Time is Honey

0:28Wait, you're listening. guy named Sunil Nakrani. Yeah, hi, I'm Sunil Nakrani. Where did you grow up? And were you just a computer kid? Like you just love computers? Or how did that, how did this all start? So you want to start from there? Yeah, I mean, a little bit. Yeah, so I was actually born in Kenya. But he grew up between India and the UK. Right. He studies hard. Bachelor's and a master's degree in electrical engineering. And in 1989, he lands a job at IBM. just as the world is encountering this new thing called the internet. Welcome. And what about this internet thing? Do you know anything about that?

1:02There's loads of useful information in here. You can get news, recipes. So, Daniil is like, Okay, why not go and study communication engineering? He goes to Oxford to get his PhD, and one day, near the beginning of the semester, while he is on one of the desktop computers in the computer lab, his whole department, including him, gets an email from one of his professors, who's an American guy. And the email just says, Hey guys, America is under attack. Come down to the common room and then watch, right? It came out of the clear blue sky on a mild fall morning in Manhattan. Oh, it's 2001. Yes, 2001.

1:38Sunil's standing there in horror.

1:43And I mean, he has a million questions. Like, who did this? Why? Is America at war? So he goes online. to get any sort of new information. But he just can't. A lot of the websites were just not responding. Some websites had just crashed. Others would, like, just keep loading, but then never fully load. There's one point there, you know, there was such overwhelming demand for, you know, news that they resorted to serving only text. Plain text. Because there were so many people trying to get to those websites. Because there was just so many people trying to get access. So much hunger of people trying to figure out, like, what the hell is going on here?

2:21Correct. And he's like, why is it that at the very moment when I in the world want to access something the most, that's when I can't access it? And this used to happen all the time. As millions of people flooded the system. Last week, a picture of a dress was posted on Tumblr. Some website. Healthcare.gov. Or picture. That broke the internet. Or video. Would suddenly become popular. In just 15 minutes, a Pokemon Go launch. And you get this crash of people. The traffic had already passed a little prediction. This flash flood. And it breaks the internet. And this situation. A demand that is beyond what they planned for.

3:00Sunil became kind of obsessed with it. And so I ended up looking at websites, how they architect some of these infrastructures. And he's like, there's got to be a way to fix this. Yeah. Like, what can I apply to solve that problem? Meanwhile, Sunil's wife is working in Atlanta. We were doing back and forth between Oxford and Atlanta. And at a certain point, he's having kind of a hard time with this internet problem. Because how do you design a system for the future when you don't know what the future is? And one day he just thinks to himself, Georgia Tech is down the street. Maybe someone there could help me.

3:31Point me to some direction that I could take. So he just emailed some people in the engineering school. I said, oh, I'm a PhD student looking to, you know, discuss some ideas. Can I come and see you? And that was basically it. I didn't describe the problem in my email. I didn't really expect anything substantial. But within 30 minutes, he gets a response from a guy named Craig Tovey. Saying, come by my office. And then a very tall guy knocks on my door and says, oh, I'm looking for Craig. This, of course, is Craig Tovey. Hi. Good to meet you. Craig's a systems engineer. Operations research. His job is to make huge operations run smoothly.

4:19Factories, shipping routes, that kind of thing. So anyway, I walked into his office. We sat down. We started talking. I started describing the problem. So Sunil is like, look, I'm trying to fix the internet. I'm trying to stop it from breaking every time one of these internet flash floods happen. He didn't say a lot, actually, at the beginning. He just kept listening. And then 25, 30 minutes later, you know, suddenly Craig Tovey said, Oh, oh, oh, wait. Craig stands up. And then went back to his desk and pulled out a paper. And he plops it down in front of Sunil. It is called The Pattern and Effectiveness of Forager Allocation Among Flower Patches by Honeybee Colonies.

4:59So at the time, I thought, why are we talking about honeybees? Yeah, why are we talking about honeybees? Well, Craig had this hunch that bees had something to teach Sunil, and all of us, really. because it turns out bees are sort of a model of how to thrive in an uncertain world. Hmm. Okay. So that's the story I want to talk about today. The story of how a multi-billion dollar tech industry used a trick they learned for millions of years of honeybee evolution to build the internet as we know it. Okay. Giddy up. Okay, so Craig and his bee study, it really began with his collaborator. I'd like you to refer to me as Tom.

5:47Dr. Tom Seeley. You could introduce me as doctor or professor, but let's quickly switch to Tom. He's a retired professor of neurobiology at Cornell and one of the world's top experts on honeybees. I did pretty much pioneer the study of how honeybees live in the wild on their own. And he started doing that work more than 60 years ago, when he was a kid. Yep. A swarm of bees moved into a large black walnut not far from my parents' house. And Tom says he would watch those bees flying in and out of this knot hole in the tree. Like shooting stars. Zooming off in all directions. And he'd see them and he'd be like, where are they?

6:26Where are they going? Like he kind of knew where they were going. There were fields of a dairy farm up on the hillside. With lots of flowers full of nectar and pollen for them to eat. They're probably going up to that. Like, I know where they're going, but how do they know where they're going? Like, what is going on here? Um, are they, they're not, aren't they just, like, flying? And if they see some pollen or nectar, they bring it home? No, no, it's way, way harder and more complicated than that. Why? How so? There's many different reasons, and this is all stuff Tom would eventually learn in college and grad school.

7:05Okay, so like first, flower patches are not evenly or clearly distributed. They're not just everywhere. Okay, fair. They're also not all blooming at the same time. Okay. So there might be flowers that are blooming at certain times of the year or even certain times of the day. Also, you need to find the flowers in bloom that still have nectar because you're in competition with all these other pollinators. Right. And you have to do all your food gathering before winter comes because, you know, then all the flowers go away. And you have to get something like for guess how many flowers if you're a bee, guess how many flowers you have to hit to get like a little bottle of honey that you would find in a grocery store.

7:49Oh, oh, OK. Like one of those bare little squeezy. Yeah, one of those little bare squeezy bottles. How many flowers went into that? That's such a nice question. I don't know, like 10 ,000? Two million. Two million? Two million. And the hive needs like the equivalent of 200 squeezy bottles of honey to survive through the winter. Wow. All of this is an efficiency game. Like you only have so much time when the sun is shining, when these flowers are blooming, that you can hit them. Okay. Okay, so they have to do all of this. It's really hard. And they have, like, nobody, they have no boss. Nobody is in charge.

8:36Queen Bee isn't in charge? So that's the thing. So back in Aristotle's time, that was the idea. For many, you know, centuries, people thought the queen was a ruler. But no, that's not correct. The queen's whole job is to make babies. Yeah. She's not telling anybody else what to do. Nobody is. The colony is intelligent in some way that the individual bees are not. And when Craig found out about this. It started with NPR. From a radio story. Tom Seeley talking about honeybees. He was just like, what? I was just in awe. Because at the time, Craig was working on robots. He was trying to get a group of them to build a car.

9:16It was an unsolved problem. And he was kind of stumped. I have no idea how to get these robots to work together to be, as a group, like more intelligent than the individual stupid robots. But here were these honeybees. Doing exactly that. And I'm thinking, wow, let's understand how the bees are doing that. Now we can copy that and apply it to robots. And Craig thought this would be pretty straightforward. I thought that biologists had figured out everything. But turns out they hadn't. That's right. If somebody asked me how much we know about how a honeybee colony works, I'd say 50%. And, you know, 50 % is not nothing.

9:58They knew, for example, about the waggle dance, which you might have heard of. Yes, I've heard that bees will waggle. And what is the waggle, though? I mean, it's quite sophisticated. It's a dance that bees do to sort of show other bees where they just came from. Oh, so it's like a choreographed map. It's like a map dance. It's like a map dance. And this is like a piece of the puzzle. Oh, yeah. Von Frisch won the Nobel Prize. But what the biologists had not figured out... Is the bigger picture. What Tom would call... The wisdom of the hive. And that is what Tom was working on when Craig called it.

10:37And Tom said, well, you know, come help me run these experiments. And the two of them would end up doing an experiment together that would give us a little peek into that wisdom of the hive and would eventually become the paper that many years later Craig would slap down in front of Sunil in his office. It's still vivid in my memory. In fact, this is one of the best weeks of my life. It's July 1991. OK, now this is in the Adirondacks, upstate New York. Like way upstate. Close to the Canadian border. Craig had to drive on these dark back roads. For an hour and a half, two hours. Looking for a little sign for this research station.

11:18Cranberry Biological Station. Cranberry Biological Station. Cranberry Lake. I'm sorry. Cranberry Lake. Cranberry Lake. Yeah. I'm always amazed that he pulled it off, that he found this biological station up in the middle of the woods, deep at night. And the reason Tom studies bees at this place, way out in the middle of nowhere, is because... there are no naturally occurring honeybee colonies. What? Wait, why would you study bees in a place where there are no bees? Right, because the absence of bees lets you set up kind of controlled experiments without any other bees messing it up. That's right.

11:56It's a BYOB, bring your own bees, set up. So Tom brought an experiment, you know, a little colony of about 4 ,000 bees. Okay. And, you know, after breakfast, we go out to where the hive is that Tom has set up. A wooden box. Two feet high. Kind of like a beekeeper would have, but with one big difference. The front is all glass. Basically a transparent hive. Yeah. So you can watch the bees inside. And each bee had to be individually recognizable. How do you do that? So you put a little two-digit number on their thorax and a little daub of paint on their abdomen. Until they're painted, they're anonymous members of a colony.

12:39But once they're painted, you get to know them. Some are nervous Nellies. Others come running. Yeah. They have little personalities? Oh, definitely. Definitely. Some get up in the morning and get things going. Others hang out till 11 o 'clock. Anyway, when the sun comes out on that first morning, a big wave of painted bees whooshes out. Except for those slackers who are sleeping in. Fine. Except for the lazy ones who are sleeping in. But pretty quickly, the ones who have gone to work, they start flying out to these feeders, these fake flower patches that Tom has put out for them. Imagine a glass Petri dish where we have high sugar content water.

13:18It's a little buffet for them. Yeah, that's right. And so the way the experiment works is that there are two fake flower patches, but they're not equal. One is better than the other. And what they want to see is like, how do the bees suss that out? and how do they make it kind of a collective decision to send more bees to one rather than the other? Whoa. Okay. Oh, so this is it. This is where you see maybe it's not just random, like they might get to glimpse hive intelligence. Yeah, because remember, this is the whole thing, right? Like winter's coming, they're on a clock, and let's see how they prioritize how they make that decision.

14:02Huh. Wait, and then how do they make one fake flower patch better than the other? Well, there's lots of ways. You know, we can put 1.5 molar sugar. One feeder might have sweeter sugar water than the other. We can change it from one and a half to two and see how the behavior of the bees changes. Or one might be bigger and one smaller. If you had a lot of bees coming to the same feeder, some of them would have to wait. But Craig says, for the purposes of explaining this, just imagine this situation. One feeder, which is close. About five minutes from the hive and another one that's... Ten minutes to fly out to the flower patch, fill up her stomach with nectar and fly back.

14:40A little further away. Okay. So... Tom is at the hive recording each bee as it leaves the hive and each bee as it comes back. Also, are you afraid? Have you ever... I mean, probably not. But like, have you ever been stung by a bee? Are you afraid of bees at all? I am somewhat allergic to bees. Oh, I did not know that. So I am scared of them. But ordinarily, honeybees are not all that aggressive. Anyway, right away, the guys noticed. There are a lot of bees going to this five-minute feeder. The feeder close to the hive is blowing up. And when they follow those bees back to the hive... To my eyes, it just looks like chaos.

15:21But if you're Tom Sealy, you'll spot... The waggle dance. A waggle dance. The bee effectively saying... Hey, go that way. And pretty soon they see another bee from that close by feeder come in and dance. Hey, go that way. And another. Hey, go that way. And another. Hey, go that way. And each time those bees dance, they're bringing more bees back to the five-minute patch. Hey, go that way. Hey, go that way. Hey, go that way. Go that way. And as all that's happening, every now and then. Hey, go this way. A bee comes in and dances for the 10-minute patch. Hey, go this way. But. Go that way. Go that way.

16:01There will be more bees going to that five-minute patch. And there will be fewer bees going to the 10-minute patch. Just because the bee's coming back twice as often.

16:17Oh,

16:21okay. So the closer flower patch gets more bees saying, go that way, that way, that way. and so more bees go that way instead of this way. Right. So now here comes one of the beautiful parts of it. And this is where it gets really interesting. If there are a lot of bees going to this five-minute patch, eventually, Craig says, there will be more and more depleted flowers. This patch, it starts to run out of nectar. Hey, go that way. Go that way. Go that way. That means that a honeybee, she's going to take longer to fill her stomach. Hey, go that way. Right. So a five-minute patch, if it's crowded, is no longer a five-minute patch.

17:00Go that way. Right. It might become a seven-minute patch. And as the patch gets more and more picked over... Now it's an eight-minute patch. Hey, go that way. And then... A ten-minute patch. And... Hey! At that point... Go this way. It's taking the bees the same amount of time... Go that way. ...to go to the close-by patch as it is to go to the one that's further away. Go this way. And because of the dancing, the hive is sort of evening out the number of bees it's sending to each patch. Go that way. Go this way. Go that way. Go this way. Then you're in equilibrium.

17:35Even though the bees don't have stopwatches, they equalize the round trip time. The hive has taken into account distance and crowdedness and figured out the way to get the most nectar in the least amount of time. And in the real world, the bees may very well be going to five different patches. Or a dozen. Not just two. And dealing with actual nature, which means taking into account different types of flowers and weather and predators. But no matter how many other variables the bees have to deal with, the allocation of bees amongst the flower patches is astonishingly efficient. It looks like there's an air traffic controller or like the hive is thinking.

18:28Yeah, that's right. I think, though, when we use the words like thinking, we're thinking in human terms. But if you say thinking, it's just a matter of taking in information, processing it to make decisions. And I think that definition of thinking applies. And the way that they were processing this incredibly complex set of information was by following one simple rule. If one flower patch has a smaller round-trip time than the others... Send more bees there. Like, whichever patch bees are coming back most quickly from... Go that way. Go that way. That's where you send more bees. And that's it.

19:11It's gorgeous, isn't it? A friend of mine once said it's very zen.

19:20But it's also weirdly, it's a very bottom line. Like, it's like, I don't care where you're going. I don't care what you're, like, it's like what really matters is when you show up with the goods. Like, show up with the goods and then we'll talk. And then we'll negotiate. So this system Craig and Tom observed at Cranberry Lake, Craig wrote it up as math. He called it... The honeybee algorithm. What does it look like on paper? Equations I wouldn't want to show a fourth grader. F sub n of X sub n, F sub n of X sub n divided by X sub n is equal to F sub m. And when Craig tried to use this algorithm on his car building robot problem.

20:07It didn't apply. It was completely unhelpful. It was such a different problem. Which is why it was so exciting to him when Sunil walked into his office more than a decade later with this other problem. The internet problem, which to Craig anyway, looked basically the same as the one the bees were facing. And so after 15 minutes, I said to Sunil, let's imitate the bees.

20:36How could imitating the bees help Sunil stop the internet from breaking? I am desperate to finally understand this, but first we need to take a quick break. All right, we'll be right back.

20:54boodoo doo doo doo doo doo ha lulu latif radio lab back with bees yeah um and you were going to tell us how they're like in our internet or something yeah basically pretty much um but to do that i needed an example so i found something in my own life which I apologize in advance for how annoying this is going to be. Okay. Do you remember the hamster dance? No, I don't think I do. Is this like a website? Yeah, yeah, yeah, yeah, yeah. Okay. I'm kind of surprised you don't remember it. It was a very early web kind of thing. But all it was, it was just a song with a web page and the web page was just a bunch of GIFs of cartoon hamsters dancing.

21:39It's the most annoying song that you will ever. Can you still sing it? I can. Would you like to hear it? Yes, please. It was, um, anyway, something like that.

21:55It's very, very annoying. Okay. And I remember when I was like in high school, I thought it was so funny. And one day I tried to show it to my cousin because I was like, oh, this is so funny. You have to see this. And I remember it like taking a really long time to load. And I was like, oh, because there's so many people who are. Who need to get their eyes on this. Right. And so what was happening behind the scenes was, well, OK, so like basic Internet 101. Yeah. Every website on the Internet, like, for example, what was at the time, hamsterdance.com, exists geographically somewhere. Like in the cloud or?

22:33No, no. Like a single earthbound computer, the owner of the website paid for it to live there. And that computer, which is called a server, is not in a house or an office. It is in a special building dedicated to servers, the server farm, right? Or a data center. Okay. So it's at a data center in like Ohio. Yeah. Wherever it is. Okay. These server farms are all over the world and connected together. They more or less make up the internet. Oh. And so. When the request comes in. When, you know, young me types in hamsterdance.com and presses enter. that server... The actual infrastructure that's holding it.

23:13...who is just sitting around in Utah or whatever on the server farm gets a little notification. Oh, someone wants to see that hamsterdance.com website. All right, not too busy with anything else. So it does a little computation. Go ahead and get that for you. Some execution of some Java code or something. Let me see here. I think it's around here somewhere. And then push that content back to UI internet. Here you go. where it pops up on my family's desktop computer back in 1999 and then, diddly diddly, and enjoy it. Then the server just goes back to hanging out on the server farm. But then, oh, another person wants to see that website.

23:54All right, let me just go ahead and get that for you. There you go. All right, where was I? Man, I just love this country. Oh, oh. Oh, no. And as it starts to go viral, a number of people are all bombarding with requests. Until that one server is like, oh my God, somebody help! can no longer serve up this website in a timely manner. And so now people have to wait in a queue, right? And this would happen all the time because back then the way servers were allocated was like you'd get how many servers you paid for. Right. Like the owner of the Hamsterdance website probably would have been like, hey, I'll just pay for the one server.