As advanced economies battle high inflation, low growth, stagnant productivity and labour shortages, could salvation be on the horizon?
Last month, economist Dario Perkins of TS Lombard described Artificial Intelligence (AI) as “the world’s best opportunity for avoiding a nasty stagflationary slump”.
It is easy to be cynical about the next big technological innovation. After all, blockchain and virtual reality also piqued our interest but have made limited material impact on the economy since. But AI could be different. It is hailed as a GPT – ‘general purpose technology’ – an innovation generic enough to be widely used, have many uses, and deliver wide spillover effects. The historical parallels certainly look promising
David Page, head of macroeconomic research at Axa Investment Managers, compares AI to the great technological revolutions of the past. While blockchain is caustically referred to as “a solution in search of a problem”, there are hopes that AI could offer a remedy to a range of very real economic concerns.
Productivity boost
At the heart of things is AI’s potential to supercharge productivity gains.
Sam Altman, the CEO of OpenAI, asserts that training models such as ChatGPT with more and more data will see their ability improve exponentially. The practical implications could be huge: according to Altman, “the technological progress we make in the next 100 years will be far larger than all we’ve made since we first controlled fire and invented the wheel”.
Altman envisions a world where “for decades, everything – housing, education, food, clothing, etc – become half as expensive every two years” as new technology drives efficiency gains.
This sounds far-fetched, and is clearly a self-interested opinion. Yet you don’t have to subscribe to Altman’s very maximalist view to concede that AI could transform productivity.
MIT researchers found that customer support agents assisted by an existing AI tool saw a 14% increase in productivity, which rose to 35% for the lowest-skilled workers. Even today’s technology – which Altman sees as “primitive” – looks powerful enough to move the dial.
Deutsche Bank analysts think that “the benefits of AI may accrue most significantly to companies that have been underperformers”, triggering an equity market boost. As with deregulation in the 1980s, the tech boom in the 2000s, and quantitative easing in the 2010s, AI-based labour productivity improvements could be “the shot in the arm that corporates and equity markets need”.
The nature of compounding means that even small productivity gains would make a big difference over a longer horizon.
Economists Martin Baily, Erik Brynjolfsson and Anton Korinek point out that if AI lifted annual productivity growth from 2 to 2.4%, the economy would be 5% larger after a decade. If we allow for the possibility that AI will be able to “improve” itself, as Altman believes, the gains could be far greater, increasing not only the level, but growth rate of productivity, too.
Analysts at Goldman Sachs forecast that if AI “delivers on its promise”, we could see annual global GDP increase by 7%.
Some economists think that firms with large workforces will be the biggest beneficiaries.
According to analysts at Deutsche Bank, AI will be able to streamline the kind of “repetitive and mundane tasks that are more prevalent in high-staff companies”. This means that the sort of ‘AI-assistance’ that transformed productivity in the MIT study could soon become increasingly widespread.
But by characterising AI as a kind of benign ‘assistant’, do we blind ourselves to its ability to replace human workers? As the chart below shows, Goldman Sachs’ analysts calculate that generative AI has the potential to substitute up to a quarter of current work. If so, we could be on the brink of a mass unemployment event (but what of the plus of a much shorter working week for many e.g. Keynes’ 15 hours?).
The good news is that, historically, this kind of displacement has (eventually) led to new jobs. According to Goldman Sachs’ analysts, “the emergence of new occupations following technological innovations accounts for the vast majority of long-run employment growth” – although this might be cold comfort to those employed in offices and administration or legal, architecture, engineering, business and financial and sales operations – with much less impact expected on food preparation and serving, construction and extraction, cleaning and maintenance operations – plus nursing and care work
AI’s ability to deal with cognitive tasks also has implications for the kinds of workers affected.
TS Lombard’s Perkins notes that, when faced with computerisation, many employees were able to ‘skill-up’ and take jobs that used the new technology. But this time it could be different: he says that AI could dislocate “a broader range of skills and occupations, at an unprecedented speed”. We might face significant downward pressure on wages as a result.
With wage growth troubling policymakers, this has implications for inflation, too.
AXA Investment Managers’ Page told Investors’ Chronicle that “on balance the introduction of AI should be disinflationary, but the exact outcome will be determined by the context into which it is introduced”.
A relatively benign scenario would see higher productivity and lower labour costs, leading to lower prices and less inflationary pressure. But Page also identifies a risk that AI technology remains in the hands of monopolist producers, rather than becoming widely available. In this case, we could find “returns accruing to a small number of people, driving some profits higher but not resulting in widespread cost reductions”.
Months – or decades?
In today’s economic climate, higher productivity and lower inflation would certainly be welcome. But how long will it take for the impact of AI to be felt?
Barclays analysts point out that “history shows that it actually takes a very long time for a single game-changing technology to materially raise an economy’s potential growth rate”. Although the car and electricity were invented just before the turn of the 20th century, US total factor productivity only grew by 1% per year between 1900 and 1920.
Economist Erik Brynjolfsson describes a “productivity J-curve”, which sets out that new technologies can only deliver productivity gains after a period of investment in new business processes and skills. The disruption caused by this transition period often results in temporarily lower productivity: in layman’s terms, things get worse before they eventually improve.
Yet there is reason to think that AI might be able to buck the trend. Brynjolfsson and colleagues point out that ChatGPT was the most rapid product launch in history, gaining 100mn users in just two months. Crucially, it was accessible to anyone with an internet connection and did not require any investment on the user’s side. In theory, AI advances could be rolled out quickly and easily, flattening the ‘J’. But whether we will continue to see unrestricted – and costless – access to AI in practice is another matter.
In a recent open letter, signatories including Elon Musk and Yuval Noah Harari called on AI labs to “immediately pause, for at least six months, the training of systems more powerful than [Chat]GPT-4”. They warned that “AI systems with human-competitive intelligence can pose profound risks to society and humanity”, adding that powerful systems “should only be developed once we are confident that their effects will be positive and their risks will be manageable”.
Setting aside these nightmarish visions of the future, even AI in its current form presents significant challenges.
Théo Kotula, ESG analyst at Axa Investment Managers, told Investors’ Chronicle that although calls to pause AI development look “unrealistic”, they “still highlight that we need to take time to better understand the risks associated with AI, and that these do not exceed its benefits”.
Kotula says that “we believe that the positives of AI can exceed the risks if AI tools and solutions are built and used following a responsible approach”, adding that “AI systems should be fair, ethical, transparent, and relying on human oversight.”. Yet there is a risk that regulators will struggle to keep up with the rapidity of new developments: “regulation has a role to play here, but it remains nascent and does not cope with the pace at which AI is developing”, he warns.
There is also the looming threat of AI-augmented hacks.
Deutsche Bank analysts think that cyberattacks will increasingly involve AI, increasing the risk that already-stretched systems become overwhelmed. In 2022, 81% of respondents to a World Economic Forum survey said that “staying ahead of attackers is a constant battle and the cost is unsustainable”. This could be about to worsen: the analysts speculate that ‘new’ forms of attack could evolve, where hackers manipulate the information fed into AI systems, or corrupt the process through which AI is created.