And today's episode, we're going to be talking about something that you do every day and that you rely on for survival, but it also can be a dangerous addiction. My name is Jonathan Cutrell, you're listening to Developer Tea. My goal on this show is to help driven developers like you find clarity, perspective, and purpose in their careers. Many people, like myself, enjoyed watching the television show Lost, and we especially enjoyed the mystery of the show. In particular, one thing stands out to me, and a lot of people probably share this, the string of numbers, 4, 8, 15, 16, 23, 42. Now, we're not going to do any spoilers on this show, but one of the reasons why this was such an engaging show and such an engaging pieces like this particular string string of numbers, is that our brains are working on making sense out of them. What exactly could they mean? And there are some things in Lost that were gratifyingly provided. The sense was given to us, the meaning was given to us at some point in the show, and other things where the meaning remained. But I want to talk about a host of biases today that all kind of revolve around this same idea, the same concept, the same behavior that we have that can be both absolutely critical to our functioning as human beings, but also detrimental to our way of understanding the world accurately. We've all heard the idea that correlation is not causation. And yet when we see strong correlations, it's very hard for our brain to not connect these things. And part of the reason for this is because when we see the correlations, typically they are presented in a way that they are kind of combined together. The correlation itself then becomes a sense-making mechanism. When we have no other context, when we have no other variables to consider, perhaps there are three or four other things that could be correlated in the same fashion. But when we don't have those other correlations in front of us, our mind gets to work at helping us understand why these two things are put together. Why is it that we're talking about them in the first place? If there's no relationship, then why would anyone go through the work of checking the correlation? Now, the interesting thing here is that there's some cultural context, there's some behavioral context around what our brain is doing in this sense-making. The behavior here that is really salient and very important was described in Danny Connman's book, Thinking Fast and Slow. He talks about what you see as all there is. When we only have these two things in front of us, they become like actors in a play. Where the play is something that we're trying to make sense out of, and the only things that are taken into account are the players, the actors in the play. We do something very similar when we see things like patterns. When we see streaks in data, or even when we see random clustering, you can see good visualizations of this with a quick Google search, but if you search the clustering illusion, what you'll notice is that there are randomly generated plots of data. Let's say on an x, y graph, there's essentially points on that graph. Even when they're randomly generated, our eyes are drawn to the clusters. It seems as though those randomly generated graphs are not actually randomly generated, that there's something that is magnetizing those particular clusters together. When in fact, there is nothing causing those clusters other than randomness. This reality is a hard one for our brain to wrap around, for our brain to accept, because when we see these patterns, our brain has a very hard time assigning the reason, assigning the meaning for those patterns to randomness. Why is this? Well, part of it is because pattern recognition is important, and very often patterns do have meaning. At a fundamental level, patterns are a very prominent feature in how humans learn. patterns are how we create schema for different types of objects, for example. Because patterns are so important to our brain, because the understanding that a pattern means something is such a critical feature of the human mind, that is an over tuned process in our brain. In other words, we see meaning in patterns where there is no meaning, where the pattern actually is in sedenal. What we're not thinking about is what we're not thinking about. In other words, we don't process the vast amount of information that we are actually ignoring. There's another bias related to this called publication bias. You may have heard of this. Before I knew what publication bias was, I thought it meant that there was some kind of conflict of interest on the publisher's part, but it happens to be something totally different. Publication bias is the idea that a paper is only worth publishing, or it's more likely to be published when there's some statistically significant finding in the paper. In other words, you could imagine that if we had 100 studies that were pointed at trying to find a correlation between chewing bubble gum and increased cognition. In 95 or even 98 of those studies, there was no particular correlation. There was no evidence showing that chewing bubble gum had any effect on cognition. But then in two of those studies, there was an incredibly strong effect, and it was replicated from one to the next. You can imagine that these two studies would be published, and perhaps the other 98 studies would either be ignored, or they wouldn't be published at all. In both of these are different types of bias. One of them ignoring those studies is something that's done after the publication occurs. But the other one is perhaps more concerning, and that is the idea that there's no publication ever done at all because the data doesn't really make for a good paper. And you can imagine why this would be problematic. If we have a mountain of evidence that shows no particular connection between bubble gum chewing and cognition, and then we have one or two studies that show a strong correlation, we might be able to, once again, question maybe there's something flawed in those two studies. The authority of 98 studies seems to outweigh the authority of two studies. You may be thinking, well, why do I care about publication bias? I'm a software engineer, and none of this really shows up in my work. And frankly, I don't care if bubble gum hasn't effect on my cognition. I'm not really reading white papers on a regular basis. But this bias is not just about the academic world. It's about what we pay attention to. Notice that all of our various incentives to pay attention to data or to information that seems to matter are very strongly weighted towards only paying attention to things that seem to matter and ignoring the vast majority of information that doesn't really have much of a signal. In other words, we're trying to find the high signal things and we're avoiding what seems to be noisy. But very often, the problem is that we're ignoring signal that happens to be covered up in that noise. And sometimes this happens in a more unconscious way, the idea that we're paying attention to those patterns and trying to assign meaning to them. But then there's other ways that it happens that are more egocentric. We'll talk about a couple more of these biases right after we talk about today's sponsor, Linode. With Linode, you can simplify your infrastructure and cut your cloud bills in half with their Linux virtual machines. You can develop deploy and scale your modern applications faster and easier. Whether you're developing a personal project or managing larger workloads, you deserve simple, affordable and accessible cloud computing solutions. You can get started on Linode today with $100 in free credit for listeners of Developer Tea. You can find all of the details at linode.com slash Developer Tea. Linode has data centers around the world with the same simple and consistent pricing regardless of location. You can choose the data center nearest to you for low latency when you're working on your project or you can choose a data center that's close to your customers. There's a lot of flexibility here. You can also receive 24, 7, 365 human support with no tears or handoffs. That's not something you're going to pay extra for. That's just what Linode does. That's how they operate. Regardless of your plan size, you can choose shared and dedicated compute instances or you can use your $100 in credit on S3 compatible object storage, managed Kubernetes and more. If it runs on Linux, it will run on Linode. Head over to linode.com slash Developer Tea and click on the create free account button to get started with that $100 in credit. Thanks again to Linode for sponsoring today's episode of Developer Tea. We're talking about this classification of various biases that we have, different kind of illusions that come out of these biases. The pattern recognition and since making that we do in almost every scenario. This is very important. We've already mentioned it's critical to our learning. It's critical to our daily operation, but sometimes it can lead us astray. We're going to talk a little bit about the practical implications of that in a minute. I did mention right before the break the egocentric biases that come into play here. One of those is confirmation bias. If you haven't heard of confirmation bias, the idea is that we already have a lot of beliefs. Sometimes those beliefs are seated in evidence other times they are cultural or they're passed on through some other means. When we seek out information, we have a tendency to believe that our pre-existing information is correct. Even if we're trying to update ourselves, even if we're kind of seeking out information, we're still seeking out information that tends to agree with our pre-existing beliefs. In this case, we are choosing to ignore relevant information. If we're talking about attention and we're talking about pattern recognition, there's another related fallacy that tends to rear its head, I guess, in human behavior quite a lot. It's called the gambler's fallacy. The idea behind the gambler's fallacy is that if we are observing a series of events, let's say flipping a coin, and that coin is flipped five or six times in a row, and each of those times it hits heads. If I had to ask you, what do you think it's going to hit next? The gambler's fallacy tends to push people to say that it's going to hit tails next. It's going to land on tails. Why is this? Well, because we believe that a streak is somehow dependent on the existence of the streak itself. In other words, these otherwise unconnected events have become connected in that there's some kind of chain that is inevitably going to be broken, that there's some tension between these events. That's somehow they've been held together, and they're about to fall off the cliff. The truth is very much different from this. Once again, this is another pattern recognition where we recognize some underlying true statistical fact, which is that if we flip the coin enough times, we're going to get a relative representation of about 50%. We take this idea, this understanding of the underlying statistical normality of flipping a coin, and we apply it to a narrative. It seems like a rational kind of reasonable understanding of the world, but because each flip is not related to the last, it's actually irrational to believe that any flip or series of flips will have any bearing on any future flips. A rational expectation might be that eventually this coin flip will land on tails again. We have not somehow entered a magical world where we will infinitely hit heads, although there is a very small chance that that could happen, but these concepts become confusing because perhaps because of the temporality of the concept. In other words, because we are connected in some way, we can visualize the data as it's flowing in, we imagine that the fluctuations should feel like a wave that we should see on a smaller scale that distribution should be more evenly spread out. The rational idea here is, eventually we will see these things converge on something around 50% heads and tails, but I'm not going to necessarily say that one particular flip is more likely to be heads or tails, it's always going to be 50-50. And to kind of give you an illustration of how this works, because you may be thinking, oh yeah, I could easily dismiss if it had five or six times in a row, I could easily say that I believe it's going to be heads or tails equally on the next flip. But what if it's been hitting it for a thousand times in a row? Or if it's never hit tails yet, all of the flips that you've ever flipped this coin, even as I say the sentence out loud, my mind says there must be a reason. Even if somebody told me this coin is a fair coin, that there's nothing tricky about it, it should theoretically represent the 50-50 split. If I see that it's hitting heads on every single flip, my brain wants to make sense of that. It seems like an anomaly, it seems that there's something happening that's not. And this is how this applies to you as a software engineer. If you stuck with me here, you probably have kind of felt some shadows of how you behave with your software, especially when you're, for example, looking for causes of failure in your code. This applies both at the minutia level, where you're looking for a bug and you start chasing down a log trail, because you see something that you think means something, but as it turns out, it didn't mean anything. You could do the same thing with statistics that you're gathering on behaviors around your application. But you could also imagine that this happens at a much higher level, that the role of randomness in our successes and failures is often discounted. It's ignored. And we look for patterns. We look for, for example, patterns of success. And when we find them, we pay a lot of attention to them. And we even look for patterns of failure. And we try to assign information to those patterns of failure. We try to assign meaning. We try to find blame. We imagine that we see all of the pieces of the puzzle that they're all laid out, like the players in the play. But the truth is, in most problems of large enough scale, the data is actually much more like that random plot. And the clustering that we see is probably more to do with randomness than it is with meaning. So what should we do? Should we ignore data altogether? Should we imagine that none of this really matters and we can't really observe anything completely? Absolutely not. Perhaps the best thing we can do is continue trying to make sense. But observe the randomness. Try to explicitly pay attention to what we previously were ignoring. Seek out disagreements and seek out data points that don't necessarily confirm your existing ones. Give credit to randomness and luck more often. We should be very wary of anybody who claims to have some secrets that no one else has that they attribute to their success. More than likely, that secret is a correlation that doesn't really have causation. Thanks so much for listening to today's episode of Developer Tea. I hope this was a challenging and interesting discussion on these various biases of attention. I hope you have an excellent week. If you would like to join the Developer Tea Discord community, you can reach out to me directly on Twitter to ask for an invite. That's at Developer Tea. You can also reach me at developertea@gmail.com. A huge thank you to today's sponsor, Lynneaued, head over to Lynneau.com slash Developer Teato get started with $100 worth of credit. Make sure you create the free account. That's a button on Lynneau.com slash Developer Tea. Thanks again for listening to this episode. If you don't want to miss out on future episodes like this one, subscribe in whatever podcasting app you currently use, where we have been releasing three episodes a week as of this year. It's very easy to miss out on something that you will care about. You will actually enjoy listening to you. Subscribe in whatever app you're currently using. Thanks so much for listening. Until next time, enjoy your tea.