A just AI transition?

Even given my cynicism regarding the current hype about artificial intelligence (AI)*, I have to admit that it’s very clear this new technology will transform the world of work. The societal excitement about Chat GPT and other large language models (LLMs) has been matched by corporate excitement. Companies across the world are experimenting broadly, many of them keen to deploy this as a cost-saving tool.

Of course, as with every technology shift, cost-saving comes in the form of replacing people with machinery. Efficiency means being able to do things more quickly with less human intervention. If the current experiments with AI deliver, perhaps companies will redeploy those humans to other work. More likely they will remove those people, and their costs, from their business.

That’s how business, and economies as a whole, operate: moving to more efficient ways of delivering what customers and society want, to enable higher profits or simply to enable companies to compete with rivals which are also trying to reduce their costs. On the whole, this is good for economies too as more efficiency allows national resources to the deployed to where they deliver most value.

But that redeployment takes time, and technology transitions are painful processes, for individuals and for society. Discussions of efficiency, cost savings or redeploying resources divorce us from the very real human and emotional impacts of these changes, which are of individuals losing their jobs and livelihoods, and subsequently struggling for money and self-esteem. Even where a technological shift does create new opportunities (which has been the case with every such transition previously and so seems likely once again), that takes time – time in which individuals feel unanchored, unvalued, and perhaps reach an age where further employment is unavailable to them. That can serve to destabilise society further. We shouldn’t let the economics blind us to the personal and emotional.

There is much talk in sustainable investment circles of the need for a just transition (sometimes a fair and just transition) to a decarbonised world, ensuring that care is taken to protect and support those individuals whose jobs are impacted by the dramatic shift to economic activity that must come as the world finally faces up to the realities of climate change. There is also likely to need to be a fair and just transition to an AI-enabled world.

Recent work from the International Monetary Fund (IMF) begins to open a window on this challenge, building on the sentiment of managing director Kristalina Georgieva in a blog from a year ago called AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity. The organisation is developing approaches to consider which jobs – and which economies overall – will be impacted by the advent of AI.

Most recently, IMF staff considered impacts in Asia. A blog this month, How Artificial Intelligence Will Affect Asia’s Economies, tries to map this out in more detail, based on analysis of the breakdown of jobs in each economy. The blog reflects a deeper discussion within the analytical note attached to the Fund’s most recent Asia and Pacific Regional Economic Outlook. This analysis suggests a greater exposure to AI impacts in the region in what IMF jargon terms advanced economies, while emerging economies are likely to face lower impacts. However, they also seek to assess whether those impacts will be positive or negative for jobs: around half the impacts in advanced economies are where AI is complementary to the job, potentially driving economic benefits; meanwhile in emerging economies the majority of impacts are where there is much more likelihood of workers finding their jobs replaced. The economists hedge this analysis with language such as ‘low complementarity’ and ‘displacement’ of work, but the thinking is clear.

The language was more blunt in some earlier less detailed work, suggesting AI “could endanger 33 percent of jobs in advanced economies, 24 percent in emerging economies, and 18 percent in low-income countries”. Those conclusions look more worrying than the most recent analysis, but even the lower levels of estimated disruption are very significant.

According to the most recent analysis, there is also a gendered split in the potential impacts, again potentially exacerbating existing inequalities:

The blog reads:

“The concentration of such [complementary] jobs in Asia’s advanced economies could worsen inequality between countries over time. While about 40 percent of jobs in Singapore are rated as highly complementary to AI, the share is just 3 percent in Laos.
“AI could also increase inequality within countries. Most workers at risk of displacement in the Asia-Pacific region work in service, sales, and clerical support roles. Meanwhile, workers who are more likely to benefit from AI typically work in managerial, professional, and technician roles that already tend to be among the better paid professions.”

Georgieva was clear about the risks: “In most scenarios, AI will likely worsen overall inequality, a troubling trend that policymakers must proactively address to prevent the technology from further stoking social tensions.”

The IMF economists are increasingly clear about what needs to be done about these inequality risks. According to them, a just transition will require:

Effective social security nets
Reskilling programmes for affected workers
Education and training to enable effective application of the AI opportunity, particularly for those economies where AI is currently seen to have low impacts – so that the positive benefits can be enjoyed
Regulation to promote ethical AI use and data protection

The IMF, in its AI Preparedness Index, suggests that there is a broad spread in the readiness of global economies for this coming wave of technological disruption:

Again, as things stand it seems that the greatest likelihood is for AI to exacerbate existing inequalities. Preparing for this major economic shift will demand fresh policies and investment. These are significant challenges for world economies, for companies as they embed AI into their workflows, and for global investors, to rise to.

*I have my doubts about each of the A and the I in artificial intelligence: calling an activity ‘artificial’ when it depends on the horrific grinding work of many people to scrub its results seems inaccurate; and calling it ‘intelligent’ when it is simply a logic puzzle about the likelihood of putting one word after another – the stochastic parrots as described in that prescient article (I particularly like the analogy of Emily Bender, one of the authors of that article and a professor at the University of Washington, that AI is reproducing text in the way people might if they had unrestricted access to the National Library of Thailand but without pictures or dictionaries to enable them actually to understand or translate the language). A more recent article touching on these matters is the excellent Ask me Anything! How ChatGPT got Hyped into Being, which among other things states this fundamental truth: “LLMs are not designed to represent the world. There is no understanding by the artificial agent (chatbot) of the meaning of the output it creates. It is us humans who create that meaning.” More directly, the word soups that I have been presented with by colleagues show very clearly the limits of the technology in doing anything without clear instruction and precise pre-existing materials to work with.

Data and inequality

Long-term readers will know that I have my doubts about the value of the precise measurement of inequality known as the Gini coefficient. It’s one of the specific examples of my general cynicism about overly specific measurements. Things are usually a good deal more complicated.

Nonetheless it is fascinating to read John Burn-Murdoch’s piece for the Financial Times this weekend, Inequality hasn’t risen. Here’s why it feels like it has, which is grounded wholly in the Gini coefficient. Burn-Murdoch is the FT’s Mr Statistics, and he is always readable and interesting, so I’ll forgive the obsession with the use of single datapoints that will always be flawed if viewed in isolation (and I’ll acknowledge also that he recognises the limits of Gini: “Aggregate inequality statistics certainly have their place, but they can mask important nuances,” he writes).

The core of Burn-Murdoch’s article is the data shown in these charts:

He notes the increased attention to inequality over the past decade, even though the Gini numbers suggest it is in fact falling in both the UK and US, and he attributes this to a squeezing of the middle class between improved treatment of the poorest paid (which appears to be the driver of more than the total drop in the Gini measurement) and an increased stretching of the advantages of those best off in our society (which has increased inequality, but less than the reduction driven by the improvements for those least well off). While it’s known that the minimum wage in the UK and its bolstering over time has helped those least well off here, it is more surprising to find that some similar improvements have also been experienced by the poorest in the US also, despite the notorious stagnation of the national minimum wage there. This squeezing effect is reflected in the title given to the article in the paper copy of the newspaper: Why the middle class is right to feel squeezed.

This blog would naturally suggest that one reason why there is increased noise about inequality now, despite the falls in the Gini coefficient, is that the problem as humans see it is unfairness, not inequality, and the bluntness of Gini fails to capture the richness of human experience. In particular, it’s hard to believe that the improvement in the incomes of the poorest is genuinely felt to be an improvement in their life experience – perhaps especially as they have experienced greater inflation than wider society over recent times. The moves in Gini may be masking insight as much as they are revealing.

Burn-Murdoch is clearly right when he highlights the reductions in sense of opportunity available to many in society – opportunity both for themselves and for their children. But, for this blog at least, these are questions of fairness, not of inequality (measured by Gini or in some other way).

I am happy to confirm as ever that the Sense of Fairness blog is a purely personal endeavour.

Inequality hasn’t risen. Here’s why it feels like it has, John Burn-Murdoch, Financial Times, 4 January 2025

Deaton’s economics: fair criticism?

It is remarkable that the International Monetary Fund, one of the bastions of our modern economic construct, should be so willing to test and challenge current economic thinking. But that is what it does in publishing a striking short blog by respected economist Angus Deaton. Deaton is best known for his remarkable work on the US epidemic of what he has dubbed deaths of despair and he also led a recently-completed eponymous review of inequality for the Institute of Fiscal Studies. Deaton offers what amounts to an apologia for modern economics, and suggests some routes that may be more productive for the future. Not only might they be more productive, I would suggest that they are also likely to be fairer.

In the blog, Deaton questions mainstream economics. He does so from a remarkably mainstream position. He won the Nobel Prize in 2015, and is a professor at Princeton. His criticism of the failings of current economics, and not least of current economic education, should therefore hit home.

The core of Deaton’s points are made in crisp discussions under a handful of bullet-point headings. These are: power, philosophy and ethics, efficiency, empirical methods and humility (doesn’t our entire world need a whole lot more of that last?). He comes most crisply to his point in the first of these: “Without an analysis of power, it is hard to understand inequality or much else in modern capitalism.” But the bullet points reflect a continuity of thought, not separate ideas. He complains at the loss of ethical thought from economics and its replacement by an emphasis on efficiency and a simplifying focus on the financial: “We often equate well-being with money or consumption, missing much of what matters to people.”

Under efficiency, he states:

“Many subscribe to Lionel Robbins’ definition of economics as the allocation of scarce resources among competing ends or to the stronger version that says that economists should focus on efficiency and leave equity to others, to politicians or administrators. But the others regularly fail to materialize, so that when efficiency comes with upward redistribution—frequently though not inevitably—our recommendations become little more than a license for plunder.”

I think that quote bears rereading.

Applying these five approaches as a new lens for approaching questions, Deaton reaches a range of fresh conclusions – or rather a reduced level of certainty – about a number of different issues. These include: unions, free trade, global poverty and immigration.

But though it is not among these bullet-points, or the issues about which Deaton now has less certainty, to my mind one of the most notable single words in the piece is ‘efficacy’. Deaton says: “today we [economists] are in some disarray. We did not collectively predict the financial crisis and, worse still, we may have contributed to it through an overenthusiastic belief in the efficacy of markets, especially financial markets whose structure and implications we understood less well than we thought.” Normally economists and investors talk about market efficiency, and certainly the financial crisis was in part due to overconfidence that markets are efficient, that they will find the right prices for things. The efficient market hypothesis – which many investors take as a certainty, even though it is merely an hypothesis, and even though there would be no ability of active investors to outperform if it were true (admittedly many are more lucky than genuinely generate outperformance, but nonetheless it is still possible to outperform a market). The crisis showed that market pricing can often be very wrong and the use of market prices as a foundation for valuations can be risky.

Deaton is clearly referencing the Efficient Market Hypothesis (and the use of ‘efficiency’ as one of his bullet-point headings makes more notable his decision not to use the term in his comment about the disarray of modern economics), but he is actually making a very different point. He is asking whether markets are always efficacious, whether they work and always add value to human society. And his clear view is that they are not always, and do not always. We should listen, particularly those of us who work in financial markets.

Deaton has never minced his words, but here he is remarkably cruel about his profession. He says he does not want to get into the question of corruption among his peers, though he notes that allegations “have become common in some debates”. But he does state, bluntly: “economists, who have prospered mightily over the past half century, might fairly be accused of having a vested interest in capitalism as it currently operates”. In a blog that clearly has real concerns about the operation of modern capitalism, that fair comment is one that should hang over the profession, challenging all to rethink with the confidence and honesty that Deaton has.