[This is one of the finalists in the 2023 book review contest, written by an ACX reader who will remain anonymous until after voting is done. I’ll be posting about one of these a week for several months. When you’ve read them all, I’ll ask you to vote for a favorite, so remember which ones you liked]

Are bees smart?

To answer that question, here’s a crab spider:

Sadly, this is not a review of a book called The Mind of a Crab Spider. But as you crab spider lovers know, crab spiders and bumble bees are natural rivals.

Both bees and crab spiders are well-matched for strength and speed, and in the Rumble with the Bumble , the crab spider doesn’t necessarily win. Bees can often evade the spider, and live to pollinate another day. Lars Chittka, who wrote The Mind of a Bee , and who can safely be blamed for this book review, got thinking. He and his lab decided to build fake robotic crab spiders, and had them really robotically attack bumble bees when they visited flowers.

I remember when I was randomly attacked by robotic crab spiders for a day, and I didn’t enjoy it much. The bees shared my opinion. Not only did the bees have a bad time, their behavioural patterns totally changed. They began to approach the flowers differently. They began inspecting flowers via quick scanning flights before landing on them, and would occasionally reject flowers even if there was no crab spider present. They seemed more nervous.

If you want to see if humans are optimistic or pessimistic, you point at a glass of water that is halfway filled and ask them to describe it. Similarly, you can do the glass half-full versus half-empty test on bees, where you give them an ambiguous stimulus - it might be sucrose, which bees love, or it might be quinine, which they hate - and see if they want it.

If they want it, they’re likely a happy-go-lucky bee with nothing on their mind. If you simulate the bee being attacked by a predator right before this test, they are much less likely to fly to the solution and much more likely to fly into the container labelled ‘Therabee’.

Does that mean bees feel emotions? If they feel emotions, would that mean bees have conscious states? Or are these all just instinctive responses?

Bees exist in that great hinterland of consciousness - the valley where we throw all manner of creatures and living beings whose experiences we remain fundamentally uncertain about. Some readers will likely enter the book believing that bees do not have conscious experiences, and Lars Chittka does a good job disabusing these people of their certainty in this belief, if not the belief altogether.

Almost every chapter discussed in this book in some way reflects on this question of complex cognition versus instinct. There are instructive questions here. What does it mean for a response to be instinctive? Likewise, what sort of response is evidence of complex cognition? Is this a useful distinction? And if you think those questions are polarising, you’re starting to see like a bee. I’ll reflect more on these questions later, but for now, let’s just talk bees.


The ‘waggle dance’ was discovered by Karl von Frisch in the months after the end of World War II. Frisch worked in Nazi Germany during the war. Frisch was fractionally Jewish and his colleagues accused him of “bigoted opposition towards antisemitism” in his writings. He seems to have bowed to the pressure, penning a nasty tract on “racial hygiene” and recommending sterilisation of mentally unwell people. A charitable interpretation of these events would argue that he hoped to protect himself and other Jewish researchers by doing so.

Nonetheless, the Nazis planned to remove him from his university post. But Frisch was saved by Nosema , a single-celled gut parasite, that wiped out several hundred thousand beehives and began causing issues for food security. Martin Bormann, chief of the Nazi party chancellery, decided that Frisch’s dismissal should be postponed to the end of the war, which meant that Frisch stayed on, which meant that Frisch was able to discover the waggle dance. So I suppose we owe a small debt to that nasty parasite, and also to Nosema.

The waggle dance is a line dance performed by honey bees - an individual bee discovers a food source, and boy, is she excited (the vast majority of bees are female). She arrives back in the hive, and begins ferociously waggling while running in a line, then does a semicircular loop and starts the dance again. The angle that the bee runs in relation to the vertical is the angle that the food source is relative to the sun. If a bee runs straight upwards, the food source is in the direction of the sun. The distance the bee runs is proportional to the distance to the food source.

As you may have surmised at some point, the sun moves. The dancers factor this in, and as they spend longer performing the dance, they change the angle of their dance in order to factor in the movement of the sun. And if the sun is behind a cloud? Bees are sensitive to polarised light, and can infer the location of the sun from this light.

An important detail here is that it is generally dark inside hives. Other bees can’t actually see the dance. Instead, bees that dream of foraging put their feelers on the dancer’s abdomen, and hold them there while the dancer shimmies. It’s probably very arousing if you’re a bee. Bees have specific dialects of their waggle dance language, and some bees can learn the dialects of other bee languages if they spend enough time on Duolingo and don’t get killed by the foreign bees.

Chittka tells you all of this, and then tells you that for most bees, the dance is totally pointless.

If you tilt the hive, the bees can adapt and continue to use the sun as a reference. But if you only give bees diffuse light, they lose their reference point. They keep waggling, but the dances become basically randomised, even with just one dancer. Other bees try to follow along, but it’s essentially a waste of time - they’re just getting a distance, with no direction. And yet, if you compare the foraging performance of bees that can’t use the dance with those who can, there is no difference.

This is because the vast majority of bees evolved in tropical Asia, where they lived in large tropical forests, which present many more problems for navigation than the comparatively open European landscapes. Bees that can waggle dance in a tropical forest are about seven times as successful at foraging as those that can’t. If you’re planning on moving your bee colony to Thailand, have them brush up on their cha-cha-cha. But sadly, the hallowed bee dances in old Bavarian halls have basically no function.


Honeycombs house larvae and store food. That’s the reason bees build them.

The honeycombs we’re most familiar with are built specifically by honey bees - there are a lot of bee species (~20,000), and not all of them are social. Of the species that build honeycombs, not all of them build in the same way. For instance, honey bees build hexagonal cells, whereas bumble bees build round cells. But round cells waste more space in the hive, since round things do not tessellate.

You could build square or triangular cells, but apparently larvae have to be raised in cells that aren’t square or triangular. Chittka does not explain why this is the case, but if you watch bees in action, it looks like a hexagon is a trade-off between the shape of a bee - roughly circular - and the waste of space problem above.

Mathematically, honey bees have done a great job. The Honeycomb Conjecture states that a “regular hexagonal grid or honeycomb has the least total perimeter of any subdivision of the plane into regions of equal area”, and it was proven in 1999 by Thomas Hales. This was to the delight of honey bees everywhere, who to celebrate, constructed a small hexagonal sculpture of Thomas Hales with a humongous perimeter.

Honey bees are the only bee species that builds double-sided hexagonal combs. The bottom of each cell has the shape of a pyramid, and the two sides of the comb are connected through the pyramid-shaped base of the cell. Honey bees also build their combs vertically, so that honey doesn’t leak out. But they don’t build them purely vertically - they keep it at an angle so that the honey’s viscosity and adhesion keep it inside the vessel. The cells are tilted slightly downwards.

That’s the structure. But as you read the next part, keep in mind the battle of cognition and instinct, the prize fight of cerebral might; it’s time to put your money where the honey is.

The first row of honeycomb cells is different from the others - it’s a foundation. It may seem like worker honey bees just use their body as a template to create the cell - but worker bees build cells for drone bees, and drone bees are larger than worker bees, by about 30%. There are pillars and cross beams that stabilise the comb. The queen gets a differently shaped cradle structure. Different workers will continue work where other workers have left off, so the bee isn’t just implementing a schematic of a cell and following it to completion. Bees will make alterations to the construction of the cells of other bees. If one bee misplaces wax, other bees will correct it.

François Huber, Marie-Aimée Lullin and François Burnens, working in the late 18th and early 19th centuries, started a malicious and fascinating program of messing with bees’ beeswax to see how they’d respond. Bees usually build downwards, by attaching honeycomb to the ceiling of the hive. If you stop them doing this, like the mean old team of François2 and Marie, they reverse all their motor sequences and build a tower like in Tower Bloxx. If you stop them building up or down, they go from side to side.

If you put glass in the way of the ‘comb - bees hate attaching their ‘comb to glass - bees will rotate their construction. And they seem to notice the glass is there, because they usually rotate their construction prior to reaching the glass wall. If you keep doing it, the bees keep rotating, ducking and diving and moving their hive around you. This changes all the dimensions of the cells, and you get cells that are wider on the outside. How do bees ‘agree’ on their ‘comb dimensions, or the new directions to take? It’s not necessarily clear.

If you keep putting glass in, by the way, the bees eventually just attach to glass. Better glass for the ‘comb than nothing. In winter, bees stop foraging and conserve energy. But in winter, in one of Huber’s glass hives, some comb broke off the ceiling. The bees awoke, fortified the dislodged comb with pillars and crossbeams, and then reinforced the other combs attached to the ceiling, in case they got dislodged. The bees, having experienced a crisis, invested in additional security.

Bees can also do it in space. There were bees on a Challenger mission in 1984 (two years before the tragedy), and zero-g-bees constructed honeycombs with cells of normal dimensions, combining that with other trivial details like learning to fly in space. The difference between space ‘combs and earth ‘combs? The bees got rid of the slight angle downwards - there’s no gravity in space, and thus no need for the angle.

Bees that are raised alone can build honeycombs, but their diameters and cell structures aren’t great. Bees that don’t go through the acadebee are not as good at building as those who do. It is easy to say that bees are just following some program, but it’s very difficult to feel that as you see the way that they build.


This is what a bee’s brain looks like, as stolen liberally from this paper:

Central body (CB) and one of the pair of lobulas (Lo), medullas (Me), antennal lobes (AL), mushroom body calyces (MBC) and mushroom body lobes (MBL).

Even if I could explain all of the intricacies of the bee brain, I don’t really think there’s space to do so here. But there are some parts of the bee brain that I think are more interesting, so I’ll talk about those and pretend like I understand the rest of it.

If you look at the image on the right above, the lobules, medullas and antennal lobes all broadly handle sensory information. Learning takes place in the mushroom body calyces and lobes. Bees utilise what in machine learning is called a ‘fan-out, fan-in’ architecture. This involves spreading out tasks to multiple destinations, where they can be computed in parallel, before sending those tasks to the similar final destinations.

There are hundreds of neuronal links between the antennal lobe and the visual system, and these ‘fan-out’ to ~170,000 Kenyon cells (a type of cell specific to the mushroom body), which then ‘fan-in’ to 400 mushroom body extrinsic neurons, which then connect back to the brain where appropriate behavioural responses are selected.

It is possible to build computerised models of this circuitry, and from these relatively simple circuits produce much more complex learning responses than you might expect.

Incidentally, before a bee leaves the hive to forage, which they do at about 2-3 weeks old, their brains enlarge drastically - their mushroom bodies grow between 15-20%, presumably in order to memorise a load of information about flower locations and the spatial environment. I remember friends at school who displayed a similar ability before they were about to enter an exam hall.


The oscillation of the brain is somewhat understudied. There was a recent ACX book review about brain waves which mentioned that there are basically no books on brain waves. Similarly, when I was doing my masters in neuroscience, brain waves came up a bunch, but mostly in a “and this elicits an alpha wave response which we’ve put into our big list of alpha wave responses and look how pretty our list is” way. I never really felt like I got an intuitive understanding of the significance of an alpha wave, or a theta wave, or a Mexican wave.

One research team studied Drosophila, or flies, when they were asleep, and found that they also have varying brain wave oscillations when they’re sleeping (which makes you wonder if those flies ended up with human-shaped sleep paralysis demons). Broadly, brain waves are linked to conscious experiences. To expand on that, measuring consciousness is hard. Duh. As Anil Seth points out in Being You , one easy bit of consciousness to measure is awareness. We have different conscious experiences when we are under anaesthetic (i.e., none), when we’re asleep, and when we’re awake. These correlate with different patterns of brain waves.

Bees need to sleep. If they don’t get their beauty sleep, their dance moves get worse. If you expose bees to odours while they are in a phase of deep sleep, they begin to consolidate memories from the previous day, and their memories improve. Rats do a similar thing, repeating activation patterns from the previous day in their hippocampus. Does this suggest these creatures are dreaming?

Humans have a pattern of brain activity known as the default mode network, which occurs when our minds wander, or we daydream. Insect brains also exhibit patterns of activity akin to having a default mode network. None of this necessarily means insects ‘think’ or that they have ‘attention’ in the same way that we think and pay attention to things. But it all suggests that they might have similar processes that generate similar, if maybe more rudimentary, forms of those experiences.


One of the most commonly cited impressive bee-behaviours is that they seemingly act in an organised way without any general command structure. But this is an emergent phenomenon. Take ‘fanning’. Bees have to keep the hive ventilated to avoid suffocation. If the hive is poorly ventilated, bees start fanning their wings to increase air circulation. If it is very poorly ventilated, all of the bees do this.

The book is filled with these scenarios, and the best way to read it is to try and solve the problems yourself. Imagine for a second you’re not sitting in your bee-tagging, brain-scanning ivory tower getting readings of how a microscopic part of the mushroom calyx responds to the sound of a frog jumping into a pond.

You’re a scientist in the past, who has to saddle horses and wait for bees to swarm so you can try to follow them, leaping fences and hurtling o’er hill and dale in a wild ride after nature.

When it comes to bee fanning, what’s the solution? How do the bees decide how many of them should be fanning? What do you think François Huber, who belonged to the ‘we do science on horseback’ generation, have guessed? There’s no communication, but as the ventilation gets worse in the hive, more and more bees start fanning their wings. How would you design bees to solve this problem? You don’t want every bee fanning their wings 24/7 or they’re wasting time, but a nice ratio of ‘bees fanning’ to ‘bees not fanning’ that adapts in order to hit your ventilation criteria.

When Huber examined the fanning problem, he came up with an elegant theory. He suggested that bees are differentially sensitive to noxious smells. So as the noxious smells get worse, the sensitivity threshold of more and more bees is reached, and more of them begin fanning until ultimately the entire hive is fanning. Termites likely have a similar set of proclivities, as they will close and open entrances into their mounds in response to humidity changes within.

Note that there are lessons for specialisation and other forms of social organisation here - if you are more sensitive to noxious smells, you spend more time fanning, and you get better at it. This may cement differences in job allocation. Small differences in preference become permanent differences between individuals. This sort of thing probably happens in humans - if you dislike dirty dishes lying around more than your slobby roommate Jeremy, you get better at doing the dishes, and you start to resent Jeremy and leave him passive aggressive post-it notes around the house.

In another example, Karl von Frisch observed that some bees were very picky about the sweetness of foods they would tolerate, and (many years later) Robert Page and others found that differences in sensitivity to sugar are present when bees are hours old, and determine whether they become pollen or nectar foragers.

None of that to my mind requires particularly complex cognition. It’s just a simple instinctive response to a problem that generates a relatively complex set of solutions. Other properties which bees have that we consider complex may also be emergent in a similar way.


No. Sorry. Not really. It’s tricky. Bees seem to have a form of metacognition, where they will leave an experiment if they are unsure about what will happen, as opposed to making a decision. I might try a similar tactic.

You have to be careful when measuring performance in things that can’t communicate. Scientists used to think that bees that were tested on colour discrimination performed to the best of their abilities. Nope. Instead, it turns out bees deploy a speed-accuracy tradeoff, where they will make rapid decisions if there is no downside to doing so.

So for decades, scientists assumed that bees could not differentiate squares and triangles, because in their experimental paradigms where they asked bees to look at squares and triangles, there was no downside. Bees just blitzed the test to get that sweet, sweet sucrose. Once penalties were introduced for mistakes, bees started performing much better and could easily differentiate squares and triangles.

This illuminates one large problem with measuring cognition in vertebrates and invertebrates. If an animal can’t do something in one test, it tells you a lot more about that test than the animals’ generalised skillset. This is really frustrating. Many animals don’t pass the mirror test, where you place something on their head and see if they try to remove it. This is frequently used to see if animals are aware of themselves. Bees don’t generally pass the mirror test. But why would they? The majority of bees have pretty similar faces.

On the other hand, Polistes wasps live in small colonies, and invest heavily into face recognition. This is because to determine their place in the colony hierarchy, Polistes wasps have fights. It’s useful to be able to recognise faces to learn the hierarchy - if I’m Wasp A, and I lose to Wasp B, and Wasp B got pancaked by Wasp C, I probably shouldn’t fight Wasp C.

Bees do have some awareness of their bodies, because before they fly through gaps they often make scanning flights to see if they need to approach the gap diagonally, sideways, or go around.

Here’s a ridiculous diagram of this, from this paper:

Does this count as self-awareness? For me, these tests all seem to fall short of the quiddity of consciousness. But they can’t all be ignored, can they?

Some of you will know William Molyneux’s Problem:

Suppose a man born blind, and now adult, and taught by his touch to distinguish between a cube and a sphere of the same metal, and nighly of the same bigness, so as to tell, when he felt one and the other, which is the cube, which is the sphere. Suppose then the cube and the sphere placed on a table, and the blind man made to see: query, Whether by his sight, before he touched them, he could now distinguish and tell which is the sphere, which the cube? To which the acute and judicious proposer answers: ‘Not. For though he has obtained the experience of how a globe, and how a cube, affects his touch; yet he has not yet attained the experience, that what affects his touch so or so, must affect his sight so or so…‘

It’s a hard question to answer in human subjects, because they can often get the information from elsewhere. It’s hard to restrict humans sufficiently to determine if they’ve actually passed this test. But bumble bees can identify shapes in the dark that they have only seen previously, or vice versa. Bees seem to have little difficulty transferring information between sensory modalities. If bees appear to hold mental representations of objects, does that take them further along the spectrum of consciousness towards higher bee-ings?

In an interview with Tyler Cowen, David Deutsch once discussed the idea of understanding, or of having explanatory power for the actions you take. For Deutsch, this distinguishes us from almost everything else that is alive. Deutsch:

[Animals] have genes which contain knowledge, but it is fixed knowledge, and it is not the kind of knowledge that constitutes understanding. Understanding is always explanatory. You can write a book on canine behavior and look in chapter 37 and it will tell you what a dog will do when such and such happens to it. Sometimes it will say, “Some dogs will do this; some dogs will do that.” There is no such book for humans because chapter 37 will be blank. It’ll say, “Humans are going to do something that neither we nor you can predict.”

It’s unlikely bees could explain their actions, if we could even find a way for them to do so. But is their cognition just a “program enacted by their genes”? Deutsch also mentions squirrels:

You know squirrels bury nuts so they can dig them up later. Well, some people did a very cruel experiment. They put a squirrel, given some nuts or something (I don’t know how they set up the experiment), on a concrete floor. The squirrel did exactly the same behavior with its hind legs with the nuts and put the nuts there and so on. Even though it was having no effect whatsoever. We see the point of scrabbling with your hind legs and then nudging the nuts over there and so on, but it doesn’t. It’s just a program being enacted by its genes.

There’s no link to the lets-mess-with-squirrel-minds study, so it’s difficult to evaluate it. But it may have been the case that they tested the wrong squirrels. Here’s what I mean - Chittka did an experiment with bees where he placed flowers with sucrose under a glass screen. The flowers were connected to a rope. If you pulled the rope, you could get at the sucrose. Bees were unleashed upon the flowers.

Many of the bees did not figure out the sucrose-rope-pulley system. But a handful of bright bees figured out that they could pull the rope. Tool use in bees! But here’s the really sweet part - remember those dumbass bees from before that didn’t figure it out? Some of them watched the brainy bees, and started pulling the rope to get at the sucrose.

There are natural variances in intelligence in bee populations, as in every population. There are also variances in intelligence between colonies. It seems so advantageous to be a fast learner, that Chittka was curious why, evolutionarily, there are still slow learners. The main argument he finds is that bees who learned faster seemed to burn brighter, but were active for less long. This produced a pronounced effect where slower learners actually gathered more resources over their lifespan.

Animals are wily and irritating to test. Simply because one set of squirrels cannot figure out the concrete problem, it doesn’t necessarily mean all squirrels cannot. It may well be the case that all squirrels cannot, but you need to do a bunch of repeated studies, and then grant application boards start asking you why you need another boatload of cash to keep being a dick to squirrels.

This all makes it really difficult to say that animals ‘cannot’ do things, and this is really annoying and I don’t have a good solution other than to say in the short run, let’s keep being dicks to squirrels.


Remember right at the start where I talked about anxious bees? Chittka says that his work suggests bees feel something:

Natural selection might not look kindly upon individuals that do not know fear, mothers who are indifferent to the loss of their offspring, or social animals for whom it does not “feel rewarding” to be in their social setting. In other words, having at least a range of basic emotions might be part of most animals’ “survival tool kit”.

Feelings are not the same as understanding or explanatory power. Emotion doesn’t necessarily equate to consciousness, and Chittka knows this:

There has never been a formal proof of consciousness in any animal, and in this book I have not supplied a formal proof for bees, either. Critical readers might counter that every single psychological phenomenon, every intelligent behavior, described in this book could somehow be replicated by a computing algorithm or a robot, and therefore could in theory be accomplished without any form of conscious awareness. They would be right. You could design a robotic system for planning honeycomb construction, you could build robots that behave as if they experience pain when damaged, and you could of course mimic the counting abilities of bees quite easily in silico. And the list goes on. But, first of all, if you wanted to build an automaton that could do everything I’ve described in this book - the dozens of “innate” as well as learned and innovated behaviors - you would have to equip your robot with a very long list of detailed instructions, and your machine would still be able to cope only with what you have programmed it to cope with. It would be helpless with any novel challenge for which you have not written any code.

If or when we get to forms of AI consciousness and intelligence, and particularly early forms of that consciousness, they might look a little like what we see from the rest of life on earth; complex behaviours that could easily be explained away by some hand-waving, but perhaps shouldn’t be. Perhaps they won’t.

As a question, it needs to be taken seriously, and we should probably be examining a much wider spectrum of animals - unfamiliarity with the creature seems to preclude a lot of science funding. Chittka notes how difficult it is to work on bees, and that his field lacks prestige, and academic recognition, and funding. And that’s bees! People know bees! People love bees!

Imagine being an oarfish specialist. Oarfish could exhibit really unique and weird forms of consciousness that are for some reason easy to study. But we wouldn’t know, because oarfish are a nightmare to find, and we aren’t funding trips to the underwater kingdom of the oarfish, and we aren’t making up a prestigious Rowers Medal for oarfish scientists to win, because no-one really cares about oarfish. I have maybe gone too deep into the oarfish analogy, but I hope you get my point.

One day, people much smarter than me will figure all of this out, but for now, this is where I fly out of the experiment chamber.

Instead, I’ll leave you with one last titbit - bumble bees used to be called humble bees. The linguistic turn wasn’t particularly exciting, but I like the old name. It wasn’t because the bees in days of yore were modest, god-fearing types - it was because they hummed. Cute, right?

If you’ll permit me to block quote one last time, I’ll leave you with Charles Darwin talking about his humble-bees and his hive bees ‘cheating’ while gathering nectar:

One day I saw for the first time several large humble-bees visiting my rows of the tall scarlet Kidney Bean; they were not sucking at the mouth of the flower, but cutting holes through the calyx, and thus extracting the nectar. And here comes the curious point: the very next day after the humble-bees had cut the holes, every single hive bee, without exception, instead of alighting on the left wing-petal, flew straight to the calyx and sucked through the cut hole; and so they continued to do for many following days. Now how did the hive-bees find out that the holes had been made? Instinct seems to be here out of the question, as the Kidney Bean is an exotic.

The holes could scarcely be seen from any point, and not at all from the mouth of the flower, where the hive-bees hitherto had invariably alighted. I doubt whether they were guided by a stronger odour of the nectar escaping through the cut holes; for I have found in the case of the little blue Lobelia, which is a prime favourite of the hive-bee, that cutting off the lower striped petals deceived them; they seem to think the mutilated flowers are withered, and they pass them over unnoticed. Hence I am strongly inclined to believe that the hive-bees saw the humble-bees at work, and well understanding what they were at, rationally took immediate advantage of the shorter path thus made to the nectar.

So, riddle me this. Are bees smart?