AI won’t cause a “jobs apocalypse”: Google Brain founder Andrew Ng — The Economist

[00:00:00] Speaker 1: There will be no AI jobpocalypse. This idea of 50% being out of work, 50% of people being out of work, maybe rioting in the streets, I don't know. I don't see any realistic part of that. And in fact, early data, even out of some of the organizations that were hyping up the 50% unemployment jobpocalypse scenarios are showing that AI is not having the negative impact on jobs that people were fearing. Because AI automates only part of someone's job, it turns out we really need people to do the rest of their job still. And if what happens in software engineering is a predictor of what will happen in other job sectors as well, I think it'll be fantastic. And if anything, we can't find enough software engineers right now. And I'm already seeing we can't find enough AI native marketers or enough AI native salespeople, for the challenges of skilling because for all of these job roles, we have to help people learn new skills, and I don't just mean learn AI skills, but it turns out that maybe I'll take one example on software engineering. A lot of people used to be maybe still are front-end engineers, meaning they build a website or an app to make it look good. But AI is pretty good at front-end engineering. So a lot of front-end developers are now full-stack developers, meaning they don't just build the part that users look at, but they build what runs in the background as well. And they build a mobile app, they build security links. And so a lot of developers are now taking on broader scope in their work, and what that means is, instead of just focusing on what it looks like, they have to learn to build all these other pieces as well, and they have to learn AI skills. Well, when they do that, they're amazingly valuable in the job market, and we just can't find enough of these people. I'm seeing early signs of a similar transformation happening as well in other job galleries. [00:02:03] Speaker 2: I'm curious, though, you know, you say there's not going to be a jobs apocalypse. I'm curious what that implies about your sense of the eventual progress of AI. Because the fear about a jobs apocalypse isn't the AI we have in 2026, it's the AI we have in 2030, or if your timelines are short, it's the AI we have in six months, right? Few people are arguing that the AI today can do every task in a job like software engineering, but the fear is that if we get this compounding rate of progress for another year or two, it's hard to see where it stops. Are you implicitly arguing that you think it will stop, that we will hit a hard limit at some sort of your scaling laws, or some architectural change will be needed? And there will be some tasks that the AI systems of tomorrow simply will not be able to do and probably never will be able to do? [00:02:56] Speaker 1: Well, I don't think we'll hit a hard limit, but progress will be incremental for a long time to come. So I would love if AI could do any intellectual task if human can. Maybe I'll go on holiday or something. But realistically, that would take multiple novel breakthroughs, and I'd love to see it within our lifetime. But whether it'll take decades or even longer than decades to get there, I don't know. And until AI could do 100% of what people can do, which seems incredibly far away, I don't think we'll get there anytime. Certainly not in six months. Almost very unlikely to get there by 2030. There's still plenty of work for people to do. And use the term compounding progress. One of the terms of hype in Silicon Valley has been compounding progress or recursive improvements and so on. And at least right now, that's much more hype than reality. For decades, computers have made computers better. I mean, we used the last generation of semiconductor microchips to build the current generation of semiconductor microchips. And yes, at the Frontier Labs, AI writes all the code, almost all the code. So, of course, AI is helping us make better AI. But it turns out that if we look at the bottlenecks AI progress, a lot of the bottlenecks are we have to build new GPU servers. We have to get power plants built. We have to run a lot of experiments to see what works. So, unfortunately, even super intelligent AI, if we can get there, needs to obey the laws of physics. And the bottlenecks are the time to build data centers and run experiments to figure out what to do, rather than just, you know, thinking in the box. Then those are fundamental bottlenecks. Because the bottleneck to more intelligence is not just to be smarter about things. That these are the bottlenecks as well that seem very difficult to overcome. [00:04:54] Speaker 2: If you're not concerned then about a jobs apocalypse, are you still concerned about some narrow impacts? I'm thinking, firstly, of the impact on early stage careers. You brought up your Stanford colleague, Bryn Jolfsson, who did some fantastic research last year, showing that it seems like the more exposed a sector is to AI, the worst job prospects are for people entering that sector. So that research was pretty convincing to me, and it seems to be borne out anecdotally, that getting a job as an entry-level software engineer with no skills, no previous work history, is getting quite hard. Because you don't have much to offer in a way that you may have if there were roles of, you know, just being a code monkey while you learn for a couple of years. [00:05:41] Speaker 1: Yeah, so Eric's canary paper was very thoughtful, I think he actually has new data since then that paints a more optimistic picture. But, to take software engineering again as an example, I think the challenge is not the lack of jobs, the challenge is the slowness of our educational system to adapt, to train graduates to do the jobs that are now in demand. So, I think it's really unfortunate that today in many universities, it's still possible to get a bachelor's degree in computer science without having made a single API call or network call to a frontier AI model. And today, I'm not going to hire a software engineer that's never done that. It's still possible to get a bachelor's degree in computer science in some universities without using an AI coding agent, and frankly, none of my teams will hire a software engineer that doesn't know how to use an AI coding agent. So, to me, the challenge is not just jobpocalypse, we don't need junior people. Oh, over this last summer, my team had a lot of interns, including high school students, current college students, fresh grads, and they were great. They did a ton of work, I'm really glad I had all these interns, but the trick is I hired only AI native interns. So, the job's out there, but we've got to fix the educational system. [00:07:00] Speaker ?: We've got to fix the educational system. We've got to fix the educational system. [00:07:00] Speaker 1: We've got to fix the educational system. [00:07:01] Speaker ?: We've got to fix the educational system. We've got to fix the educational system. We've got to fix the educational system. We've got to fix the educational system. We've got to fix the educational system.