Any who have kept up with the rush of headlines about the impact of AI technologies will know that they often wildly disagree. Some paint an optimistic picture of this future. A recent piece in The Economist, for example, argues that by lowering costs of production, AI-based automation will create more demand for goods and services. The economy may need fewer checkout attendants at supermarkets, but that's not a problem - more massage therapists will make up for that!
Other articles lean toward the heavily pessimistic view, predicting mass unemployment and worse---the rise of superintelligent machines dominating humans.
In my new book with Simon Johnson – Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity — we attempt to set out a middle road through all of this, outlining the very serious risks that the present dominance by a small number of enormous tech firms poses, as well as some potential routes towards a future of better work and better automation. Our aim in writing it was to challenge some of the simplistic narratives of ‘tech = progress’ and refocus attention on the quality of work. As I outline below, we also wanted to move towards a focus on machine usefulness rather than machine intelligence and — through changes in government policy and regulation — help reinstate human agency in this tech revolution.
What is clear from the data we have in the US over the past century is that wage inequality has increased in the labour market since the 1980s. From the 1940s to the 1970s, real wages across almost all demographic groups grew in tandem. But then from around 1980, we see a change. Wage gaps are not only becoming wider, but the real wages of many groups of low-education workers — especially men — begin declining.
How do we reconcile this with the core tenet of economics that suggests that increases in average productivity — and clearly digital technologies have done that — should ultimately lead to higher wages because they push up labour demand?
The answer is that automation — which is one of the main things that many digital technologies are used for — has more complex technological implications than what is generally offered.
One way of seeing pre-AI automation is that – deployed in a certain way – it shifts what it is possible to do with capital. With investment in sophisticated machines, existing tasks get automated, displacing work that would previously be performed by humans. Pascual Restrepo and I have called this ‘so-so’ automation: the productivity benefits are actually quite small, but the labour and wage implications are quite significant. Looking at changes in wages for different demographic groups with varying levels of task displacement over time, our research shows that automation focused on replacing existing tasks has been a major driver of inequality, and depressed the wages of lower-skilled workers.
With AI now coming on stream, the question is what extra effects it will bring into play. Work that David Autor, Joe Hazell, Pascual Restrepo and I have done recently shows that, frustratingly — despite AI having the potential to do many different things and to create new tasks — it is currently mostly being used in the automation of existing tasks, adopted by firms to do simple, routine, clerical work. This means that it has the ‘so-so’ effect: disrupting wages and labour markets, but for small productivity gains.
What then should we do to create a better future? Our argument in the book is that we need a reframing of technological progress. Rather than focusing on autonomous machine intelligence to displace existing routine tasks and perform ‘so-so’ automation, we should focus on maximising machine usefulness to humans. Take a calculator for example: it is not an intelligent machine, but it is incredibly useful. It saves time in computational tasks and enables human users to do more sophisticated things.
The diverse skills that gardeners, carpenters, electricians or tailors bring — they are very skilled, adaptable and creative and we should not try to fully automate these jobs, but rather have useful machines that help these workers amplify and improve those skills.
‘Good’ automation like this can produce large productivity gains and be complemented by new tasks for labour, it empowers workers rather than disempowering them. But the problem is that right now AI seems to be drawing us more and more in the ‘bad’ automation direction.
Why is this happening when it has such negative effects - and when better paths for technology and AI are possible? Firstly, it is a result of an excessive focus on cost-cutting. Secondly, it is incentivised by tax regimes that typically — in both the UK and the US — levy four or five times as much on labour as they do on capital spending on software and equipment. Finally, there is a current lack of countervailing power from labour movements which, as we outline in the book, have been a major factor throughout history of moderating the adoption of technology and promoting workers’ rights.
How then can we promote the conditions where ‘good’ automation focused on machine usefulness prospers?
Firstly, we need to address the current fiscal asymmetry between capital and labour. In the UK context, for example, this could mean balancing the full expensing on machines introduced in the Chancellor’s latest budget with incentives for companies to invest in workers or training. We should also explore how to effectively subsidise the deployment of automation that is in the ‘machine usefulness’ bracket and encourage wider diffusion of worker-friendly technologies.
Secondly, in the face of technologies that often reduce us to passive automata, we need to reignite human agency. This will mean new forms of collective organisation where workers are given a voice that is not focused on conflict. There are models in worker councils in Germany that could be useful, or we may need to find new organisational models. In addition to this, we also need to tighten regulation of business models based on data harvesting and surveillance that create emotional outrage but reduce genuine engagement.
To be clear though, there is no silver bullet. We want to challenge and change the narrative on AI and automation, with policies and institutions that promote technologies that make humans more productive and empowered. But we will have to work to create these things, and do so through stronger institutions, better regulation and a renewed idea of democracy that has human flourishing at its heart.
Watch Daron's presentation at our January conference for the Pissarides Review into the Future of Work and Wellbeing: