Prologue:
Honeybees swarm in late spring and summer, when home starts to get crowded. 2/3 of the hive (like ten thousand worker bees plus queen) fly off to find a new home. They hang like a beard in a cluster on some branch while choosing new home. This book is about how they decide which home to go to. Why write the book?
- Condense the research on honeybee decision making over the last 60 yrs
- Has intriguing similarities to primate brain decision making process, no interesting to non-social animal scientists too
- Help improve human decision making
Chapter 1: Introduction
We’ll be talking about just one type of honeybee, Apis Mellifera. Queen doesn’t tell the hive what to do, she just dispenses like 1,500 eggs every day. The only thing she decides is whether to rear additional queens. She secretes queen suppression hormones, workers pick this up and spread it around, so all workers know queen is fine, no need to rear another one. When they swarm, old queen flies off, 1/3 of hive that remains behind rears a new queen. When they swarm, they only go like 30 meters from the hive, hang out for a few hours or days, disperse several hundred house hunters to survey 30 square miles for potential new house sites. Thoreau said groups don’t live up to their smartest individual, but degrade to their stupidest. But maybe by studying honeybee behavior, we can figure out how to make group decisions that are way smarter than any one individual could have been.
It all started in 1944, when Karl von Frisch discovered that honeybees dance to tell the hive about food sources (got the Nobel for it). Bee makes a circuit, waggling on the way, then walks back to where she started, and might do another circuit, for seconds up to minutes. It’s dark in the hive, but the worker buzzes while she waggles. The amount of time she waggles is proportionate to how far away the food source is (1s of waggling=1000 meters). Angle of waggle run, relative to straight, is angle relative to direction of the sun. Lindauer was his student, and he made a lot of seminal discoveries in this field. Drafted into Hitler’s workforce, but sent home when he got shrapnel in his leg, saved his life probably. He saw von Frisch speak, and saw a new type of humanity in science, where people create rather than destroy, and decided to become a biologist.
Lindauer began to see dancing bees on top of swarms, over other bees. They’re not carrying pollen, or regurgitating nectar, also they’re really dusty. What are they up to? House hunting, they all advertised different sites at first, then increasingly smaller sites, and a few hours before they flew, only one site, which Lindauer followed them to, and found that it corresponded with the site they were recently dancing about. So I guess humans can read bee dances?
Chapter 2: Life in a honeybee colony
Today there are 20,000 species of bee! All from a common ancestor vegetarian wasp from the Cretaceous period. Most bee species are actually solitary. But we’ll talk about group living honeybees. The queen, in the first weeks of her life, flies out of the hive and mates with 10-20 males in the area. She stores this sperm for the rest of her life in her spermatheca, about 5 mil sperm. With each egg she lays, she decides whether to dispense a few sperm to fertilize it (for a female) or not (a male). Whether egg grows into worker or queen depends on if the queen deposits the egg into a regular cell or a special large queen cell. Drones don’t help with hive work, but once mature, they go out every day to search for queens to mate with, they tend to congregate in certain areas, no one knows why. So they contribute critically to the hive’s evolutionary success by passing on its genes. Honeybees are individuals, but also superorganisms, because they do lots of things collectively, like regulate temperature and CO2 in the hive and make decisions together.
Most social insects go dormant in winter, but not bees. They huddle and take turns shivering their flight muscles, thus producing heat. They stockpile like 40 pounds of honey in summer, and this helps them get through the winter. So ideally they need about 4 gallons of space in a nest cavity. Bee keepers house bees in 36 gallon containers, so bees gather way more honey than they actually need.
Like trees, bee hives require cross-fertilization to avoid inbreeding problems. Only 25% of new bee hives survive their first winter, but 80% of established hives survive winter, probably because they had a some honey leftover from previous year.
Due to a combination of factors in April or so (colony crowding, plentiful food reserves, lots of foraging available) hive decides to prepare for a swarm, starts raising 10 or so new queens. The current queen is fed less and less, stops laying eggs, is shaken continually by workers so she has to stay on the move. She loses 25% of her body weight, so now she’s able to fly, normally she’s too heavy. Workers on the other hand, eat enough honey to increase their body weight by 50%. The wax glands on the underside of their abdomens also get huge, you can see wax scales visibly on their bellies, in preparation for making tons of new comb at new site. Lots of the workers also get really lazy and just lie about. Though some go out to hunt for potential nest sites. At some point, they come back and start piping against the lazy sisters with a thorax vibration pressed against them. These lazy sisters start shivering their muscles to a flight ready temperature. Once the scout bees contact enough warm bees, they know it’s time to buzz run (running while buzzing, bumping lazy bees). Then they swarm.
The stay at home bees now don’t have a queen, but lots of new workers are emerging and within a few days, they’ve replenished their numbers, and a new queen has emerged. This new queen announces her presence by piping in short bursts against the combs. All workers stand quite still while she does this to minimize background noise. But the other virgin queens still in cells (kept in place by the wax caps the workers are not removing) quack to let the queen know she has competition. At this point, the queen might decide to do a secondary swarm, might be safer than staying and battling all the emerging new queens. This might happen a few times as new queens emerge, until hive is so depleted it can’t support any more swarming. Now they just let the rest of the queens, if any, emerge. Now, if any emerges, she’ll chew a hole in ay remaining queen cells and sting occupants to death. If two emerge at once, they battle to the death, trying to seize and sting each other. Eventually one queen remains, she goes on mating flights, then starts egg laying.
Chapter 3: Dream home for honeybees
Earliest record of humans keeping bees is from an Egyptian relief from 2400BC.
So now Seeley graduates from Harvard, returns to the lab in Ithica where he’s been working summers to continue studying bees. First he wants to observe several wild hives to see what they have in common, what do bees like in a home? Climbs trees with ladder while all bees are in hive (right before sunrise), spoons calcium cyanide in there (which reacts with moisture in the air), and plugs the hole with rags before the gas gets out and kills him. Chop down the tree, cut out the hive part, get it back to the lab.
So turns out, bee homes (based on 21 hives) were about 45 liters. Nest openings were all a single knothole or crack, about 10-30 square cm, typically low down on the nest and south facing.
So now he starts making experimental nest boxes to see which ones swarms preferred. They selected them based on several attributes: entrance size, entrance direction, entrance height, cavity volume, and they like it when there’s comb already in there. So it looks like 40 liters with a small ish entrance, several feet off the ground, is ideal.
He learned to make artificial swarms from his PI (shake bees into an empty container and feed them with a lot of sugar syrup. He starts taking swarms out to Appledore island off the coast of Maine. No bees are there, no trees either, so they have to concentrate on the artificial nest boxes he provides. He set up the nest boxes with a red filter on one side so he could see in but it would be dark to the bees.
When scout first sees a potential site, she goes in and out a bunch of times, if it’s good, she goes home and waggles about it, and keeps visiting that site. She paces the inside, and does really tiny flights also. With subsequent visits she penetrates deeper and deeper. She covers the whole inside surface eventually.
So how do scout bees measure the inside volume? Seeing, walking? Turns out they’ll use vision if there’s enough light and they can’t walk around on it (fluon coated), but in nature, they usually measure it by walking. Using a bee treadmill (cylindrical nest box that he can rotate with the bee or against to change the amount of walking she does) showed that if she walked more, she thought the box was better, recruited more scouts to look at it.
Chapter 4: Scout Bees Debate
In the 1970’s, he finally has cheap access to video equipment, so he can now have a comprehensive record of all the house hunting scout’s dances, and number tags glued to the back of each individual bee. About 3-5% of bees in a swarm participate in the debate dance for homes. Worker bees only live 3-5 wks in active months, so many generation of bees pass when no one has to be a scout. But once they’re needed, they spring into action. Some foragers turn into nest scouts, but not all, they carry certain genes. Gene Robinson showed that the offspring of some drones are more likely to be foragers than others for example. Foragers know to turn into scouts when they’re homeless but not hungry. That’s why to make an artificial swarm that sends out scouts you need to keep the bees really well fed for several days.
At first lots of bees are dancing for different sites, this reduces, until finally, everyone is dancing for one site.
Chapter 5: Agreement on Best Site
So we know bees choose one site unanimously, but is it the best one? Turns out yes. They’re choosing between 24 sites on average in natural swarms, but unlike humans or other animals, they’re not using heuristics to make the choice. They’re deeply investigating lots of different options (at least 6 attributes of each site). So they’re actually using all the info, no filter.
So Seeley goes back to Appledore island, sets up various nest boxes which differ in goodness, and see which one swarms choose. Set up 4 acceptable and 1 ideal nest site. In 4 out of 5 trials, the best nest box was chosen.
Later did an experiment to see if larger nests help colonies survive the winter. Installed 2 swarms in test boxes, one large and one small, large did better.
Chapter 6: Building a Consensus
A scout bee discovers a site, she recruits other scouts, who also go check it out to make an independent evaluation, then they might also dance for it. How do they reach consensus?
Scout bees waggle more enthusiastically based on the absolute goodness of a nest site, so it turns out they can detect that, not just relative goodness.
Seeley had observed before that foragers do more dance circuits (faster and longer) when a nectar source is rich. And Lindauer had said before that bees advertising a not so good nest site do “lackluster” dances. So now Seeley wants to prove it quantitatively.
They noticed that scout bees advertise the site they found over a few successive trips to the site and back to the hive, but stop after like an hour. Why? The bees did many more dance circuits for the 40 liter box than the 15 liter box. Also, generally bees were more likely to dance if they found the 40 liter than the 15 liter. The individual data is pretty noisy actually, but overall there’s a preference via more dancing for the better site. The bee that discovers a site is likely to dance, thereby entering it into the debate. Also, they didn’t see any scouts visiting more that one site.
Why do bees stop advertising their site? Is it because they become aware of a better one? Nope, they just stop, usually before they’ve followed dances for any other sites or checked those sites out. Seems like they have some kind of built in process that makes them stop advertising their site after an hour. Probably helps to build the consensus.
Chapter 7: Initiating the move to a new home
Got a Von Humboldt prize to study in Germany. Bees need to be at a flight-ready temperature to fly, otherwise they just fall to the ground. Now they have video equipment that shows how hot bees are. So they see that about a half hour before swarm takes flight for new home, the temp of the swarm starts to rise exponentially. Turns out the signal to do this is piping, which is when a worker presses her thorax against a bee and makes a vibration against it that rises in pitch. Note that piping is only done by scout bees returning from the winning site, this might inform other scouts that their site doesn’t have enough support, so they stop advertising it. Once temp is high, scout bees run over and through bee clusters while running, or “buzz running.”
But how do bees know when to start piping? It’s not when consensus is reached about a site, because they saw one swarm start to fly while still debating two sites, then split itself in two and then circle back. It’s by “quorum sensing” which just means once a minimum number of scout bees are hanging around the winning site (20-30). Consensus would definitely be a better system, but it’s not practical. It’s very hard to poll all the scouts and it’s slow.
Chapter 8: Steering the flying swarm
If only a small number of bees knows where the swarm is going, how do they all fly there together smoothly?
Bees navigate using the angle of the sun, keeping a running tally of how far they’ve traveled (by visual passing of stuff before their eyes) and then use landmarks for the home stretch.
Queen releases 9-ODA from her mandibular glands. Bees keep flying around her if they smell it. Doc painted worker bees with it, and the other bees swarmed normally because they smelled 9-ODA in their presence. When they get there, a few scouts fly ahead into nest site, release a Nasonov pheromone, which guides swarm into the home stretch. But how to the 5% of bees who know where they’re going guide? Through olfactory cues? Streaking along the side of the swarm repeatedly to guide? Or actually computer simulations have shown that if they just fly there, and other bees try to avoid collisions and follow close neighbors, that’s enough to guide the swarm.
Seeley painted shut the Nasonov gland, found that swarms still flew directly to the nest site, so they’re not signaling with scent glands. Except for the home stretch, the swarm took twice as long (20 min) to get into the actual entrance without the scouts signaling for them where the entrance was with pheromones. Using point tracking algorithms and high speed video, they found that streaker bees guide the swarm, fly on top.
Chapter 9: Swarm as cognitive entity
Natural selection has organized bee swarms and primary brains in some similar ways to optimize decision making.
Monkey task: lots of randomly moving dots, but a few are moving together, is it left or right? Starts in middle temporal area (MT) which processes sensory input about movement. Each neuron in MT is tuned for a particular direction of movement. Next lateral interparietal (LIP) area sums noisy input from MT, integrates it into a decision about direction. Importantly, LIP neurons corresponding to different movement directions are mutually inhibitory and this inhibition is proportional to it’s rate of firing (strength of stimulus from MT). So a consensus is reached in the LIP quickly. Once a threshold is reached, this goes to the frontal eye fields and superior colliculus, where motor output is generated, and monkey moves it’s eyes in one direction to indicate which direction the coherent dots are moving in.
He’s thinking of the visual field of a swarm as all the stuff the scouts see. She waggles proportionately to the goodness of the site, so it’s graded just like firing rate in neurons. Each scout is a site-specific sensory unit, because she reports on just one site. When all scouts report the sites they found, this is the swarm’s sensory representation of the surrounding landscape. This is like the MT neurons all firing about what they saw about the direction.
6 important aspects of honeybee decision making:
- Lots of scouts are surveying, so swarm gets a rich representation within just a few hours. Also having so many bees integrating information averages out bee-to-bee variation.
- Lengthy period of information gathering (several hours or days)
- Each scout evaluates site independently after she’s been recruited there. So an evaluation error by one bee isn’t propagated.
- Positive feedback happens with recruited bees recruiting more bees etc. And since recruiting is done based on goodness of the site, the goodness of site signal gets amplified.
- Signals eventually die as scouts stop dancing for a site after a set amount of time. This way, information about inferior sites naturally dies away as they failed to recruit enough bees to keep propagating the dancing for it.
- Speculative: scouts may choose between exploring more to find new sites or exploiting ones they already know about, and this might depend on the number of dances in the swarm. Have they found enough options?
However, they do two things not optimally. Scouts dance for different sites at different times, so it’s hard to compare all options side by side. Also, the individual data is noisy. For each given bee given a certain nest site, they’ll do a different number of circuits in the dance. The swarm integrates over many individuals over a long period of time to reduce the noise from these factors.
For the integration portion (the LIP area in monkey brain) swarms use the number of bees at a potential site. The number of bees at the site represented the total number of dance circuits produced over the last several hours. The inhibition part we see in LIP neurons is just less scouts being available for recruitment to other sites. Also, both primate brains and honeybee swarms are leaky, in that unless new info comes in, the signal declines. This is good for decision making because it increases the amount of time you’re collecting noisy information, so you have a longer period of time over which to integrate. So you can avoid making “fast mistakes.”
Using computer models, Seeley lowered or raised the rate at which dance behavior naturally decays in scouts on each subsequent visit to the swarm (15 fewer circuits per visit). If it decays slower (by 5 per visit) the swarm makes faster, but less accurate decisions.
The final step is turning that integrated info into action. In primate brains and bee swarms, this is done by reaching a threshold, which usually yields a good decision because the strength of the integrated info usually reflects relative level of evidence from each alternative. In bees, this is quorum sensing. Scouts at a site sense a quorum, fly back and start piping. So all non scouts start warming up, all scouts still dancing for other sites stop. Once everyone is warm, the pipers start buzz running, and they take off as one. Quorum size matters. In computer models, turning quorum size down makes the swarm make hasty and inaccurate decisions. Turning it up yields slightly more accurate decisions but much more slowly. This suggests that the quorum size of natural swarms is optimized to balance accuracy and speed.
For the inhibition of non ideal sites, the system needs to have few uncommitted bees, so a vote for one site is a vote away from another. So likely this more comes into play in the latter half of nest-site decision making.
Interestingly, these principles of decision making are quite similar in rock ants, though they evolved it independently.
Chapter 10: Swarm smarts:
What can we learn from honeybee decision making to improve human decision making?
- Members should have shared interests and mutual respect for each other (enough to hear others out on their points of view). But humans don’t’ often share a singularity of purpose like bees, but we can still improve this in human groups. Group leader could remind everyone of their shared stake in the welfare of the group. We can also choose reasonable people who are respectful of others and good at spotting problems etc. Unfortunately, we often can’t choose the members of a group. But we can instate norms of behavior (like encourage all members to speak directly to moderator to keep personal battles more subdued).
- Minimize leaders influence on group decisions. Bee swarms make collective decisions without a leader, but this is unlike most human groups. So perhaps the leader should be as impartial as possible, and just contribute facts like the scope of the problem, resources available, the rules of conduct etc. Leader should be impartial information seeker, more like a moderator, and encourage airing of doubts and disagreements.
- Seek diverse solutions to the problem. Can do this by making sure group is sufficiently large, consists of people with diverse backgrounds and perspectives, foster independent exploratory work by individual members, create a social environment where everyone feels comfortable talking about their solution ideas. You can do this by presenting problems well in advance of the meeting for example, so people can privately explore solutions. Also encourage members who haven’t spoken yet to speak up, this adding a wider range of possible solutions.
- Aggregate the group’s knowledge through debate. Bees don’t just imitate other dancers, they go check out the site themselves. Human groups could do three things: encourage frank debate, encourage good communication, important to critically form your own views about what you have heard.
- Use quorum responses for cohesion, accuracy, and speed. If a human group needs to reach a unanimous decision, you could poll people anonymously several times throughout the debate to see how close you are to a consensus. If individuals set a high threshold, like lets say 80% and decide to switch over if they’re in a 20% or smaller minority, then unanimous consensus can be reached quickly.
Questions:
Why is a bee colony a consistent size (well is it actually?).
The book makes it sound like it is. But why don’t they just make more bees if they have the space and the food? Is there a control on it, like maybe if colony gets too big they can’t communicate effectively?
Is there a direction that evolution takes to integrate information at higher levels of complexity?
One theme of Seeley's book is how similar biological principles seem to apply at very different levels. The reproduction of a hive is very similar to the reproduction of a fruiting tree. The decision making of a hive is very similar to simple information processing schemes used in the brain. Bees in a hive specialize in ways where they can no longer survive alone. There are other principles less discussed in the book, like symbiosis between different types of organisms.
Do you think that these types of principles are predictive at any level, such that we could predict the path of evolution just be assuming increasing levels of complexity? This is a more general idea, not meant to predict the specific fate of some species. For instance, right now, as far as we know, there is no eusocial superorganism composed of "hives" of other superorganisms. Should we expect this in the future of evolution? We've already saw this happen at a few different levels - endosymbiosis (like mitochondria inside the eukaryotes), multicellular organisms (e.g. colonial protists), multicellular organisms with specific tissue types and organ structures, eusocial species of insects, mole-rats, and shrimp.
Do you think that Seeley's experimental methods are justified? Who is this science for, and how do we decide what is ethical in science?
One thing that stood out about Seeley's experiments was how many bees (and sometimes whole hives) he killed to learn something. Sometimes, it feels like an experiment could have been done less invasively, but at a higher cost of money and time. For instance, what if Seeley used something like ladders and laproscopes to explore beehives in the wild. What if the parameters of excellent beehives were determined by observing hives in the wild only, rather than setting up hives where swarms were destined to die?
Should the immediate utility of experiments like these be considered? For instance, would you feel differently if these bees were killed because it was likely to contribute to curing cancer? For a long time in science, the predominant ethical idea was that animals were disposable and for humans to use. Now the predominant ethical idea is that anticipated animal suffering needs to be weighed alongside benefit to humans (but we credit insects with very little capacity to suffer). What could we learn that could shift our ethical framework again in the future? Do you think that scientific ethics will evolve dramatically within the next decade, or do you think that our current framework is stable?
A discussion on consciousness
Do you think that a bee hive could have consciousness?
The superorganism of the hive acts as a single entity in many ways. To us, the most striking aspect was that reproduction takes place at the level of the hive (rather than the level of the individual). Another aspect that was emphasized in the book was how decisions were made by the hive rather than by individuals. Since we've been recently covering various theories of consciousness, we wondered if we could apply those to the types of information processing that hives use.
One aspect that was emphasized in many of the other books we read about consciousness was the need for some integrated information. Christof Koch made the distinction between processes in our brains that are unconscious, which are usually set up in a very feed-forward manner, vs. those that are conscious, which act in a networked way where the nodes modulate each other's responses (has some intrinsic causal structure).
The bee decision making process, at least in the swarming example, felt like a very feed-forward algorithm to me.
- Scout be discovers location, returns to hive and waggle dances with duration and enthusiasm proportional to "goodness" of the site.
- Through random interactions, the scout be catches the attention of other scouts with its dance. Longer dances will tend (on average) to capture the attention of more scouts. Those scouts are recruited to sites.
- Those scouts come back and dance with duration and enthusiasm proportional to "goodness" of the site. At the same time previous dancers stop dancing.
- This same process is being replicated at many sites on the swarm, such that there is a competition between sites. Scouts that are recruited to one site are therefore not recruited to another site. In this way, better sites with longer waggle dance garner more support than other sites over time.
- Eventually there are enough bees at the "good site" (~20-30 bees), that a "quorum" is reached. Counting the 20-30 bees at the prospective site is probably assessed visually, tactilely, or olfactorily. Once the quorum is reached, the scouts return and signal to the swarm that it's time to go. Then they move to the site.
There doesn't seem to be much information integrating here (e.g. there don't seem to be expectations or models that the hive has that is providing feedback to the incoming information of the scout bees). Do you agree or disagree with this assessment? How can we define something like "intrinsic causal power" for a bee hive like this? Can you think of an example of what "intrinsic causal power" within a bee hive would look like?
Say that the superorganism of the bee hive is conscious. Do you think that that predisposes its sub-units to be unconscious? In other words, can a conscious being be composed of smaller conscious beings?
In order for some organism to be conscious, do you think that its component parts must not be conscious? In Koch's book, he conceptualizes the "parts" that contribute to an intrinsic cause-effect structure as logic gates. Often neurons are thought of in the same way. It seems like for a cause-effect structure to be specified, you need the component parts to act in reliable ways. Would this pressure to act reliably for the good of the hive minimize pressure at the level of the individual bee to develop complexity within its own brain? There is some "noise" in neuronal firing. Could a conscious bee, acting in slightly unpredictable ways, just be considered a type of noise? In Koch's IIT theory of consciousness, there is an "exclusion" axiom, which states that the conscious system must be specified over a single set of elements. In that theory, does that axiom prevent one conscious entity being composed of other conscious entities?