If AI is making us more efficient, why isn’t the economy improving?
Download The Business Of podcast on your favourite podcast platform.
AI should boost productivity and growth. So why isn't it showing up in the numbers? UNSW economist Professor Kevin Fox explains
About the episode
AI is supposed to be boosting productivity. So why isn't it showing up in the numbers?
UNSW economist Professor Kevin Fox explains why traditional measures of economic growth may be missing part of the story. If you're using free AI tools for financial advice, health guidance or landscaping ideas instead of paying a professional, real value is being created, but it may no longer be captured in measures like GDP.
Professor Fox explains what the data says about AI's impact on the economy, why productivity gains have been hard to measure, and how economists are rethinking the way growth is recorded.
The Business Of podcast is brought to you by UNSW Business School, produced with Deadset Studios. It’s hosted by Professor of Practice Fahim Khondaker.
Want to know more?
Want to follow the latest research and news from UNSW Business School and AGSM @ UNSW Business School? Subscribe to our industry stories at BusinessThink and follow UNSW Business School and AGSM on LinkedIn. Or listen to more episodes of The Business Of and subscribe to our UNSW Business School and AGSM channels on YouTube.
Edited transcript
Prof. Kevin Fox 00:00
The big part of the picture here is invisible. That's how households are using AI for all sorts of things, including engaging with the government, choosing running shoes, and getting general health advice. They're not paying for the services as they may have done in the past. So that activity is disappearing from GDP.
Prof. Fahim Khondaker 00:22
AI is set to reshape the global economy. Businesses are automating workflows, investors are pouring money into technology stocks, and data centres are popping up across the country. So when you look at the numbers, is our economy actually better off? The way we measure GDP was never designed to capture these powerful AI tools or to account for what happens when an algorithm starts replacing services. So, how do we know whether AI is actually making our economy stronger? This is The Business Of – AI Explained, a podcast by the UNSW Business School. I'm Fahim Khondaker, Partner in Data and AI at BDO, and Professor of Practice at UNSW. Joining me is Professor of Economics, Kevin Fox. He's the Director of the UNSW Centre for Applied Economic Research and a member of the Leadership Executive at the UK Economic Statistics Centre of Excellence. Welcome, Kevin.
Prof. Kevin Fox 01:15
Thank you.
Prof. Fahim Khondaker 01:16
Whenever we talk about AI, or whenever people talk about it, some of the words that come to mind are jobs, productivity, and economic growth. What do we know about that at this stage?
Prof. Kevin Fox 01:27
Well, when it comes to jobs, I think this is the one people are most concerned about. Is AI going to make us all unemployed? At the moment, it doesn't look like that. AI is mainly being used to augment jobs. With any major technological change, like AI, a general-purpose technology coming in, we do get a bit of reorganisation of the workforce, but we also have new jobs emerging. So we're not suffering from mass unemployment at the moment. What's happening, though, is that certain tasks within occupations are being done by AI. So there's automation happening there on very routine cognitive tasks. How that's playing out in terms of productivity at the moment is unclear. For any particular job, you can see productivity gains, but that's not necessarily showing up in the aggregate data yet. There was a lot of perhaps overemployment immediately following the pandemic. A lot of people changed jobs, and part of what we're seeing in some occupations is simply due to the continued shakeout after the pandemic. There do seem to be some slowdowns in employment growth in sectors where they're most exposed to AI. But much of the slowdown in junior hiring started around 2022, so it's unclear whether this is all due to AI or still part of the ongoing shakeout. So that's one reason why it's a bit unclear. Another thing is that AI coming in means people can use it at home, not only in the workplace. In fact, AI can be used for about 88% of occupations in advanced economies. However, they're not used that intensively, and the use is very shallow. In other words, only 21 of the tasks in a median occupation could be done using AI. The big part of the picture here is invisible. That's how households are using AI. Households are using AI for all sorts of things, including how to engage with the government about, say, taxes, getting financial advice, cooking recipes, choosing running shoes, general advice about health; these activities, which they might have engaged with a person on in the past, they're not doing that now. They're not paying for the services as they may have done in the past. So that activity is disappearing from GDP. Instead, households are using a free platform, typically a free AI agent, to help them navigate tasks they would previously have hired a professional to handle. Now this is not unusual in the digital economy. We've seen this for a number of years now with things like the whole process of taking a photograph used to involve a separate camera, buying film, and getting the film developed. All those things had costs, just like sending a letter. You needed an envelope, some paper, and a stamp. We don't send too many letters these days. We send texts and emails, and for each one the marginal cost is zero. Same with photographs; you take more photographs than ever before, but for each one, its marginal cost is zero. So all that activity that used to be measured in GDP is no longer there.
Prof. Fahim Khondaker 05:13
First of all, it sounds like you're measuring a lot of things, which, as an economist, I imagine there are economic models for that. Would you say the models are still fit for purpose in the world of AI, or do we need to change the models much?
Prof. Kevin Fox 05:26
Okay, so if we want to understand what's happening with productivity in the economy, it would be sensible to capture what's going on in the household, the household production. So it would be advisable to try and do a better job of measuring the impact of the digital economy, the spread of AI, on that. Otherwise, we're just seeing stuff exiting our main measure of economic activity, Gross Domestic Product (GDP), without taking into account the huge boom in household activity.
Prof. Fahim Khondaker 05:57
And just out of curiosity, is that as simple as updating the census we do every four or five years with a few more questions, or how does that actually happen?
Prof. Kevin Fox 06:05
Unfortunately, it's not that easy. The problem with household production is complex, which is why it typically hasn't been included in GDP. The reason it hasn't been there is that it's very difficult to measure. National statistical offices have made a crack at this, so we want to do a better job of measuring it. Now, what's happening? You said, models. Well, it's really frameworks. The United Nations has a System of National Accounts. This is a framework they recommend countries adopt for measuring economic activity. And the main part of that, the most well-known part of the System of National Accounts, is GDP. So there's just been a revision of the United Nations framework. It's called the SNA 25, and it includes recommendations to make a better effort to measure the impact of AI on the economy. Now, it provides a framework. It doesn't tell countries exactly how to do that. So, one of the recommendations is that national statistical officers are currently looking at ways to implement it. The leading agencies are looking to implement around 2030–31. So, it's going to be a while before we get a good read on many of the things we're currently interested in; there's a lot of debate about AI that isn't data-informed. That will change, but it's coming. Like, the ABS has a plan to do a better job of understanding what's happening in firms through the Business Characteristics Survey, but that data isn't available yet.
Prof. Fahim Khondaker 07:44
Could you give us a quick overview of what we mean by productivity? And when we talk about productivity growth, what does that mean?
Prof. Kevin Fox 07:51
Productivity is often misunderstood. It's often confused with profitability. Firms interpret it as, "Well, I spend this much money; how much revenue do I get for that money? The way it's measured by the Australian Bureau of Statistics is output growth divided by input growth. That is, if you can get more output growth than your inputs are growing, then you're better off.
Prof. Fahim Khondaker 08:16
And then, if we're talking in the context of AI, what are we talking about when we say we're going to invest in AI to get productivity growth? What do we mean by that?
Prof. Kevin Fox 08:24
Yeah, so if you're investing in AI, you're growing your capital stock. There'll be the data centres, but there's also the knowledge. There's the R&D that has happened. So you get all that activity around AI investment being capitalised, and then you're hopefully getting some extra value added, some gross domestic product from that. So you look at this measure of output growth divided by input growth, and if AI is giving us a whole lot more output growth than inputs are growing, then that's productivity growth.
Prof. Fahim Khondaker 09:02
So you mentioned 2030–2031 is when we're going to see some actual data. So in the meantime, what are we doing? Are we using assumptions and estimates?
Prof. Kevin Fox 09:11
Yeah. So we are getting data from countries such as the UK and Ireland. The US is moving forward to get a better grasp of things before 2030–31, and the ABS (Australian Bureau of Statistics) as well, but a broader picture. I think we'll need a few years of data before we can identify any patterns.
Prof. Fahim Khondaker 09:34
For spoiler alert, those of us who haven't seen any of this, are we seeing a significant uplift in productivity at the household level?
Prof. Kevin Fox 09:42
Well, first, at the industry level, if we look at the market sector, we're not seeing a big boom in productivity at the moment. Just like we're not seeing mass unemployment at the moment, so the productivity picture isn't clear. Some people think in the US we're starting to see an uptick, but it's a bit cloudy because of other factors in the economy. It's not just AI coming in, and even if it is an impact of AI, is it the productivity arising because of the construction of huge data centres? I mean, the amount of investment going into data centres is massive. It's that infrastructure spending is more than many countries spend on infrastructure in a year.
Prof. Fahim Khondaker 10:29
And you talked earlier about things coming out of the economy because households can now do it themselves using AI. Can we dive into that a little bit more? So, what happens in the traditional way to the economy when those sorts of roles just disappear?
Prof. Kevin Fox 10:45
Yeah, so if households stop paying financial advisers and use AI instead, there'll be fewer financial advisers, and so that activity in the economy will disappear. But that doesn't mean it's not happening. The households are producing those financial services themselves by using their time and prompts to get the advice they want from a digital platform. In this case, AI, but that's unmeasured currently, and we're working – myself and a team – ways of trying to measure that sort of activity in the household work that's ongoing with the Productivity Institute at the University of Manchester and the Economic Statistics Centre of Excellence at King's College London.
Prof. Fahim Khondaker 11:34
But it's not just going to be financial planning, right? It's going to be lots of roles, probably all at once. What do we do about that? How do economists measure the economic impact of that?
Prof. Kevin Fox 11:45
Well, that will appear in the national accounts as usual. If an entire occupation disappears, well, it won't happen overnight. There'll be a transition, and as we've discussed, other jobs will emerge, but that will be measured. There are new jobs within the classical labour accounts that Australia has, for example, that didn't exist. You know, there weren't AI and data officers in a company. These are new jobs. There's always some transition between occupations, and it usually happens rather slowly. With AI, people have suddenly seen it everywhere, and it can be used in almost 90% of occupations. And that's what has people worried: will it make us all unemployed? It hasn't happened yet. Superintelligence may change that, but that may be a way off.
Prof. Fahim Khondaker 12:46
And is that the main reason we need to measure this: to predict that transition and adjust policy and economic measures from that perspective?
Prof. Kevin Fox 12:53
We need to understand the economic impact because it may help inform policies on AI use and legislation that could protect workers. It could also be about protecting the environment. At the moment, it seems like there's at least politically, in many countries, a desire to be at the front of AI adoption, including approving the building of a lot of data centres, yet we don't have proper accounts of the water and energy that these data centres are using or could be using in the future. A lot of the estimates come from the industry itself, and you know that sort of self-reporting perhaps shouldn't be believed. It's not in the public interest to do so. For example, in Sydney, the industry says that by 2030, about 2% of the city's water usage will go to data centres. Independent assessments say 25% of usual Sydney water consumption will go to data centres. This is a problem if it's 25%. That, you know, if we have a drought, it will have a material impact on people's wellbeing.
Prof. Fahim Khondaker 14:13
Here's another podcast you might be interested in. Have you ever wondered what you'd do if someone made a deepfake of you? I've just listened to an episode of The Conversation Weekly about how far celebrities are going to protect themselves from exactly that. It also covers what the rest of us, without the massive budget, should be thinking about. The episode is called, "How celebrities like Taylor Swift are fighting back against AI deepfakes." Search The Conversation Weekly, where each week they speak to an academic expert about the big story in the news, and to understand how they got here, wherever you get your podcasts.
Prof. Fahim Khondaker 14:45
Let's bring it down to a slightly more micro level; let's say a workplace, and AI is obviously an opportunity to improve the output of that business or the productivity, not just productivity, the profitability of that business. How would I go about boosting my profitability using AI, and what's the conversation happening in the economics world?
Prof. Kevin Fox 15:07
We can relate this to earlier technology that emerged and the delay in productivity growth observed at that time, known as the Solow Productivity Paradox. This relates to the introduction of personal computers into workplaces through the 70s to the mid-80s. A lot of investment went in, and Solow quipped that you can see the computer revolution everywhere except in the productivity statistics. And we can say the same thing about AI perhaps now. So what happened? How did we resolve that productivity paradox? Well, in industrialised countries, productivity growth took off around the mid-'90s. The most popular explanation for this is that workplaces needed to reorganise around the new technology. Just buying a personal computer, putting it on a desk, saying, "Well, now we've got a computer, productivity is going to go up." That didn't really happen. That doesn't work. You need to rethink how you can do business. So, if people are having routine cognitive tasks taken out of their occupation, what else could they be doing? How could the workplace be reorganised? What are the opportunities that opening up that extra time for workers? What does that give the workplace? Does it mean you just cut your costs, or does it mean you can expand into different areas? Just do things differently, and the real benefits – the productivity that'll lead to profitability – will be in working out how to better organise the workplace and jobs, so that you are taking full advantage of the new technology.
Prof. Fahim Khondaker 16:56
Controversial question: Will it require economic intervention, or will the free economy sort this out by itself?
Prof. Kevin Fox 17:04
I doubt that the free economy will sort things out smoothly by itself. There will be some adjustments in the labour market, as some jobs and occupations may change. Do we really need so many people doing coding these days when AI is so good at it? So there could be some need for safety, social welfare, a safety net. We already have quite good safety nets, so it may not need anything too much more. Well, something that I'm worried about is the large tech companies self-regulating simply because we don't. The public doesn't know what they're seeing. They've expressed some anxiety about their own progress and actually have colluded to slow down progress. Activity, you know, when people and businesses get together to collude, to deprive the public of a better product, that would usually be subject to legal action by the government. In this case, we don't know why they're wanting to slow down. It makes us nervous, and we're willing to accept that. Well, if they're slowing down, then they must. There must be a reason. I think that should be out in the open. I don't think these firms should be hiding what they're seeing. It makes people fearful, justifiably fearful of the technology, and that is unhelpful for adoption and for society in general.
Prof. Fahim Khondaker 18:47
In terms of adoption of AI in Australia, it's part of our national AI plan for the federal government and things like that. I'm really keen to see: are there sectors that come to mind in your research and where you think we're primed to benefit from AI? And I think we talked a little bit about knowledge-based jobs in particular. But are there other sectors, not just knowledge-based ones?
Prof. Kevin Fox 19:09
So it's not just the desk-based jobs where AI is being used. So about 90% of occupations can use AI. There's very little end-to-end automation, even in knowledge jobs, for tasks that could potentially be automated; only about 10% of those tasks have even been tried yet. So it's unclear how that's going to play out. But I would say it's a bit the wrong way to think about AI to say it's only about jobs and the service sectors and knowledge sectors, because it can have industrial applications as well, and in the health sector as well, like emergency departments under stress. Emergency department staff can use AI to handle routine cognitive tasks, allowing them to focus on non-routine work. So it's going to be in not only knowledge-based jobs; it's going to be throughout the economy.
Prof. Fahim Khondaker 20:16
You mentioned healthcare, and it reminded me of the view that AI could help with longevity, life expectancy, and related areas. Would see that flow through in the economy as well, in theory. Again, is any modelling being done, or are frameworks being followed to assess the impact of that, and how's that being managed?
Prof. Kevin Fox 20:36
That is a very good question, and it's very difficult to answer. We get better healthcare. We have a healthier workforce. You should see some productivity benefits from that. But that investment in better health and the resulting productivity comes with a great big lag. And so we've been interested in this issue for a while. Does increasing the welfare, health and welfare of your workforce have immediate productivity benefits? We would think yes, but it's hard to identify. However, with the promise of breakthroughs in medical technology now enabled by AI, most serious diseases will be cured within a decade. Some people believe that will be quite sudden. Now, whether we see that as people being healthier, enjoying more leisure or enjoying a longer work life, that's yet to be seen.
Prof. Fahim Khondaker 21:38
But big tech companies have now warned us that they feel it might be going a little bit too fast from their end, and they may have various reasons for that. What do you make of that conversation and the warning?
Prof. Kevin Fox 21:52
It is extremely worrying, and the response to it is, too. That, well, we'll just leave it to them. Then that seems inappropriate. Any company that produces a product that could harm the public needs to disclose what that harm is. We can think about tobacco. For many years, the tobacco companies tried to deny that their products could cause harm. We don't want companies behaving like that. Now we've seen in Australia these AI agents running rogue and hacking government sites, and they're no doubt hacking private companies' sites as well. The producers of those products can't just shrug, I believe, and say nothing to do with us. It's our agent gone rogue. I don't think that's acceptable, and I think you know this is not acceptable as an economist. This is as a regular citizen. I am very concerned about that, and I think legal action needs to be taken against firms that produce harmful products.
Prof. Fahim Khondaker 23:11
What are you optimistic about from an AI perspective and AI in the economy?
Prof. Kevin Fox 23:16
I believe it's a great new technology. It presents us with many opportunities. It takes away routine cognitive tasks from us, things that we really don't want to do. AI will do for us, generally pretty well. It gives us more time to either consume leisure or to work on other things, and you know, the number of things it can give advice on is quite startling. So I'm very impressed and optimistic about what AI can do to provide us with information that can help us make better decisions. What I'm pessimistic about is the significant risks and downsides. With employment, I think we'd need superintelligence before we really need to start worrying about that. But there will be some difficult job transitions that will happen. People are going to feel justifiably threatened by some of the aspects of AI, and that's a job for the managers to see how they're going to respond to that. Are they going to say, "Well, that's just tough. We don't need so many workers anymore. Or are they going to say We have opportunities now to take our business in different directions?
Prof. Fahim Khondaker 24:49
I'm going to give you a crystal ball. Where do you think AI is going to take our economy and society at large?
Prof. Kevin Fox 24:55
I think there will be some productivity growth coming from it. It will eventually show up in the statistics, so this will be good in the long run. Productivity growth means we're getting more output for the same amount of inputs, and that makes us better off. There will be some redistribution issues, and not everyone will benefit equally; there's a role for policymakers in ensuring that some of the gains from that are shared across the economy.
Prof. Fahim Khondaker 25:31
Great, Kevin, thank you so much for joining us.
Prof. Kevin Fox 25:34
My pleasure.
Prof. Fahim Khondaker 25:36
Kevin Fox is a Professor of Economics at UNSW. If you enjoyed this conversation, hit subscribe wherever you're listening or watching, and leave us a rating or review. It helps more people find the show. The Business Of is brought to you by the University of New South Wales Business School and produced with Deadset Studios.