CSN Podcasts
Luis Puelles on neuroanatomy and prosomeric model CSN Podcasts
Zoltan Molnar on subplate neurons and cortical development CSN Podcasts

Mandyam Srinivasan on honeybee cognition and waggle dance

Episode 18 14.03.2012

PLAY EPISODE

Season 2012

0 Followers 22 Episodes

Description

A honeybee learns a color in five visits, generalizes matching rules across sensory modalities, and signals food distance to nestmates through dance. How does a brain with fewer than a million neurons achieve cognitive feats that challenge our understanding of intelligence?

Subscribe for more from the Convergent Science Network podcast series.

Srinivasan explains that insect compound eyes create a fundamentally different visual world than vertebrate camera eyes. With the two compound eyes too close together for effective stereo vision, bees rely on optic flow, the apparent motion of images across the retina during flight, to gauge distance. His tunnel experiments demonstrated that bees measure distance in units of integrated optic flow rather than absolute meters, which means flying over a featureless lake versus a textured forest produces different distance readings. The system works because all bees from the same hive take the same route, so calibration errors cancel out in the waggle dance communication.

The waggle dance itself encodes both direction and distance to food sources. Direction is referenced to the sun’s position or the sky’s polarization pattern, while distance is conveyed by the duration of the waggle run. Srinivasan describes how recruited bees evaluate the ratio of caloric return to energy expenditure, effectively performing cost-benefit analysis before choosing which advertised food source to visit. Intriguingly, angular precision in the dance increases with distance, compensating for the fact that a fixed angular error maps to a larger search area at greater range. The evolutionary origins of the dance may trace to solitary butterflies that perform waggle movements without an audience, suggesting the behavior was co-opted for communication from a pre-existing motor pattern.

The cognitive capabilities of bees extend far beyond navigation. They learn colors in five rewards, discriminate wavelengths with near-human precision, and exhibit color constancy across lighting conditions. Most remarkably, bees trained on a delayed match-to-sample task using odors spontaneously transfer the matching rule to visual stimuli they have never been trained on, demonstrating cross-modal concept learning. The mushroom bodies, which expand dramatically when bees begin foraging, likely serve as the invertebrate analog of the hippocampus, though the physiological basis of bee memory remains almost entirely unknown.

Tagged as:
Bees Compound Eyes honeybee cognition Optic Flow waggle dance

About the author

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

Timestamp

00:00:03 - This is the Convergent Science Network podcast. Leading researchers in the domain
00:00:10 - of neuroscience, brain theory and technology are interviewed by Paul Verschure and Tony Prescott.
00:00:19 - So this is Paul Verschure with Parta Mitra with the Convergent Science Network podcast.
00:00:26 - And in this episode, recorded as part of the CSN Barcelona Cognition,
00:00:30 - Brain and Technology Summer School, we're talking to Majiem Srinivasan, also known as Srini.
00:00:36 - And Srini, in your lecture, you focused very much on the mental life of bees,
00:00:44 - if you want, with a special focus on memory.
00:00:48 - Well, part of it, yeah. The first part was more on low-level vision and navigation.
00:00:52 - Right. And the second part was really more on the cognitive bit.
00:00:55 - So we can start with the first part or the second part, whatever you prefer.
00:00:59 - Good. Well, before we do that, if we think about, let's say,
00:01:04 - the vision or the cognition in bees, could you give us an intuitive description
00:01:09 - of the world in which a bee lives? Yes.
00:01:12 - Well, I suppose it's okay. I mean, it probably depends on whether it's flying
00:01:19 - outdoors or staying indoors in the hive.
00:01:23 - Outdoors, I suppose the world is very similar to what we experience.
00:01:26 - I mean, it's mostly visually oriented or visually dominated,
00:01:31 - I would say, with a bit of olfaction thrown into it.
00:01:35 - When you get close to a flower, you smell it and you either go towards it because
00:01:39 - you recognize the smell or you avoid it because the nectar is not good over there.
00:01:44 - But within the hive, it's of course a completely different situation.
00:01:47 - It's totally dark and there's a lot of pheromonal contact and there's a lot
00:01:52 - of acoustic signaling going on.
00:01:54 - And that's stuff that a lot of other people are studying. I don't know a lot
00:01:58 - about that, but it's certainly a lot of, it's mostly acoustic,
00:02:00 - I would say, inside the hive. Okay.
00:02:03 - So, starting out then with the visual capabilities of bees, this would then
00:02:07 - support their navigational skills.
00:02:10 - Absolutely. So, how do you decompose vision in bees?
00:02:14 - Okay. I mean, as you probably know, insects have these compound eyes,
00:02:19 - which are many, many facets. facets, and we have the so-called simple eyes, camera lens type eyes.
00:02:25 - And so optically, superficially, there's a difference between insect vision
00:02:29 - and our own vision in terms of just how the information, the visual information
00:02:33 - is collected or sampled.
00:02:36 - But the more fundamental difference, as I was saying in the talk,
00:02:39 - is the two eyes of an insect.
00:02:42 - When I say two eyes, I mean the right compound eye and the left compound eye.
00:02:45 - They're very close together.
00:02:47 - So it becomes very difficult to do stereo if you're an insect because the baseline is very small.
00:02:55 - And the only way you can do stereo, the only circumstance under which you can
00:02:58 - do stereo is when your object of interest is very close to your eyes,
00:03:02 - which is when you get a much bigger disparity, of course.
00:03:04 - So this is where it looks like insects have gone a different route.
00:03:09 - And they... Sorry, am I distracting you? No, no, no, no. I'm distracting myself.
00:03:13 - Go ahead. Go ahead. Which is where they...
00:03:17 - They have to use a very active mode of sensing the world in three dimensions,
00:03:22 - and this is done by physically moving in it and measuring the optic flow that
00:03:28 - various objects around them generate.
00:03:30 - Right. So something that's very close to you, if you're moving in a straight
00:03:33 - line, something that's very close to you moves by very rapidly in your visual
00:03:36 - field, and at high speed tells you that this is very close.
00:03:40 - And something that's very far away, on the other hand, moves very slowly,
00:03:43 - and that tells you that that object is at infinity.
00:03:45 - And what we've done is a nice series of experiments maybe the first ones I suppose
00:03:50 - with insects to show that insects really gauge distance based on how rapidly the images are moving.
00:03:56 - So this is done with flying them through tunnels and moving the patterns on the walls as you know.
00:04:02 - But now in insect vision which is a very active area of research,
00:04:06 - there's if you want let's say a prototypical design of a vision system fairly
00:04:09 - hierarchical where you slowly move to let's say a bit more complex and integrative,
00:04:15 - processing of vision-derived signals, but in what sense does the B visual system
00:04:21 - really deviate from that sort of standard template?
00:04:24 - What are the unique features? Okay, what we're finding, I suppose the fly,
00:04:27 - most of that work, the initial work, has been done with the fly.
00:04:30 - And there, you know, the motion-setting system has been studied very well there.
00:04:36 - But I think it's just one part of what goes on in an insect visual pathway.
00:04:41 - I mean, there's lots of other things is happening. For example,
00:04:44 - in bees, again, we're finding what seems to be almost certainly multiple parallel
00:04:48 - pathways, so some of which function the way Reichardt described it,
00:04:52 - with this correlation type motion detector.
00:04:54 - But all of these other mechanisms that are used to sense range,
00:04:59 - for example, they don't seem to obey the Reichardtian laws.
00:05:04 - They seem to be more accurate measuring devices for measuring image velocity,
00:05:10 - independently of the spatial frequency content.
00:05:12 - And independently, largely of the contrast of the scene as well.
00:05:17 - And if you think about it, you need that, because you want to know how far away
00:05:20 - a surface is, regardless of what its texture is, spatial texture,
00:05:24 - or what its contrast is, right?
00:05:26 - So you need a robust system, and the Rijkaard system does not deliver that robustness to you.
00:05:31 - So there must be other systems there which are acting in parallel,
00:05:34 - or which are processing the, you know,
00:05:37 - using several different Rijkaard detectors maybe, multiple channels to get that,
00:05:42 - what do you call it, that vertical information on velocity. So that's one thing.
00:05:48 - Apart from that, of course, honeybees have excellent color vision,
00:05:53 - trachromatic color vision, which they need to detect flowers, recognize flowers.
00:05:58 - And they're really a beautiful learning machine. You can train the bee to learn a color in half an hour.
00:06:05 - Five rewards is enough for a bee to learn a color. So they do great with that.
00:06:12 - And of course, they have a beautiful polarization sense as well,
00:06:16 - which you probably know.
00:06:16 - So to analyze the polarization pattern of the sky and use that as a compass.
00:06:21 - If the sun is hidden behind a cloud, then they can't use the sun anymore,
00:06:24 - but they can still use the polarization pattern of the sky to work out which direction to fly.
00:06:28 - So all of that is there in the bee, which I think, well, color vision certainly
00:06:32 - is not as well developed in the fly. and with polarization vision,
00:06:35 - no one really knows for sure. Right, okay.
00:06:39 - You had mentioned that when you had these artificial small tunnels set up,
00:06:44 - you could fool the bee into thinking that it's flown a smaller distance.
00:06:48 - A larger distance. Flown a larger distance.
00:06:51 - So it's not entirely invariant to the environmental views. Exactly, exactly.
00:06:56 - No, you're absolutely right. But what's neat about that is that,
00:07:00 - as I was saying, and when someone asked the question.
00:07:05 - So you would say, for example, if bees flew in different landscapes,
00:07:10 - they would give you different distance readings, right?
00:07:13 - Because the optic flow they've experienced could have been different.
00:07:16 - Let's say you fly over a plane, over a lake, for example, where there's almost
00:07:21 - no optic flow as opposed to going through a dense forest.
00:07:24 - But what happens is that a bee that goes and finds a food source after it's
00:07:27 - flown over a lake comes back and does a dance and it signals a certain amount
00:07:32 - of units of optic flow, which is the measure of distance.
00:07:34 - And then all the other bees take the same route.
00:07:37 - So they experience the same environment. So whatever calibrations or miscalibrations
00:07:41 - are happening, they're the same for all the bees, and so they cancel each other out, right?
00:07:46 - So that's the way I think, that's why the system probably works.
00:07:51 - But again, there's a little unsolved question there because different bees can
00:07:56 - fly at different heights above the ground.
00:07:58 - And so if you fly higher, you will experience a lower amount of optic flow,
00:08:02 - so the integrated flow will be lower.
00:08:04 - The question is then, do bees actually measure height and take that into account
00:08:09 - when they're doing their odometry, or do they all fly at the same height?
00:08:14 - We don't know that. There's some anecdotal observations which say bees fly typically
00:08:18 - about two meters above the ground, but whether that's really the case,
00:08:22 - no one really knows for sure.
00:08:24 - But if you would have something like a polarized light.
00:08:28 - Compass, if you want. You could always use that to recalibrate and make yourself,
00:08:33 - your distance estimate independent of your altitude.
00:08:36 - Well, the polarization compass will tell you only which direction you're flying
00:08:40 - in. It won't tell you your height about the ground.
00:08:42 - Then you could ignore it because you just instruct your fellow bees to fly in
00:08:47 - a certain orientation with respect to that reference.
00:08:49 - How do you know how far to go, though? The optic flow will depend on how far you are.
00:08:55 - So you have to give you have two bits of information to specify a location in
00:08:59 - a plan, right? You need the direction and you need the distance.
00:09:03 - So you've got to specify the position of the food source in polar coordinates.
00:09:07 - So the polarization pattern of the sun, they give you the compass direction,
00:09:11 - so it tells you, okay, go this way.
00:09:12 - Then you also have to tell them how far to go before you start looking for the food.
00:09:16 - But you could also argue that the
00:09:18 - more minimal model will just take the orientation until you hit a target.
00:09:22 - Yeah, you could, but you see, what is also interesting and probably subtle is
00:09:27 - that when these bees come back home and dance and they're advertising direction
00:09:31 - as well as distance and they're also passing out these nectar samples to these other bees.
00:09:37 - So quite often a bee will stop this dancing bee and beg it for nectar samples
00:09:41 - and this other bee will, you know,
00:09:43 - regurgitate some of the nectar that's brought out. And so this bee can actually
00:09:47 - assess how good that nectar is.
00:09:49 - And based on the distance signaled, it can decide whether it's really worthwhile
00:09:53 - going all that way or not.
00:09:54 - So these bees are evaluating dances that are being produced by different forages
00:09:58 - coming back from different food sites, which are advertising different food
00:10:01 - sources, and they're making up their minds.
00:10:03 - So do I go a long way to get to a good food source? Or can I fly a shorter route
00:10:08 - to get to a not-so-good food source?
00:10:11 - There's a trade-off there. And apparently they're working out the trade-off
00:10:14 - in their minds. So what kind of correlations did you find?
00:10:17 - Between? Well, for instance, the animals that decide to go off on a long foraging run.
00:10:23 - What's the difference in the nectar? So it seems like, so this is not our own
00:10:27 - work, but other labs have investigated it.
00:10:29 - And it looks like what these bees are doing is that they're looking at the ratio
00:10:32 - of calories brought in in terms of energy from the sugar, nectar,
00:10:36 - versus calories expended to get to the food source. And they're trying to maximize that.
00:10:41 - But then still an alternative could also be that you sample the nectar to get
00:10:46 - the chemical fingerprint of your target.
00:10:49 - Yeah. And this could help in your navigation because then again,
00:10:52 - you could go for some odor plume that matches that template to get to your target.
00:10:56 - Sure. No, you could do that.
00:10:58 - It's harder though because what happens is typically if you say,
00:11:03 - okay, this baby comes back and gives you a scent of lavender, right?
00:11:06 - So because you go looking out for lavender, you don't know which direction to
00:11:09 - go. And remember, it's not just a simple diffusion process.
00:11:12 - People in the old days used to think, you know, these scents just diffuse nicely
00:11:15 - and you have a nice diffusion gradient.
00:11:17 - And you just go up the gradient and you'll find the food source, right?
00:11:20 - No, it's not like that. When the wind is blowing, the whole thing is very turbulent
00:11:23 - and you get these little filaments of, you know, scent that are moving around.
00:11:29 - And you've got to be lucky to hit one of those filaments. Sure.
00:11:31 - And then you can zigzag, as you probably know, across the filament and get to
00:11:34 - that. You can do it, but it's tedious.
00:11:37 - Of course. Tedious. Sure. And there's some evidence that insects do,
00:11:40 - when they're close to the goal, they do start looking for the scent.
00:11:44 - And certainly they're using that information, too, to guide their final thing. Yeah.
00:11:48 - But you've shown in your controlled studies that once you've put one of the
00:11:53 - bees through your tunnel with a controlled distance,
00:11:57 - it can signal distance to the bees, which then will fly and will skip sources
00:12:05 - of food to go to the food source at the right distance.
00:12:09 - They're definitely paying attention to the dance, no doubt about that.
00:12:13 - I mean, I guess Paul's question was, yeah, why do they even need a dance?
00:12:17 - But there's no doubt that they are using it. That is definitely the case.
00:12:22 - I was just trying to see what's the minimal, what would be the minimal model?
00:12:25 - The minimal thing would certainly be, yeah.
00:12:27 - I mean, if you got the scent given to you at the hive and the nectar tasted
00:12:34 - good, you could just go out looking for that.
00:12:36 - Exactly. It'll take you a long time. It won't be very efficient.
00:12:39 - And you get some vector, some heading vector, right? Some heading vector.
00:12:42 - Okay, we get a heading vector. Okay.
00:12:44 - Okay. Okay, but while they're going through all that trouble,
00:12:46 - you might as well even provide the distance, right? I mean, it's one more parameter.
00:12:50 - But remember, we were trying to solve the problem, how to deal with altitude,
00:12:53 - right? So actually, it's interesting.
00:12:54 - So this is another thing that came up in the discussion.
00:12:57 - How does a dance actually evolve, right? And so people are not really sure,
00:13:01 - but there's a lot of evidence that other insects will dance too,
00:13:04 - completely out of context.
00:13:06 - Like which ones? There's a species of butterfly, for example,
00:13:08 - apparently, which is a solitary butterfly.
00:13:11 - So it flies a certain distance, and then when it lands, it does a waggle dance.
00:13:16 - It waggles its abdomen, just like the bee.
00:13:18 - But without an audience. Without an audience, exactly. But no one really knows why.
00:13:22 - But it's there. It's some kind
00:13:24 - of epiphenomenon. So maybe it's something metabolic or some other thing.
00:13:28 - And so maybe that has been picked up by these bees and being exploited as a signaling mechanism.
00:13:36 - Has anyone measured the accuracy with which the distance is signaled?
00:13:42 - So if different bees end up at different distances from the target or at different
00:13:49 - angles, has anyone measured the spread?
00:13:51 - So, yeah, typically the distance, the scatter and the distance,
00:13:57 - the Weigel durations is about 10%. And so it's always a percentage of the mean distance.
00:14:06 - But what is interesting is that with the angular error, as we were discussing
00:14:11 - the other day, it looks like when the food source is close by,
00:14:16 - there's a certain angular error in the direction indication.
00:14:19 - But when the food source is very far away, the angular error is much smaller.
00:14:24 - So, they're trying to account for the fact that if you have the same angular
00:14:29 - error, that will signal a much bigger patch when you're further away compared
00:14:32 - to when you're closer, right?
00:14:33 - So, they're trying to compensate for that. There's some built-in compensation, I guess.
00:14:37 - But this is interesting because then apparently this compensation comes at a cost.
00:14:41 - Because when something is nearby, so it's easier to find, they seem to put in
00:14:45 - less effort to communicate that parameter. Yes, yes. So, what's that cost exactly?
00:14:50 - You're saying cost in terms of doing an accurate dance? Right, exactly.
00:14:57 - It seems to me... So you're saying, why are they being sloppy when...
00:15:01 - Yeah. So that's interesting.
00:15:02 - Maybe, I don't know what the energetic requirements are for doing a dance,
00:15:08 - and whether precision involves more concentration.
00:15:11 - Are they really being more sloppy, or is it that when the waggle dance is longer,
00:15:16 - longer the audience gets to integrate for a longer period of time and therefore
00:15:21 - gets a more accurate estimate.
00:15:22 - The actual variance itself, so people have measured physically the variance
00:15:26 - in the axis orientation, and that seems to go down.
00:15:32 - But is that a simple consequence of having a longer down?
00:15:36 - It's possible. It's possible that it's a measurement. It's possible that even
00:15:40 - with the humans who are measuring these dances, maybe there is an element of
00:15:44 - increased accuracy simply because they have a longer sample,
00:15:48 - right? That's a good point.
00:15:50 - But that's only true when the Wegel dance is a highly controlled,
00:15:54 - low noise expression of these parameters. Is that true? Yeah.
00:15:58 - Yeah. So you mean how noisy is it? It's pretty noisy. It's not clean.
00:16:03 - It's a longer integration. Yeah.
00:16:05 - It wouldn't possibly help you to have a bit more accurate estimate.
00:16:08 - And the extreme limiting case is when the foot was very close to the hive,
00:16:14 - when, as you probably know, there's no longer even a waggle dance.
00:16:18 - So the waggle disappears completely and it becomes just a round dance.
00:16:21 - Right. And that simply says, okay, I'm not going to tell you exactly which direction to go.
00:16:26 - Just look within a small radius, 50 meters of the hive, and you'll find it.
00:16:30 - So it becomes very imprecise as you get very close to the origin.
00:16:34 - But the two interesting aspects of this, I find them interesting right now,
00:16:38 - is that you could also argue that on the one hand, it's the dancing bee itself
00:16:44 - who has integrated more or less information.
00:16:47 - Information like if you have to unwind this clock you
00:16:50 - have been winding up the clock for a shorter period of time
00:16:53 - when you found the nectar nearby so there's less to
00:16:55 - display in in the waggle dance would that be an alternative interpretation that
00:16:59 - would make sense yeah i mean the the the original people think
00:17:03 - the way this whole dance first started was in the tropical
00:17:06 - bees which were living not inside these
00:17:09 - dark you know hollow chambers uh you
00:17:12 - know nests that we find in the the european bees now but
00:17:15 - outdoors and um you know and and there most of
00:17:18 - these primitive bees now have dance on a horizontal surface
00:17:22 - not on a vertical surface and they're doing a kind
00:17:24 - of a scaled down version they're simulating the
00:17:27 - flight to the hive so i mean to the food source this is just basically pointing
00:17:30 - in the direction of the food and they're doing it coming back so it's exactly
00:17:36 - as though they're doing us yeah yeah a miniature version of the actual flight
00:17:39 - so in a sense that makes sense So they're coming back and they're sort of replaying that.
00:17:47 - But that's interesting because that creates a new problem for the observer.
00:17:50 - Because now the observer must extract this direction of movement independent of their own position.
00:17:56 - Okay, in the...
00:17:58 - In the outdoor arena, that's fairly easy, I think, because you basically have
00:18:03 - to face in the direction that the bee is waggling, and then you know where the sun is.
00:18:08 - So I say, okay, I'll fly in such a way that I keep the sun over there,
00:18:11 - or the polarization pattern in the sky if the sun is not visible,
00:18:14 - if there's some patch of clear sky.
00:18:16 - Okay, I'll say, okay, I know the orientation of the vector should be that.
00:18:19 - I'll hold that and go along, and I'll go for a distance that tells me,
00:18:25 - you know, how many optical flow units I should experience.
00:18:29 - And then when I get close to that, I'll start looking. Right.
00:18:32 - But now if I take a bee and I observe the bee dance, but now I rotate it a few
00:18:36 - times and then I let it go, will it still orient itself correctly?
00:18:41 - Theoretically, it should. You're not in practice.
00:18:44 - They're not all theoretical physicists. That's a good point.
00:18:47 - That's a good point. I don't know if anyone's done that.
00:18:49 - Because that's a bit the issue. That's very interesting. Because the way you
00:18:52 - describe it is that you physically orient yourself. Okay. As far as people know,
00:18:57 - no one really knows whether bees have a vestibular organ as yet.
00:19:02 - If they do, of course, they could get messed up.
00:19:04 - Absolutely right. Now, the way people think, they assess, they calibrate their
00:19:10 - orientation is the dance is done on the vertical plane inside the hive.
00:19:13 - And the direction of the sun is symbolized by the vertically upward direction,
00:19:19 - which is the direction of negative gravity.
00:19:22 - And people think they have a sense of gravity.
00:19:25 - Looking at the way, for example, your abdomen hangs in relation to the thorax,
00:19:30 - that tells you which direction gravity is, and that's your direction calibration.
00:19:35 - That's what people think. But in addition to that, if they have something that
00:19:39 - senses your angular velocity, as in a vestibular thing, no one really knows.
00:19:44 - But then how do they map that back in the horizontal plane relative to the sun?
00:19:49 - Because then gravity is not helping you. Yeah, yeah. It's very interesting. Completely unknown.
00:19:53 - Okay. It's very interesting. Also, how do they hold this information, you know? Exactly.
00:19:57 - There's the automatic information. They go some distance and it can be,
00:20:02 - you know, several minutes before they come back home and dance, right?
00:20:05 - So where is that information stored? Is it stored as a charge in,
00:20:10 - you know, the membrane potential of some neuron? or is it stored as an activity
00:20:15 - in some kind of place cell?
00:20:16 - My own feeling is that really the mushroom bodies and the insects really are
00:20:20 - like the hippocampus invertebrates and there might be place cells.
00:20:24 - So there could be certain place cells. As you go along to a food source,
00:20:28 - presumably successive place cells are lighting up.
00:20:32 - Now that there is more bee genetics and molecular biology coming on,
00:20:38 - are people thinking of manipulating bees to knock out specific genes.
00:20:43 - Certainly, knocking out the dancing gene, for example.
00:20:46 - Has that been done? No, no.
00:20:49 - I'm sure they're trying to do that. Or are people breeding bees for certain traits?
00:20:54 - Yeah, they've been doing that for a long time, even before the gene was mapped, of course.
00:20:59 - I mean, basically the idea is to breed bees that will produce a lot of honey.
00:21:03 - No, I meant more in the context of the waggle dance.
00:21:06 - Bees that are better signalers? or worse. Interesting.
00:21:12 - It would be very interesting. If you can find out a combination of genes that,
00:21:15 - for example, are responsible for the dance, that's a very basic thing, right?
00:21:19 - Could one not even screen these forward mutagenesis studies?
00:21:25 - I'm sure that's being done. I'm sure that's being done.
00:21:27 - I'm not a molecular biologist myself, but I'm sure there are people doing it.
00:21:30 - There's Gene Robinson in the US who's been doing stuff along those lines.
00:21:35 - I don't know if he'd actually succeeded yet, but that would be really exciting.
00:21:37 - And we started discussing whether this was learned behavior or innate behavior.
00:21:41 - You had some comments on that? Yeah, this is just a guess. I can't be sure.
00:21:45 - But I think probably the basic dance is pre-programmed. It's there like a child
00:21:51 - that has its gate already built in.
00:21:53 - But it's probably learned and refined during the first few days.
00:21:58 - The first three weeks I spent just not doing any foraging. As you know,
00:22:01 - they're just acting as nurse bees inside the hive.
00:22:04 - So there's no need to do any dancing. but it's only after that when they start to go and forage.
00:22:10 - That they start to need to dance. And that's the point, by the way.
00:22:13 - I don't know if Nick mentioned that, but when a bee first starts to go out and
00:22:18 - forage, the mushroom bodies expand hugely.
00:22:22 - So that map is being laid down and there's another piece of evidence that...
00:22:27 - That's interesting. It's really like the London taxi drivers.
00:22:30 - But are the mushroom bodies this very central structure in the bee brain? Yeah.
00:22:35 - Expanding in some uniform fashion or is it certain lobes of it that expand more than others?
00:22:41 - Oh, I don't know about the details. The calyces are the cup-shaped structures.
00:22:45 - They enlarge hugely. And there's still a debate about whether the number of
00:22:50 - neurons is actually going up or whether it's just the density of the neuropil.
00:22:54 - So it could be that the dendrites are getting longer and developing more synapses.
00:22:59 - There's a bit of a debate, I think, about whether the number of neurons,
00:23:03 - whether new neurons are being created.
00:23:04 - I don't know if that's still the case. But certainly all those calyces get very big.
00:23:10 - So then the question becomes just a bit of shift and also towards memory, if you want, right?
00:23:14 - Because you were talking about something like a place cell type response in these mushroom bodies.
00:23:18 - So if we start with, let's say, the basic task these bees have to solve,
00:23:22 - which is, okay, here I go. I'm navigating out.
00:23:24 - I'm finding some nectar and flying back home. And now I'm going to tell everybody about it. Yeah.
00:23:28 - So where's the memory of that? What's the memory of that? What are the physiological
00:23:32 - correlates of it? Nobody has the foggiest idea.
00:23:35 - That's a million-dollar thing. I mean, where is this information stored?
00:23:39 - Again, my guess is that it'll have to be somewhere in the mushroom bodies.
00:23:42 - It'll probably be some place cell that's firing that says, okay, this is the location.
00:23:46 - And that has to be translated somehow into producing a dance.
00:23:50 - Right. But let's approach it maybe different. Let's see, what do we know about
00:23:54 - B-memory structures that could help us understand answering this question?
00:23:59 - Unfortunately, almost nothing.
00:24:01 - Well, we know a little bit about this. Yeah, all that people know is that if
00:24:04 - you knock out the mushroom bodies, a lot of learned associations are lost.
00:24:09 - But that's about it, really.
00:24:11 - That's about it. There's a little bit that you were mentioning that certain
00:24:17 - things, the bees are easier to train, certain tasks, the bees can be trained in easier.
00:24:24 - Whereas other tasks are difficult. And you also had some time courses associated
00:24:28 - with that. Yeah, so there's gradations of learning. So, for example,
00:24:31 - the simplest kind of learning would be color learning.
00:24:34 - It's just so fast and so robust. If a bee learns a color, we learned it,
00:24:39 - as I said, in half an hour, five visits.
00:24:41 - But that means the bee gets rewarded to find that color, and it will find it within five trials.
00:24:48 - That's right. So the idea is, okay, do the reinforcement five times.
00:24:53 - So the classic von Frisch experiment was to have trained them to come on a piece of blue paper.
00:25:01 - Which had a drop of sugar water. And then, of course, he did a very nice...
00:25:05 - He just didn't simply do two different colors because they could be discriminating
00:25:10 - the colors not on the basis of color, but on the basis of intensity,
00:25:13 - on the basis of brightness.
00:25:14 - So what he did was he had a blue sheet that he rewarded the bees on.
00:25:20 - And then so they came five times, got rewarded on that.
00:25:23 - And then he gave them a test where this blue sheet, he took away the sugar water,
00:25:28 - there was no food anymore. and the blue sheet was placed in the midst of,
00:25:31 - or in the general vicinity of a bunch of other sheets of different gray levels.
00:25:37 - And so then he said, okay, let this train bee come and choose where it wanted
00:25:42 - to land. It picked the blue.
00:25:43 - So regardless of the intensity, it was perceiving hue as a separate quality.
00:25:48 - Right, exactly. And then landing on that thing. So that, again,
00:25:52 - takes only five rewards.
00:25:54 - And what's the generalization capability of the bee?
00:25:58 - If I now put this blue patch of paper among many other different kinds of blues, Oh, yeah.
00:26:05 - The delta-lambda discrimination is almost as good as that of a human.
00:26:10 - It's about five nanometers.
00:26:11 - So if you do experience the spectral lights, five nanometers.
00:26:14 - They also have very nice color constancy, the way we do.
00:26:18 - So they can perceive the shade of blue to be, well, they can recognize this
00:26:22 - irrespective of the illumination, largely.
00:26:24 - So evening versus midday, so that color constancy competition is going.
00:26:29 - All animals need it, probably.
00:26:30 - And bees have it too, of course, which is nice.
00:26:35 - Well, that's an exciting issue as well, but maybe we should get back to that
00:26:38 - later before we understand memory, or at least understand what we don't understand about it.
00:26:42 - So I think the behavior leads a long way ahead of the physiology,
00:26:47 - I think, and the circuit knowledge of what the circuits are doing.
00:26:51 - Right, but then let's look at it from a performance perspective.
00:26:54 - I mean, how many landmarks can I memorize as a bee? Okay.
00:27:00 - It depends on how you train them.
00:27:06 - I don't know if anyone has looked at that systematically, but certainly you
00:27:12 - can, in terms of discrimination, you can train bees to distinguish between horizontal
00:27:18 - versus vertical patterns.
00:27:20 - You can train them to distinguish colors, as I mentioned.
00:27:24 - You can train them to distinguish between odors, different scents,
00:27:27 - and you can train them to associate odors with colors.
00:27:34 - So you can have a bee come into a maze, it gets a whiff of scent,
00:27:38 - and then you can train it to say, if you smell lavender, then you go to a decision
00:27:42 - chamber, you have to pick the blue disc.
00:27:45 - If you smell lemon, when you go to a decision chamber, you've got to pick the
00:27:48 - yellow disc, for example, so they can learn to make those associations. Um,
00:27:54 - You can also train them to do kind of a delayed match to sample task.
00:28:00 - You're familiar with the delayed match to sample task? Yeah,
00:28:02 - of course. Just define it.
00:28:04 - So the idea is that the classical experiment is you flash a stimulus,
00:28:09 - let's say a color to a person, blue.
00:28:12 - And then later on, a few seconds later, they're given a choice between blue and yellow.
00:28:16 - So if you see blue as your sample stimulus, you've got to pick the matching
00:28:20 - stimulus with blue. What's the delay the bees can handle? About five seconds. Okay.
00:28:24 - And then after five seconds, it drops off rapidly or gradually? It drops off gradually.
00:28:30 - In about five, it's close to random.
00:28:33 - Okay. But they can also learn non-matching.
00:28:36 - So pick the stimulus that does
00:28:38 - not match. And they can also learn to match across stimulus modalities.
00:28:43 - So you can train them to do the matching task using sense and then expose them
00:28:50 - to a visual task on which they haven't been trained and they will do the matching.
00:28:54 - So, they've learned the concept of matching from the smell task and they're
00:29:01 - applying it to visual tasks. So, they generalize the rule. Exactly.
00:29:04 - Both for matching and non-matching. Yes.
00:29:07 - Okay. Isn't that nice? That's amazing. That's pretty amazing.
00:29:09 - That's really cool. Okay.
00:29:10 - Still don't know how it's useful in nature. Here we're treating these animals as a lab rat.
00:29:17 - I can't imagine that it's… Okay. A certain context, maybe when you land on a
00:29:21 - flower and you get some good reward there, you might want to seek out a similar
00:29:25 - flower which has the same color.
00:29:27 - So in that sense, there's a thing. But you see, in the lab experiment,
00:29:32 - you're not being rewarded on the sample stimulus.
00:29:36 - I find it hard to think of a situation in nature where this particular task has to be applied.
00:29:44 - But they can do it, they can do it, you see. Look, but why is that so hard?
00:29:47 - I mean, here you are, I'm Mr. B flying around.
00:29:50 - I'm visiting different flowers, different colors, different scents and everything.
00:29:54 - And now there might be contingent relations among these flowers.
00:29:57 - That's the rule, right? So maybe when I visit, let's say, a yellow neutral flower,
00:30:02 - I should not go to the blue one because I get absolutely nothing.
00:30:07 - Well, if I first go towards the yellow one and then to the red one that was
00:30:10 - hidden behind it, I get a reward.
00:30:12 - So now I can start to learn contingencies in my environment. Okay, that's sure.
00:30:16 - You could do that, but you've got to...
00:30:24 - Okay, let me say this one thing. I don't know if that answers your question.
00:30:32 - They can learn... It's called... What do you call it? It's symbolic delayed master sampling.
00:30:38 - So not just direct matching of stimuli, but saying, for example,
00:30:42 - if I see blue at the entrance,
00:30:46 - then I should pick the vertical grating versus the horizontal grating in one decision chamber.
00:30:54 - And then you can cascade this. You can go to a second decision chamber where
00:30:59 - you have a choice between two other stimuli, for example, a radial pattern versus
00:31:03 - a set of concentric circles.
00:31:06 - So if you see blue, you pick vertical and you pick a radial.
00:31:10 - If you see yellow, then you pick the other stripe orientation and the other pattern.
00:31:16 - So all those contingencies, they can learn those contingencies.
00:31:20 - Now, so you think that'll be useful in nature?
00:31:25 - For problem solving, I mean, these animals will fight themselves in complex situations.
00:31:31 - They have to navigate through, let's say, dense growth.
00:31:34 - There might be relationships between flowers. Yeah. So, chances are in nature,
00:31:40 - you probably won't find the stimulus changing in the same location.
00:31:44 - But you could certainly apply it to two different trajectories.
00:31:46 - For example, if I come here, if I come to this location, you know,
00:31:50 - I recognize this blue, blue something here.
00:31:52 - And that tells me how I should proceed. Whereas if I see something else here.
00:31:56 - Right. Moreover, it can pick up regularities among plants.
00:32:00 - Because let's say the plant from which you want to get the nectar,
00:32:04 - the flower, might not be easily visible from their altitude.
00:32:08 - But however, the tree just next to it is. Sure. Right. And maybe the flowers
00:32:14 - you like grow close to certain trees.
00:32:17 - This is completely reasonable. So now I can extract these regularities from
00:32:21 - my environment and immediately apply them.
00:32:23 - I agree. Well, thank you for pointing that out. How's my bee psychology going?
00:32:28 - I feel a lot more encouraged about what I'm doing.
00:32:32 - Great. We're so pleased to have you here. What got you interested?
00:32:37 - You're giving me a reason to live.
00:32:41 - Great. But what got you interested in bees in the first place?
00:32:44 - Oh, it was purely accident.
00:32:47 - Everything I've done in my life has been just not planned at all.
00:32:50 - So I did my undergraduate in electrical engineering, as you probably know, in Bangalore.
00:32:55 - And when I was doing my master's, my professor suggested, and that was,
00:32:59 - by the way, my master's was, and it was called Applied Electronics and Servo
00:33:03 - Mechanisms in those days.
00:33:04 - So it's mostly control theory and electronics.
00:33:09 - Too soft? I'm sorry.
00:33:12 - And with the Masters, my professor said, why don't you do something biological?
00:33:17 - Why don't you try to model a biological system using the control system theory
00:33:20 - that you've been studying?
00:33:21 - And so we decided to try and model the human eye movement system as a target
00:33:26 - tracking system, a moving target.
00:33:29 - So that's how I got interested in this thing. And when I went to the U.S.
00:33:32 - To do my Ph.D., I was looking for some project in the area of interface between
00:33:38 - biology and engineering, and it turned out the only person who was doing anything
00:33:42 - in that area was the person who was working on insect eyes. So I got into fly vision in that way.
00:33:47 - And then when I went to Zurich, Zurich in Switzerland was the,
00:33:51 - you know, one of the world's, you know, sort of leading areas in bee work.
00:33:56 - So there, there's a lady called Miriam Lehrer, who's unfortunately passed away
00:34:00 - now, who is the world's expert in training bees.
00:34:04 - She was with Rudiger Wehner? Exactly. So yeah, it was in Rudiger Wehner's lab.
00:34:08 - Yeah. So that's where I learned all about bees.
00:34:10 - So it was purely by accident, but it was a wonderful accident.
00:34:13 - They're just amazing creatures.
00:34:15 - Right. And so, but how long ago was that your first encounter with the bees?
00:34:20 - Well, if you ask me to give away my age.
00:34:24 - No, no, we won't go that far. Well, that would be, that would have been 1982
00:34:28 - when I was a young assistant professor.
00:34:31 - Okay. But then, so the bee is still your main target preparation in the empirical work?
00:34:38 - I would say so, yeah. We're starting to work a little bit with birds as well,
00:34:41 - but bees are the main thing.
00:34:44 - Thing yeah so so then there are two issues to explore right so so with respect
00:34:49 - to the to the b cognition we touched a little bit on this issue of memory um but now so is this sort of,
00:34:58 - this ability to extract these symbolic rules would you see that as really the
00:35:03 - the highest level of of of b cognition that can be achieved or other tricks in their cognition bag?
00:35:11 - Well, there are a few things which are really kind of striking.
00:35:15 - That's one thing. The other thing we were talking about the other day was maze learning.
00:35:20 - They learn to go through labyrinths and various kinds of labyrinths.
00:35:24 - But there's also the business of breaking camouflage and perceiving camouflage
00:35:32 - objects which they normally would not see.
00:35:35 - How does that work? Well, for example, you probably know this picture of this, uh.
00:35:41 - Dalmatian dog that's hidden behind a pattern of camouflage dots.
00:35:44 - It's a dog I never see. You never see it, right?
00:35:47 - But once someone traces the outline for you and you see that same image again, it pops out.
00:35:52 - You see the Dalmatian every time. I always say no, but it's true.
00:35:56 - Well, bees seem to have that property too.
00:35:59 - So if you can give them a hint, initially when you show them two camouflaged
00:36:02 - objects and try to train them to distinguish between them, they don't seem to do it.
00:36:06 - But if you give them a hint by showing them un-camouflaged versions of the same
00:36:09 - two objects, and you train them to discriminate those two, and then you show
00:36:13 - them the camouflage objects, and they can pick them out.
00:36:16 - And not only that, once they've learned how to break the camouflage,
00:36:20 - you can give them novel camouflage objects.
00:36:23 - And without the pre-training, they will do the task, learn the task.
00:36:27 - So you really taught them a different way in which to see the world.
00:36:32 - That's pretty impressive. Which is not bad.
00:36:34 - Do you think all these skills have given the bees an evolutionary advantage? advantage.
00:36:39 - Are there more bees than would be there otherwise?
00:36:44 - I wouldn't be surprised. I mean, I don't know to what extent it depends on how
00:36:48 - much predation there is and how many creatures are out there trying to eat these creatures.
00:36:54 - But that certainly is the fact that they can sting has certainly kept them alive for quite a while.
00:37:01 - And as you know, there's also these wonderful bee mimics.
00:37:05 - There are lots of insects which don't sting, which have evolved to mimic the
00:37:09 - bee because other birds will stay away from them because they look like a bee.
00:37:12 - They taste beautiful, but the birds just avoid them because they've been conditioned
00:37:16 - to avoid any bee. But that's more the sting, though.
00:37:20 - That's the sting. How are their cognitive abilities helping them?
00:37:24 - I think all I can say is it certainly helps them become more efficient foragers.
00:37:30 - I mean, this is where recent beautiful work that was done, not in our lab,
00:37:35 - but in the lab of a chap called James Neer, and he discovered that when a bee comes back and.
00:37:42 - Dances and to advertise a food source and another bee is watching this dance
00:37:46 - and it has been to that food source and has had trouble it's been attacked by
00:37:51 - a spider for example this bee will then head butt this dancing bee and stop
00:37:56 - it from dancing because it's a dangerous food source and this
00:38:02 - stopping is very target specific so it's it stops this dance only when the bee
00:38:09 - is advertising adding that particular food, so nothing else.
00:38:12 - And also, only if this bee has come back badly damaged.
00:38:17 - If this bee has had a fight with a spider and it had actually won, no problem.
00:38:22 - Okay, right. So I think that's getting to the point where these creatures are,
00:38:28 - I would say, almost human.
00:38:30 - That's pretty impressive. I never stung a spider though, but there's a work on that.
00:38:35 - But the thing is that the memory of these bees that would be providing,
00:38:40 - let's say, the core infrastructure for these cognitive capabilities seems to
00:38:46 - have, let's say, varying time windows in which it operates and is stable.
00:38:50 - So how many, let's say, would the distinction between the short and long term
00:38:55 - memory be sufficient to describe bee memory or do you see more stages?
00:38:58 - Well, there's also working memory.
00:39:00 - So this delayed master sample is a kind of working memory, right?
00:39:03 - And that lasts about five seconds.
00:39:05 - And there is the short-term memory, people say, which lasts about an hour.
00:39:09 - And then beyond that, it gets put into long-term memory.
00:39:13 - Now, exactly where the short-term memory resides, is it in the mushroom bodies
00:39:17 - or is it somewhere else? Right. No one really knows. I mean, the...
00:39:22 - It's very sad. I mean, the main problem, I think, is, as usual,
00:39:26 - the funding for insect work is not as good as it is with vertebrates.
00:39:29 - So, really, the physiology and even the anatomy. Well, the anatomy,
00:39:33 - thanks to people like Nick, is really doing very well. But the physiology is really suffering.
00:39:37 - But now about the stability of this memory. For instance, if the bee comes back
00:39:41 - and it dances, does that compromise the memory of that location,
00:39:45 - you think, in any way? Good point.
00:39:49 - A good point. I always looked at that. I mean, why should it?
00:39:52 - Well, it has to recall it, right? So the memory might become instable because
00:39:56 - it has to be recalled, replayed in some form.
00:39:57 - So you're saying every time you recall, you might lose the memory trace?
00:40:00 - Well, there are some theories of memory that go in that direction,
00:40:04 - right? Because it means recall would mean you have to make the memory again
00:40:08 - accessible and you pay a price for that. Oh, yeah, that's a good point.
00:40:13 - I don't know. I wouldn't be able to answer that question. But as you probably
00:40:16 - know, the dance is done only after the bee has wilted the foot several times.
00:40:22 - And also, by then, this particular experienced bee will not even be relying
00:40:26 - on its own dance information to get to the food source.
00:40:30 - It'll be using the sequence of landmarks and things that it's learned to go along the way.
00:40:34 - Right. So, that's why, although on a cloudy day, when the sun is covered and
00:40:41 - the whole sky is cloudy and there's no polarized light or sun, there's no dancing.
00:40:45 - The bees don't dance. But the bees that already know the food source will continue
00:40:49 - to forage because they don't need that information anymore.
00:40:52 - I mean, it's like you and I, you know, we know a familiar place like this beach
00:40:56 - that we're going to. I don't know it, but you know it.
00:40:58 - But I would be using vector information because you've given me that vector
00:41:01 - information, but you won't be
00:41:02 - using that. We're just using a sequence of landmarks, right? That's right.
00:41:05 - That's how these bees do it. The experienced bees do it that way.
00:41:07 - So the information hidden in the dance is really very crude. Yes.
00:41:10 - Crude information about the environment. And it's kind of dispensed with,
00:41:13 - you know, once the bee has advertised the food source and enough recruits have
00:41:16 - been collected, they basically forget about it. The dance is not done anymore.
00:41:20 - Okay. But when a new food source comes up, then, of course, the dancing starts again.
00:41:25 - Okay. But so now another part of your work, which might also be sort of have
00:41:30 - its roots in your engineering background, is you have been mapping a lot of
00:41:33 - this understanding of insect vision and behavior onto machines, onto robots.
00:41:38 - Yeah, that again, you know, that's also something that we didn't think of about ourselves.
00:41:43 - And it sounds a bit stupid because, you know, this first thing that we published
00:41:47 - on bees navigating down corridors, we didn't even think of it as anything that
00:41:53 - had potential engineering applications.
00:41:54 - But then after that work got published, a number of labs started to build robots
00:41:58 - that navigated down corridors using the same principle.
00:42:01 - And so we were actually sort of rather, you know, latecomers to this thing.
00:42:05 - And in fact, I wasn't even really pursuing that a lot until...
00:42:10 - A few US-based military funding agencies kind of just tapped us on the shoulder
00:42:14 - and said, hey, you know, would you like to work on this? And here's some funding.
00:42:19 - So that's how we got into it, really, ourselves.
00:42:21 - But now tell me, so to what extent have these principles really generalized successfully?
00:42:27 - What can you really achieve? How close is it to what you see in these insects and so on?
00:42:31 - Yeah, so as I was saying briefly the other day, so we're not really doing what
00:42:36 - you would call as biomimesis.
00:42:38 - So we're totally not building a compound eye.
00:42:43 - I mean, it's probably a good idea to do that. If you have the expertise and
00:42:47 - the technology, you'll probably learn something nice.
00:42:50 - But our idea is to implement the principle. So instead of using a compound eye,
00:42:54 - we started out by using just a single camera, off-the-shelf camera,
00:42:58 - but building a specially shaped reflecting surface.
00:43:02 - Just a mirror, but a specially shaped mirror.
00:43:06 - So you can do, you can play, we like to do things with mirrors. We're fond of mirrors.
00:43:11 - So with a mirror, for example, you can have either a spherical mirror.
00:43:14 - The trouble with a spherical mirror is that the radial gain is not constant.
00:43:21 - So what happens is if you look at the world with a hemispherical mirror,
00:43:25 - then the central part is magnified and the peripheral part is compressed, right?
00:43:31 - So you don't have uniform gain, elevational gain, as you might say.
00:43:34 - So we tailor the shape that produces uniform elevational gain,
00:43:38 - which is kind of useful because then you don't lose resolution.
00:43:41 - You make, what do you say, optimum use of the resolution, no matter where you're
00:43:45 - looking, right? So we use that on our aircraft with a standard camera.
00:43:49 - And that functions almost as well as a compound dye.
00:43:52 - There are a few blind zones, of course, like directly behind the camera you
00:43:56 - can't see, and then behind the mirror you can't see, but you've got a good field of view there.
00:44:00 - So that's what we do. So we implement compound panoramic vision in that way.
00:44:06 - And we don't build a flapping wing vehicle because that's too hard and we're not experts.
00:44:10 - We leave that to people like Mike Dickinson and so on. So we just build a vision
00:44:15 - system that uses some of the.
00:44:19 - It's a global principle that we discover from insectivision.
00:44:22 - So, for example, the finding that you need to measure optic flow in the two
00:44:25 - eyes and balance them in order to fly down the middle of a corridor.
00:44:31 - But the actual computation of the optic flow, again, we don't do it using the
00:44:36 - biological algorithms because we find that the biological algorithms don't do the job for us. Why not?
00:44:40 - Because they don't signal velocity reliably. They confound image velocity with spatial frequency.
00:44:50 - They're not robust to changes in contrast. They're just a mesh.
00:44:55 - Well, just to clarify, though, when you say biological algorithm,
00:44:58 - you really have in mind the Reichardt model or models that people have made.
00:45:03 - People have made. You don't actually know what is going on inside the fly's
00:45:07 - brain. We don't. Yeah, exactly.
00:45:08 - The true biological algorithm obviously works, right?
00:45:12 - Yeah. The real biological algorithm actually works. It still hasn't been,
00:45:16 - yeah, we don't know exactly what's happening in terms of the nervous system to generate that.
00:45:19 - But behaviorally, the animal behaves as though it is sensing velocity very, very robustly.
00:45:24 - Because what I found interesting, I mean, it sounded a bit normative what you were saying, right?
00:45:29 - Because in some sense, the biological algorithm as a biological algorithm must
00:45:33 - be incredibly precise and robust.
00:45:35 - Because, you know, it's computed in this really minuscule computational system.
00:45:39 - When I say biological algorithm, I mean I mean the known biological algorithm.
00:45:42 - Our interpretations are hypothesized. Hypothesized, or the recordings from neurons,
00:45:47 - the models of neurons that,
00:45:49 - that respond to motion, do not seem to do the job. So that's perhaps the limitation
00:45:53 - of the people who did the modeling rather than of the fly. Well, yeah, okay.
00:45:57 - In a way, they're trying to model the neuron's response, and they're probably
00:46:00 - right. But they have done a bad job.
00:46:02 - Well, this particular neuron may show that response, but maybe they're not looking at the right neuron.
00:46:07 - But there's something interesting here, Srini, that I think,
00:46:09 - of course, I understand you want to defend your colleagues, okay? That's all right.
00:46:13 - But the point is that, indeed, people have taken the physiology,
00:46:16 - given that some sort of functional interpretation.
00:46:20 - But if you apply it to your aircraft, it is exposed to just different conditions
00:46:23 - because you're not tying this airplane to a table and show it fixed stimuli.
00:46:28 - It's now flying around. And the input sampling, the dynamics of the stimuli
00:46:33 - has really changed completely.
00:46:34 - And that's where these algorithms, of course, collapse because they have been
00:46:38 - calibrated in highly controlled, rather artificial situations.
00:46:41 - So I think an interesting consequence of this observation is maybe that both
00:46:46 - the experimental context, the experimental paradigms have sort of been biasing
00:46:50 - our interpretation of what these systems do.
00:46:52 - And on top of that, possibly, our algorithmic function interpretation has just
00:46:57 - maybe been completely misguided.
00:46:59 - This is actually a consequence of your work.
00:47:02 - Yeah. I mean, the other way, I suppose, the way we're doing it presently is
00:47:07 - to put on a different hat completely.
00:47:09 - So when I say I want to measure optic flow, I simply, you know,
00:47:13 - We've developed a whole bunch of just machine vision-oriented algorithms, which work well.
00:47:19 - I mean, they give you the right answer. They've probably got nothing to do with the biology.
00:47:22 - So that's how we do it. But in the future, maybe there's a learning method or a...
00:47:31 - I don't know, genetically, what's the word, genetic algorithm-based approach,
00:47:35 - which will give us something that produces a circuit that measures velocity accurately.
00:47:41 - Because there's another aspect to this, right? That in some sense,
00:47:44 - what you're also using is computational hardware that has certain capabilities.
00:47:48 - Yeah. And you're exploiting, let's say, algorithms people have developed using this kind of hardware.
00:47:53 - But biological hardware might have to optimize different parameters.
00:47:58 - Exactly. Than the engineered hardware, because in the case of the bee,
00:48:02 - it must be flyable, it must be very compact, it must be energy efficient,
00:48:05 - and so on. It's optimizing so many different criteria.
00:48:08 - Right. I agree completely, and it's not necessarily tuned to exactly what we want.
00:48:12 - So, therefore, if you talk about biomimetics, I think it might also be a good
00:48:17 - way to actually benchmark and validate our understanding of the real biological system.
00:48:21 - I like to call it bioprincipics rather than biomimetics. So,
00:48:25 - you abstract the principles.
00:48:26 - You don't lavishly copy the biology. Just abstract the higher order principles
00:48:30 - and try and implement them, right? Sure.
00:48:33 - And then test our ideas that way about whether it's using this principle or not. Absolutely.
00:48:38 - But earlier you said yourself that's not really what you do with your airplanes.
00:48:45 - With the aircraft, no, what we do is we want to see, okay, we know the fact
00:48:50 - that insect is now using optic flow to control its landing.
00:48:53 - Can we get an aircraft to use optic flow again to control its landing?
00:48:57 - Representing the exact way in which optic flow is computed.
00:48:59 - We don't know in an insect, but we'll use our own way. We'll use our own engineering-based way.
00:49:04 - And that works. And that works. But this is funny because in some sense,
00:49:09 - in the literature, people would make you believe that they do know how flies
00:49:13 - or flying insects compute motion. They say it's the Reichert Correlator.
00:49:16 - So why do you say we don't know how it's computed?
00:49:19 - Because it doesn't fit the behavioral data.
00:49:21 - Okay, tell me. See, the thing with the Reichert Correlator, The only thing it
00:49:25 - can reliably tell you, really, if it comes down to it, is the direction in which something is moving.
00:49:30 - Beyond that, the information is very ambiguous.
00:49:34 - So if you want to use it as a bang-bang controller to control the direction,
00:49:39 - keep flying forward, right?
00:49:41 - So if the world moves to the right, it means you veered off to the left,
00:49:45 - and you generate a compensator. You're off to a turn back to the right.
00:49:48 - For doing that kind of bang-bang control, it's great. It's very reliable.
00:49:52 - But when you're turning to the right, you want to know how rapidly you're turning.
00:49:55 - It did not give you a reliable answer. And then there's another thing that always
00:49:59 - worried me about the Reichardt Correlator is that in some sense,
00:50:02 - it tells us that neurons can multiply.
00:50:05 - And biophysically, I find it always difficult to comprehend how you could do that.
00:50:08 - Well, actually, we had a cute idea that I published as part of my PhD thesis
00:50:12 - a long time ago, which I don't think anyone's really picked it.
00:50:14 - You could do it very easily.
00:50:15 - Very easily. Okay, just a quick...
00:50:18 - A coincidence detector, a neuron, and two trains of random spike trains coming
00:50:22 - in, two different frequencies, right? They're random, they jitter.
00:50:26 - So the probability of a coincidence is proportional to F1, frequency 1, and frequency of F2.
00:50:31 - So the spike rate at the output will be proportional to the spike rates in the
00:50:35 - input, if you assume randomness. Right.
00:50:38 - If there's no randomness, if they're periodic, then that doesn't work because
00:50:42 - you could have either perfect, you know, synchrony or no synchrony.
00:50:45 - So certainly evidence for this kind of... But the moment you have noise,
00:50:47 - and noise is actually helpful in this case, you will get a beautiful product.
00:50:52 - I don't know if anyone's found a neuron like that, but I'd love to see a neuron
00:50:57 - because we published that a long time ago.
00:51:00 - We also briefly discussed the possibility there are other sensors,
00:51:03 - like it is actually sensing drag in the air or something like that.
00:51:07 - Right, right, right. Possibly.
00:51:08 - Yeah, certainly. Combining the visual information with other sources.
00:51:11 - There are the antennae doing all kinds of things. as we were saying,
00:51:17 - they're probably tactile sensors as well.
00:51:19 - By the way, just as there's a beautiful rat whisker story we heard just now,
00:51:25 - bees look like when they come in close to a surface to land,
00:51:30 - when the surface is oriented nearly vertically, they're using their antenna
00:51:33 - to make the first mechanical contact.
00:51:36 - And also, it seems like the antennae tend to be perpendicular to the surface.
00:51:42 - As they're coming in and hovering, they have a perception of the surface slant,
00:51:46 - And you can even fool them by producing optical illusions which simulate different
00:51:51 - surface lands by having texture gradients.
00:51:53 - So obviously the eye is, the visual system is analyzing surface orientation.
00:51:58 - And maybe that's one where we could apply this model too. Mm-hmm. Yeah.
00:52:03 - Excellent. So then you haven't said much about the birds yet,
00:52:08 - and still I want to hear something about it.
00:52:09 - Oh, we're just starting, yeah. Okay. But at the level of intuition…,
00:52:14 - So now we talked about the bee, we've talked about this sort of bioprincipics
00:52:18 - and extraction of core design principles of the bee brain.
00:52:21 - Would you believe that you will also find some of these design principles back into the bird brain?
00:52:26 - At least so far, we found a couple of similarities.
00:52:30 - One again, and this is, again, we haven't tested a whole range of birds.
00:52:34 - But if you take one of the standard birds in Australia, it's called the budgeriga.
00:52:39 - Do you get budgies here? In the animal store, yeah. In the pet stores.
00:52:44 - There it's a kind of an iconic Australian bird, native bird.
00:52:48 - Anyway, there, if you fly them down a tunnel, they show very similar behavior to what the bees do.
00:52:54 - So we haven't been able to actually physically move patterns in these tunnels
00:52:57 - yet. We're starting to do that now with a long tunnel.
00:53:00 - But you can manipulate the optic flow by having static patterns,
00:53:03 - which are, for example, horizontal stripes on one side and vertical stripes on the other side.
00:53:08 - Then you imbalance the optic flow that way, and they behave in exactly the same way. Right.
00:53:13 - This is interesting in the context of our notion of convergent evolution and
00:53:19 - understanding engineering principles by looking at convergent evolution.
00:53:23 - Clearly, if the birds are using the same optic flow mechanisms as the bees are,
00:53:28 - it will have to be convergent simply because the common ancestor didn't fly.
00:53:33 - Yeah. The other thing that seems to be uniformly true in many species,
00:53:36 - including humans, is the fact that motion perception is largely colorblind.
00:53:40 - So humans, as you probably know, although we have beautiful color vision,
00:53:44 - the motion sensing system is almost colorblind.
00:53:46 - It's driven only by the luminance pathway, red plus green.
00:53:49 - If you look at the bee, again, it's colorblind. It's driven only by the green
00:53:53 - receptor, all of the motion sensing.
00:53:55 - And if you look at the spectral sensitivity function of the green receptor,
00:53:59 - it sits bang on the sum of, if you take the red plus green cones,
00:54:04 - into our luminous pathway, it sits exactly on top of that.
00:54:08 - So it's as though that system is adapted exactly to our environment, exactly the same thing.
00:54:12 - And now we're finding that even birds are like that. At least these buzzard
00:54:15 - guys are behaving as though their perception of motion is also colorblind,
00:54:19 - although these birds have even better color vision. They're tetrachromatic, right?
00:54:24 - So the system is going out of the way to make the motion detection colorblind.
00:54:28 - Exactly. Do we understand why? But wait, but there can be… Is that more efficient? Yes.
00:54:32 - I think the common notion is that the early creatures, and then again,
00:54:37 - correct me because I'm not an evolutionary biologist, is that most of the early
00:54:41 - vision systems did not have color.
00:54:43 - If you needed to have basic motion sensing just in order to move around.
00:54:48 - And not bump into objects and just navigate safely, and then later on when flowers
00:54:56 - evolved in the scene and it became important to recognize objects based on color,
00:55:00 - that's when color came in.
00:55:02 - So, many people, and this is still a hotly debated topic, but many people think,
00:55:06 - you know, color vision sort of co-evolved with bees and flowers at the same time, pretty much.
00:55:10 - So, there's no fundamental need to have color to begin with.
00:55:13 - But now, on this issue of convergent evolution, right?
00:55:17 - So, you could also argue that there can be, in the end, a common ancestor that
00:55:22 - just had to move and crawl, right, in a visual world.
00:55:26 - And that could have picked up these kinds of responses to motion.
00:55:30 - You don't need to fly to pick that up. So if you just have to speculate, how would you see this?
00:55:36 - Isn't it convergent evolution to a similar solution between flying insects and birds?
00:55:40 - Or just really a very ancient common ancestor just crawled around?
00:55:45 - I would say a very ancient common ancestor is my guess, yeah.
00:55:48 - Primitive things like photo taxes, going towards a bright light source is probably
00:55:51 - a very fundamental thing.
00:55:53 - Going towards something that smells good is probably a good thing.
00:55:57 - But on the other hand though I mean hummingbirds and bees fly in the same way
00:56:01 - and the common ancestors didn't fly so not everything is convergent we have
00:56:07 - to be a little careful about that in terms of optic flow,
00:56:11 - do you think that it is indeed in the common ancestors are the parameter,
00:56:19 - regimes right even though it had to sense motion do the birds and the bees have
00:56:24 - a a more refined sense of the speed.
00:56:29 - Than, for example?
00:56:31 - Some crawling primitive animal, let's say.
00:56:35 - Oh yeah, it would have a more refined sense. Like a snail, for instance,
00:56:39 - you know? Yeah. It will respond to moving visual stimuli.
00:56:42 - See, my guess is that...
00:56:45 - Do flying creatures have to have more precise flow detection?
00:56:52 - Oh, sure, yeah. Yeah, they need to have also, depending on the speed of flight,
00:56:56 - it seems like certainly the nervous system is very different,
00:56:59 - right? The dynamic properties are very different.
00:57:01 - Right from the photoreceptors, you have a much higher flicker fusion frequency
00:57:05 - if you've got a fast-flying creature.
00:57:09 - And I would say even the motion-sensing units are tuned to their speed.
00:57:13 - So I would say that a thing like a primitive worm would probably have, again, the same kind of,
00:57:18 - Same kind of basic structure for motion
00:57:20 - detection, because it's not very complicated when you think about it.
00:57:23 - If you just want to tell which direction something is moving,
00:57:25 - you just need to have one inhibitory synapse, right, between one photoreceptor and another one.
00:57:30 - And that could have evolved as early as lateral inhibition, you know? Exactly.
00:57:34 - And so you just modify that a little bit. Well, actually, for any sensor,
00:57:38 - it is sort of the endocannulas.
00:57:40 - But you could even go to other modalities. If you go to some mechanical sensing
00:57:42 - or chemical sensing, all you want to do is measure a difference.
00:57:46 - That's what this is. What is then generalized to some optic sensor.
00:57:50 - You just measure a difference.
00:57:51 - But as you were also pointing out, the Rijkaard detector, which does that, is not adequate.
00:57:57 - It's not adequate. So therefore, there is something more. Sure.
00:57:59 - Yes, there is something in common.
00:58:01 - But there's more. Yes. And that is really necessary for this behavior.
00:58:07 - And that's where maybe you even have parallel pathways. My guess is that,
00:58:12 - at least in the B, there must be several parallel pathways that are sensing
00:58:15 - different aspects of motion.
00:58:17 - For example, this optical avoidance and flying down the middle of a corridor,
00:58:21 - that motion sensing seems to be non-directional. It's not directional at all.
00:58:25 - So if you think about it, if you want to avoid an obstacle, you don't really
00:58:29 - care which way the thing is moving. You just want to avoid it because of its high speed, right?
00:58:33 - So maybe it's computationally simpler to just do some non-directional motion
00:58:37 - sensing rather than compute motion direction.
00:58:39 - It's just not necessary. And this thing has completely different properties.
00:58:44 - You know, I think things are changing now. I probably shouldn't say this,
00:58:48 - but as long as Reichardt was alive, there was a big school of tubing and following the Reichardt model.
00:58:54 - But I think now things are starting to unravel.
00:59:00 - Okay. But so to finish up, there are two questions that we actually pose to
00:59:05 - everybody we interview for the podcast.
00:59:07 - So there But here you come out of engineering, discover the bee,
00:59:11 - do all this fantastic work, have it fly now on these airplanes, move to the bird.
00:59:18 - So, you've been exposed to these different disciplines, trying to extract these
00:59:23 - engineering principles of biological systems.
00:59:25 - So, if we would like to follow in that tradition, what's the law of Srini we
00:59:29 - should adhere to? What's Srini's law? Srini's law? Yeah.
00:59:33 - Follow your heart. Don't worry about where your next job is going to be.
00:59:43 - No, just follow your heart. And if you enjoy what you're doing,
00:59:46 - things will come to you naturally.
00:59:48 - Do you want me to say briefly what my next thing, I don't know if I will ever
00:59:52 - get around to doing this, but what my next, I would love to be able to look
00:59:57 - at higher emotions in simple nervous systems.
01:00:02 - Things like, for example, joy, disappointment, fear, anger, pain.
01:00:13 - I think there's a whole thing there, and I believe there's an entire continuum
01:00:17 - and there's no such thing as, you know, I find it hard to believe that invertebrates,
01:00:21 - for example, don't feel pain, whereas vertebrates do. They know hard and fast.
01:00:24 - So would you speculate then that
01:00:26 - these animals would qualify for a very primitive form of consciousness?
01:00:31 - I would think so. I would think so. Especially, you know, that headbutting that I mentioned to you.
01:00:36 - That's, you know, I find it hard to believe that's all done in a purely a reflexive
01:00:40 - way, especially because it's so subtle, right?
01:00:45 - Well, headbutting is not that subtle, I have to say.
01:00:47 - No, but the things that control when it headbutts and doesn't headbutt,
01:00:50 - it's very precise, right?
01:00:52 - I mean, the fact that it's exactly that food source is being signaled,
01:00:56 - and also on how dangerous the predator there is.
01:00:59 - Right. I mean, you can't say this is all being done by some stupid automaton, right?
01:01:03 - Well, I think to any of us who've actually watched flies and things closely
01:01:08 - behaviorally, there's no doubt in our minds that they are constants.
01:01:11 - There's a lot going on, you know. For example, you jab a cat or a dog and it
01:01:16 - flinches and you say it feels pain.
01:01:17 - And you jab an insect and it again does the same kind of reaction,
01:01:20 - but you say oh, it can't be pain because it's an insect. It must just be a reflex.
01:01:25 - How do you know, right? This is a great topic for another podcast interview
01:01:29 - that we're definitely going to have in the future. Sorry, I didn't want to launch
01:01:31 - you off on this. No, no, this is good.
01:01:33 - But then the final question for me is really, so since Since Partha is so successful
01:01:38 - and generates all this money, we can travel around the world on his budget,
01:01:41 - you know, without any trouble.
01:01:42 - So five years from now, we're going to go visit your lab and we're going to
01:01:45 - remind you of a hypothesis that you generate today that you feel most passionate
01:01:50 - about and that you claim will come out five years from now. So what's that hypothesis?
01:01:55 - Oh, my hypothesis will be that insects feel pain. Right.
01:02:02 - You're really going to work on that? Oh, look, if I get the funding,
01:02:05 - if I get the funding, I'd love to.
01:02:07 - Okay. I've got some ways in which they can be tested, I think,
01:02:10 - which will be a little more telling than just jabbing the insect and saying
01:02:14 - it's a reflexive thing because that doesn't convince people.
01:02:18 - It convinces me, but it doesn't convince other people.
01:02:20 - But I think there must be ways. You can never make a conclusive proof.
01:02:23 - But again, as for example, now that they convinced that fish feel pain,
01:02:27 - for example, do you know how they finally did it? No. In fact, it involved a bee.
01:02:32 - They took a bee sting. They took a sting out of a bee and stung the lip of the
01:02:37 - fish and the fish twitched.
01:02:39 - And they said, aha, the fish feels pain. It seems like a very primitive,
01:02:42 - unsophisticated experiment.
01:02:44 - But the world was at the right time to accept that at that point.
01:02:47 - And they said, aha, fish feel pain.
01:02:49 - And now from now on, we've got guidelines for experimenting with fish.
01:02:53 - So that means now we have to slap a bee with a fish and see if it twitches.
01:02:57 - I think that's what you're proposing now. Yeah, the fish will probably twitch, right?
01:03:03 - Because the bee will have stung the fish. And the bee doesn't feel the pain, obviously.
01:03:10 - You proved two things there. The bee doesn't feel pain and the fish feels pain.
01:03:16 - I'm so happy we resolved that issue. Srini, thank you very much for this wonderful
01:03:20 - interview. Thank you. Thank you, Paul.
01:03:22 - Thank you. The CSN Podcast was produced by the Convergent Science Network of
01:03:29 - Biometics and Biohybrid Systems,
01:03:32 - a project funded by the European 7th Research Framework Programme.
01:03:37 - Music.

CSN Podcasts

Mandyam Srinivasan on honeybee cognition and waggle dance

Season 2012

Description

CSN Podcasts

Episodes

Peter Gärdenfors on conceptual spaces and knowledge representation

Friedemann Pulvermuller on word meaning and embodied semantics

Nick Strausfeld on brain evolution and cambrian explosion

Moshe Bar on proactive brain and prediction

Guenther Knoblich on joint action and entrainment

Timestamp

Be the first to leave a comment

Leave a commentCancel

Legal

Follow us on

Mandyam Srinivasan on honeybee cognition and waggle dance

Description

Episodes

Timestamp

Be the first to leave a comment

Leave a commentCancel

Tags

Legal

Follow us on

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Cancel subscription