Episode 43 of the podcast is now live! Listen over on Podbean, or wherever you get your podcasts. This week, I cover Chapter 9 of Honeybee Democracy by Thomas Seeley. This chapter compares what we have learned about the decision making process of honey bee swarms with what neuroscientists have discovered about the decision making process of monkey brains. Yes, really!
Homestead Updates
Agnes, my red hen who I had previously treated for fluid buildup in her abdomen, passed away recently. The day after my last episode went live, I found her lying on her side in the coop during my morning check. I had felt like her time might be approaching but her recent adventure to a neighbour’s yard had me feeling like the old gal might have longer left than I anticipated. Sadly, that was not to be, and she was so weak that we decided the best thing to do would be to euthanize her at home. My husband performed a cervical dislocation, a humane method of euthanasia, for me as I couldn’t bring myself to do it. I performed a necropsy and determined that her decline was due to heart failure. What was interesting, though, is that heart failure in chickens often presents as an enlarged heart as it must work ever harder to effectively pump blood throughout the body. Agnes, however, had a small, pale, soft/floppy heart. Likely because of the poor circulation this would result in, her crop was extremely full as if things weren’t passing effectively into her digestive system. I found no blockage, though, and there was food and feces in all the appropriate places. My suspicion is that her digestive tract was moving things through extremely slowly due to a lack of adult blood supply. As a result, she was eating a lot more (filling her crop) but still felt hungry, poor thing. It wasn’t my best necropsy in the sense that I didn’t parcel out her reproductive system or lungs but I found the cause of death so it did the trick. RIP, Agnes. You lived your best hen life and I am glad to have known you!
One of my ginger hens was looking a little unhappy last week so I scooped her up for a check. I found a large amount of feces stuck around her vent. Closer inspection showed lice egg build up that the poop was getting stuck to. Her vent looked inflamed and there was dry skin. I treated for lice with DE, gently washed her vent area, removed the eggs, and then applied a simple salve to her vent. She seemed to feel better almost instantly and has been acting normally ever since. I just can’t seem to keep my flock free of these annoying lice!! I have decided to try a product I’ve seen around a few times: First Saturday Lime. Let’s see how that works out!
Speaking of the chickens, I witnessed a showdown between the flock and a cat! None of them seemed to know what to make of the other. Neither the cat or the chickens were aggressive, they were just staring at each other looking bemused. The rooster was alert but not going for the cat. When the cat saw me, she ran away. Beautiful thing that is clearly someone’s well loved pet. Clearly, pet cats are not sure what to make of giant birds like chickens!
Spring is definitely here now and we have had some glorious weather (into the 80sF!!) so I have been making the most of it. I’ve been working on clearing up the garden beds, I repaired the side/backyard path from my pipe excavation, I’ve started attacking the weeds that took over my side bed, which has been ignored for a year, and various flowers are slowly coming into bloom.
I have been spending time just wandering the property, finding quiet places to sit, and just watching nature. Due to this, I found a groundhog and their burrow at the back, by the swamp! I really love groundhogs. They’re so cute and much bigger than I think they will be! I hope this one sticks around and raises a family back there. It’s a good spot away from any neighbourhood dogs.
It’s also officially Spring for us because our osprey pair have returned! This pair has been nesting on top of a billboard on a major read near us. My husband and I have loved watching them nest and raise young each year up there, and so were crushed and furious when we noticed that the company that owns the billboard ripped the nest down and put spikes up to prevent further nesting. Apparently, it’s legal to remove the nest when the birds aren’t using it (and osprey fly to warmer climates for winter) and so they seized the opportunity. I was gratified to learn that I’m not the only one who is upset by this; many people on my birding groups complained to the company, and it was even covered by a few local news channels! I am pleased to report that, not only have the osprey returned, but they are building a new nest using the spikes as support. Ha! Take that, heartless advertising company!!
Hive Updates
I did an inspection on Tuesday, April 6th. Weather was warm (65F+) and sunny with a light breeze.
This colony is bursting with brood! So lovely to see all those capped cells. A new generation has already emerged based on the nurse bees I saw, and there are lots of eggs.
Only 5-6 capped cells of drone brood so I appear to be safe from potential swarming for now. I’ll keep a close eye on the number of drone brood moving forward, as well as how much space they have.
Previously, I had given them a deep box filled with honey frames (from my deceased colonies) to make sure they would have enough food for our cooler nights and any potential cold snaps. Well, they don’t appear to have used much of it! There is so much honey throughout this colony.
I decided to remove the hive wrap and the moisture box but left the feeder on for now. I’ll likely remove it at the next inspection as I’d like them to eat through that honey (I can’t extract it for our consumption due to the treatments I used in the Fall).
Next inspection will involve a mite check, I think.
*
Chapter 9: Swarm as Cognitive Entity
I’m a systems neurobiologist who studies how the three pounds of goo we call a human brain makes decisions.
-William Newsome, 2008
Quick note: I struggled a bit with this chapter. First, I clearly will never be a neurobiologist because all the information on neurons was almost painfully boring for me, which I’m not too proud to admit! Second, I found this chapter to have especially dry language so I sometimes struggled to condense things down into an accessible format. All this is to say that any errors are entirely my own, and there might be a few things that are lost in translation. I highly recommend reading this chapter yourself to fully grasp the material covered. I’d like to think this episode did a decent job of explaining things, though!
So far, we have looked at how honeybee swarms decide where to set up their new home. Each mechanism of this house-hunting process has been identified and discussed in the previous chapters, and now Seeley invites us to step back from the detailed analysis and instead consider what has been learned about the general features of a swarm as a decision-making system.
In doing this, we can see comparisons of the mechanisms used by swarms to those of primate brains. Yes, really! Both seemingly disparate systems have been shaped by natural selection to efficiently collect information and act on it in a manner that benefits the whole.
Just as individual bees within a swarm have limited information so do the neurons within the primate brain. In other words, “the decision-making process is broadly diffused among an ensemble of relatively simple information-processing units.” Pg.199.
Conceptual Framework for Decision Making
Neuroscientists study monkeys, using them as a kind of human surrogate, to dig deeper into the mystery of the human brain.
The example of note for this chapter involves investigating a monkey brain while the monkey receives visual stimuli and then needs to make a decision. A monkey is placed before a screen with dots moving around on it. Some of the dots move right and some are moving to the left. The monkey has been trained to focus its eyes in the direction that has the most dots moving toward it. While the monkey is gazing at this screen, its eye movement is being tracked while the scientists also record the neural activity. This has allowed neuroscientists to identify the neural processes involved in this particular decision-making task (whether the monkey looks left or right).
When a decision is to be made, the starting point within the monkey’s brain is the middle temporal (MT) area of the brain, which processes sensory information about the motions witnessed.
Each of these MT neurons have a receptive field that corresponds to a specific portion of the monkey’s visual feed, and each is sensitive to a particular direction of movement. So an MT neuron fires when it detects movement in one direction but would remain passive (would not fire) if motion occurred in the opposite direction.
Working together, MT neurons provide information on the strength of rightward and leftward motion throughout the full visual field.
Next in the process, the Lateral Intraparietal (LIP) area of the brain comes into play: these neurons receive input from the MT neurons and are organized into direction-specific integrators. What this means is that, if a monkey is shown a display with rightward moving dots, the LIP neurons that act as rightward motion integrators will fire, and their firing rate will increase in relation to the stimulus strength (more rightward moving dots will cause more rightward LIP neurons to fire).
Crucially, the various integrators corresponding to different motion directions are mutually inhibitory. This means that stimulus strength is key. Even if the firing rates of LIP neurons associated with rightward and leftward motion increase at the same rate, only those with the strongest stimulus will continue to increase their firing rate, while the weaker will decrease. This assists in the monkey discriminating between simultaneous right and left movement of visual objects.
The activity of these integrators eventually reaches a threshold and that results in a decision being made. So a monkey exposed to a mix of dots moving in multiple directions can choose the direction in which most of the dots move, and then choose to focus its eyes in that direction.
You might already be seeing the similarities in this process and that of the decision-making of a swarm but let’s look at that more closely. . . .
The Sensory Transformation in a Swarm
We know that scout bees fly for several kilometers in all directions to look for prospective nest sites and that, should a desirable nest cavity be found, she will return to the swarm cluster and report her finding through her waggle dance. The strength (number of dance circuits) of her dance indicates the quality of the found site with high dance circuits indicating high quality.
We know that each scout acts as a site-specific sensory unit, reporting on just one nest site at a time, much like the MT neurons that report on just a section of the visual field.
The display of bee dances can be thought of as the swarm’s sensory representation of the landscape of potential nest sites, just as the MT neuron firings form a sensory representation of visual stimuli in the monkey’s brain.
Seeley identifies several ways in which scout bees build their swarm’s sensory information:
Several hundred scout bees make up the sensory apparatus of a swarm within a few hours. These intrepid explorers can gather a wealth of information on potential nest sites. Scouts can discover, inspect, and report on almost a dozen sites in just one afternoon.
Scouts collect information over several hours and even several days. A long period of discovery and information acquisition means that a swarm collects a larger, and therefore more reliable, amount of information.
Each scout makes an independent evaluation of a site. We know that the majority of scouts that report on a site were recruited to it. A recruited scout will first inspect the site and she will then decide how strongly she will dance for it depending on how high quality she has found it to be. This individual inspection means that a reporting error can be corrected. A scout who dances strongly for a poor site will be quickly ‘overruled’ by recruited scouts who find her location poor and then do not dance for it (or dance more weakly).
Scout reporting leads to recruitment of additional scouts. A strong reporting of a site leads to recruitment of more scouts, who in turn recruit others, and so on. This is called a ‘positive feedback loop’, and results in the better site monopolizing the dances, which leads to the swarm’s attention (or sensory input) focusing on the higher quality locations.
Scouts reduce their dance response over time. Although a good nest site does not decline in quality, scout bees report on it less as time passes. This means that inferior sites will be forgotten over time as ‘older’ scouts lose their drive to dance and no recruits return from the weaker sites to report on it. Instead, the strong dances continue and gain additional support. This ‘decay’ in dance response contributes to the swarm focusing on the higher quality sites.
Scouts may choose between exploring and exploiting. It’s not known at this time for sure but it’s entirely possible that scouts choose between exploring unknown sites versus exploiting known sites. They could do so by sensing the abundance of dances on the swarm. This would function as a kind of regulation of sensory input; increasing exploration when the swarm has little information (few dances for sites), and decreasing when an abundance of information has been gleaned (many dances for sites).
These 6 features foster successful swarm decision making but there are two features that actually hamper successful decision making.
First, scouts make reports individually at varied times. For instance, depending on the time of discovery, most dances at the swarm could be for a poor site because a better location has yet to be found and recruited for. In this instance, just because many scouts are dancing does not mean the site found is of high quality.
Secondly, individual scouts produce a wide variation in the number of dance circuits per dance. Thankfully, to offset this ‘noisy’ reporting, a swarm integrates its sensory information over many hours and across hundreds of bees.
The Decision Transformation in a Swarm
After the first stage of decision making through sensory input has occurred, the monkey brain and the honey bee swarm moves on to stage two: decision transformation.
The primary function of this stage is to integrate all of the ‘noisy’ information, allowing the brain or swarm to know how much evidence overall has been collected for each potential outcome. This allows a decision to be made.
We learned previously how, in a monkey brain, the LIP neurons integrate information received by the MT neurons. In response, the LIP neurons compile the level of input (stimulation) and adjust its output (firing rate) accordingly.
This decision transformation process is similar in a honeybee swarm. For the swarm, the integrator for each potential nest site is the number of bees that visit it.
We have learned how a scout’s dance strength recruits more scouts to the advertised site, and how these new scouts will then dance for a site found acceptable with the strength of their dance indicating the site quality. This means that, over time, many different dances will occur on the swarm and at varying strengths.
Since the highest quality site will continue to attract more scouts then the best site will accumulate the highest number of bees that visit it.
Previously, it was mentioned how the integrators in monkey brains are mutually inhibitory; as evidence builds in one integrator it inhibits the accumulation of evidence in all other integrators.
Similarly, with honeybee swarms, the fast rise of bees visiting the higher quality site is accompanied by a decline of bees at all other sites. In this case, inhibition is due to the finite pool of uncommitted scout bees that can be recruited to support a found site. If the recruits were originally visiting a poor site, when they retire from visiting and dancing (as we have learned all scouts do), and re-enter the neutral recruit pool, they are more likely to be recruited to a better site whose dancers have continued to grow over time.
An additional shared design of integrators in monkey brains and honeybee swarms is that collection of evidence in any integrator will decline over time if no new evidence is collected. Think of the integrators as ‘leaky’; without a continuous influx of information, all the previously collected input will ‘leak’ out.
This same mechanism is seen in the way a scout bee’s drive to visit and dance for sites declines and then ceases over time.
Why might this be the case?
Several models developed by mathematical psychologists have found that this mechanism enables a decision making system to update itself should the situation change and a new alternative be discovered.
This mechanism also results in lengthening the time period in which evidence is collected, preventing ‘fast mistakes’.
Seeley sees the same function in his honeybee swarms.
Working with Kevin Passino (see previous chapter) once more, the two men designed a mathematical model of the nest site selection process of a swarm. This model simulated the activity of 100 scout bees presented with 6 nest sites that varied in quality. Each simulated scout was equipped with all the known behaviour of scout bees.
They first tested the model and compared the results to those seen when observing natural swarms. It worked beautifully, replicating real world examples.
Then they modified the model to include scout bees who behaved slightly differently from their real world counterparts. For instance, they modified the dance decay rate of the scout bees. Real scouts reduce their dance strength by 15 dance circuits per trip, on average. They tweaked the system to see what would happen if the decay rate was raised to 35 circuits per trip, and then lowered to 5 circuits per trip.
At the lowered decay rate, which caused the scouts to dance for longer, the model swarms made faster but less accurate decisions. Their decision-making was poor because the information received was lingering for longer; meaning that a poor site discovered early would have more support than a good site discovered later in the process.
At the increased decay rate, the swarms decision making process was much slower but had a high degree of accuracy. It took the swarm a long time to decide because even the best sites had scouts that would stop visiting more rapidly, causing a longer time to reach the information threshold needed to make a decision. The decision was more accurate because the threshold would eventually be reached as normal, just at a much slower rate.
Looking at these results, we can see that the dance decay rate witnessed in natural swarms is a good balance of speed and accuracy.
The Action Transformation in a Swarm
The final stage of any decision making process is the act of making the decision itself; selecting a single response to the options at hand.
In monkey brains when making eye-movement decisions, and in a honeybee swarm choosing its new nest site, a response occurs when one of the integrators reaches a threshold level.
This usually results in a good decision because “the relative level of evidence in each alternative’s integrator normally reflects the relative strength or quality of each alternative.” Pg.212
In Chapter 7, we learned how a honeybee swarm senses when a threshold level has been reached through quorum sensing. Once a quorum has been sensed, the scout bees begin to stimulate the swarm to prepare for flight via worker piping.
The beauty of this method is that, even if some scouts are still reporting for losing sites, the activity of the bees preparing for flight will drown out these potentially confusing messages, ensuring consensus for the chosen site to be reached.
Quorum size is a key component in the accuracy of this decision making system as demonstrated when Seeley modified his mathematical swarm simulation. If he adjusted the number lower than the natural model of about 15 bees at a nest simultaneously, his model swarms made fast, poor decisions. Adjusting it upward resulted in slower and only slightly more accurate decisions. Once again, the natural system seems to have found a good balance between speed and accuracy when making a decision.
This makes sense if we consider how natural selection has honed this process to one that offers the greatest chance of success, as the swarm has just this one chance to succeed.
Seeley notes, however, that it is possible that the quorum number could be lowered in an emergency, such as extremely bad weather or the swarm being on the verge of starvation. This would allow a swarm to seek a new nest site much faster. Whether this is in fact the case awaits further study.
Convergence on Optimal Design?
Thirty years ago, computer scientist Douglas Hofstadter suggested that “ant colonies are no different from brains in many respects”. In both systems, groups of ants and groups of neurons are themselves not individually intelligent but function together as a higher-level intelligence.
Thirty years on, much more is known about the decision making systems of primate brains and insect societies, and all this new knowledge seems to support Hofstadter’s statement. Namely, “that evolution has built intellectual strength in ant (and bee) colonies and in primate brains using fundamentally similar schemes of information processing”. Pg.213
Looking at the decision making systems of primate brains and honeybee swarms, we can identify 5 critical elements:
Sensory units that provide evidence, with each reporting on just one option, and with its strength connected to the quality of said option.
Integrator units that collect evidence provided by the sensory units, with each integrator collecting for just one option/possibility (such as nest site quality)
Mutual inhibition of the integrators; the growth of evidence in one suppresses the growth in all other integrators.
‘Leaky integrators’; growth of evidence requires continuing input.
Threshold sensing; the first integrator to reach the threshold ‘wins’, and a decision is made based on this integrators evidence.
Why might this kind of convergence have occurred? “A strong possibility is that this striking similarity exists because this design is a means of implementing robust, efficient, and possibly even optimal decision making.” Pg.215.
Now, let’s look at something called the sequential probability ratio test, or SPRT. This mathematical system implements the statistically optimal strategy for choosing between two alternatives by specifying when to stop integrating further evidence in order to achieve a given error rate. This minimizes the time needed to make a decision for any desired level of accuracy. Simply put, this test achieves the optimal balance between accuracy and speed when making a decision.
A computer scientist at the University of Bristol in England (my alma mater!) named James Marshall, worked with a team to examine theoretically how honeybee swarms might implement optimal decision making when faced with the choice of just two nest sites.
Working on the premise that evidence for one alternative is also evidence against the other alternative, they posit that all evidence can be accumulated as a single total.
This ‘random walk model’ posits that evidence accumulation can be viewed as a random walk along a timeline where evidence for one alternative increases the total while evidence for the other alternative decreases it.
This random walk model of decision making implements the statistically optimal SPRT.
In the case of a swarm deciding between two possible nest sites, we can see how evidence for a site is also evidence against the other.
Seeley points out that further study is needed when examining this binary choice scenario in honeybee swarms.
Of course, usually swarms are faced with deciding between several sites, not just two, and even near the end of the decision making process, the race to reach the threshold limit can occur between more than two sites. But since the SPRT is effective when several alternatives are available, it is possible that primate brains and honeybee swarms independently evolved the same basic decision making mechanism because it results in, or is close to, optimal decision making.
“If this hunch proves correct, then we are looking at an astonishing convergence in the adaptive design of two physically distinct forms of “thinking machine” -a brain built of neurons and a swarm built of bees.” Pg.217
*
And that’s it for this episode! Next, I will be finishing the book with Chapter 10 and the Epilogue. Thanks so much for sticking with me! Once we have covered this book, I’ll be moving on to other topics such as top bar beekeeping methods, and a list of some homesteading/farming memoirs that I have found both helpful and a pleasure to read.
I hope you’re all doing well and staying safe in this mad world of ours. And I hope you’ll join me again in 2 weeks! In the meantime, take some time to relax outside and enjoy the lovely weather!
