Abhas Gupta maps work tasks on a 2x2 grid of uniqueness versus complexity. Sweeping floors is repetitive and simple; software development is novel and hard. Jobs in that high-uniqueness, high-complexity corner, like loan underwriting and brand strategy, are most resilient to automation.
Chapters
So this is how— I'll start with looking at the framework. It's essentially like a 2 by 2, so think of 2 axes. On the x-axis is uniqueness. So something that's low on the uniqueness is something that's really highly repetitive, like ordering at McDonald's. It's the same menu every single time. There's only 25 different possibilities. very repetitive. And then something that's highly unique is something that's non-repetitive. Something like, I don't know, arguing— sorry? You want to throw out some options? Michelin restaurants, like good restaurants that they have all these weird combinations. Yeah, yeah, yeah. Interestingly, I would put that on the low end of the spectrum because in order to do that, they are very— Let's think about that. Let's take that example. The McDonald's is something— Yeah, OK. So the y-axis is complexity, low and high. So a Michelin restaurant, you can think of the chef who's creating the actual menu as being really high complexity and not very repetitive. You do that— it's not like you create the menu only once, essentially, right? But the actual execution, the making of the meals, is highly repetitive and probably low, mid of complexity. There's probably some new techniques that they use or whatever. So this is sort of the 2 by 2 grid. And what I'll do is start to plot how different tasks fit on this grid and then kind of use that to sort of explain where our jobs are going. All right, so let's look at the first. So I'll go through each quadrant and talk through some examples. Something that's not very complex and also not very repetitive is like sweeping the floor. I would think of things in this category tend to be like a high school education is probably appropriate, or things that are low on this y-axis. High school education is probably appropriate. Kids can probably do half this stuff. That's sort of low, whereas up here I would think more like graduate students, real specialists in those fields. And then you know about the x-axis already. So sweeping the floor, picking fruits off of trees, or picking vegetables, whatever. Dispensing cash at a cash register. Kind of put that— that's something that it's not particularly unique. It's very repetitive. You know it's whatever the amount is, whatever you get in minus the amount, give the change back. You're picking from 6 different containers to give someone their change back. Pretty straightforward stuff. Checking out a customer, similarly, you're scanning all the tags and figuring out what the total is. So let's go over to this other quadrant now. This is stuff that is unique and low complexity. So things like driving on the highway. Driving on the highway is something that's a little bit— it's not completely— it's a little more unique because turns are different. You're going into new territories. You kind of— there's construction that's going on. There's all kinds of challenges that are being raised. But it's also still relatively low complexity, because you basically have 2 lane markings, and you're trying to stay between them. And that's kind of the idea. And sure, there's some exceptions, but for the most part, it sort of fits in that category. In the top left, now this is stuff that is low on the unique scale, so it's very routine, and it is highly complex. Interestingly, I put folding laundry on that because that's one of the classic problems in AI. But folding laundry is a really challenging problem for machines to do. It's like they look at a piece of clothing and they don't know which side to pick up. And if the cloth is folded under, they can't forecast where that piece is actually folded under. It's a very human thing that we can do instantly, but machines can't do very easily. Building a regression model. How many data scientists in the room right now? Wow, really? Are you guys all entrepreneurs? Yes. Okay, well, you might have to hire some data scientists at some point. But so, so building a regression model, you know, you've got a data set. I mean, I'm sure you guys have done this in school, but you've got a data set, you're trying to fit that data set to some kind of Regression model. And so you sort of take the data, run it through your model generators, and then test the model. And that's sort of the kind of 3-step process. Highly complex in terms of understanding, are these the right inputs? Does the actual model fit? Interpreting those results are complex things, but still relatively low on the uniqueness scale. It's kind of repetitive. Every time that you create a model, it's pretty similar to the models that you've created before. At least the process is pretty similar. So you're saying that the model All right, so now let's look at the highly unique, highly complex category and what fits into there. So things like recognizing faces, relatively tough thing to do. A lot of nuance to a face. Humans do it instantly. Machines have had a really tough time. I think this is sort of the first decade in the history of computing where we're now recognizing faces is becoming a lot more tractable of a problem. City driving, there's all kinds of craziness that you can't even predict. Sometimes there's not even lanes, cars that are merging. There's all kinds of things that make city driving more complex and pretty random, or pretty unique, I should say. Things like underwriting loans. People have different situations. You've got to think through what's their background. So every single person's financial history is a little different than the rest. So that's something that I kind of put in that category as complex and pretty unique. Things like writing code takes a lot of creativity. Some people will call this whole category like the creative category. But writing code, it takes— you've got to think through what's the most elegant approach to— elegant way to approach this problem, those kinds of things. And then finally, I think things like building trust. So one human interacting with another human, that's a pretty complex— looking at understanding their facial expressions, understanding how their body language, maybe making forecasts or predictions about what this person may like or not like. I could say, maybe with your Converse shoes, I should kind of dumb down my language. I don't know. Or I should be a little more free-willed in my language with you. So that's something that I might pick up as a cue on how to build trust with someone. Or someone that's uptight or someone that's not, you can kind of— anyways, you know what I'm talking about. And then something like crafting a brand strategy. How many marketers are in the room? 2. OK. Great. All right. So thinking about what does the brand represent? What is the right— what messages are we trying to put out there in the world? What's going to get us more customers? That's a very creative, strategic process. I think that's— and it's unique to every brand. So I put that highly unique and highly complex activity. All right? Any questions?
