AI technologies bring greater accuracy, precision and speed to analysis and forecasting. They can conduct risky activities and repetitive tasks that are boring for humans. They are also readily available and make it easy to scale up technological services and save costs. But how are these effects felt around the world, especially against the backdrop of widening inequality?
While we’ve witnessed a decline in the incidence of extreme poverty over the past few decades (progress that is now being reversed by the impacts of the pandemic, climate crisis and rising debt), inequalities have continued to rise globally, across income, wealth, assets and human development. Those who already have deep pockets or high levels of development have amassed more proportionately, leading to widening gaps between and within countries. So, what does AI mean for the world’s most vulnerable people?
Research on AI and automation finds that countries face different potential levels of automation, owing to differences in employment, the automatability of jobs, labour market structures, education, skills levels and government policies. Jobs and industries involving more routine and codifiable tasks face a greater risk of automation, while others that require high human capital or technical skills are harder to automate and thus more resilient. China and India are among the countries whose populations face the highest risk of automation.
Job opportunity vs vulnerability
The development of AI solutions also includes tasks that depend on human oversight. Most machine learning applications use supervised learning, which requires labelling datasets correctly to train the AI models. Sometimes termed “ghost work”, this is often outsourced to a remote workforce and includes tasks such as tagging and labelling images or other content, moderating, proofreading, editing and sorting. While these new forms of work offer valuable opportunities for income generation, they also remove many of the traditional safety nets protecting workers and risk perpetuating poor working conditions and deepening existing job vulnerability.
While AI offers many potential benefits, in the short term it may lead to issues such as job displacement and skills polarisation, resulting in wage inequality. Different countries and regions may be differently prepared to deal with affected workers. Mitigating mechanisms such as policies protecting vulnerable workers, economic safety nets for the unemployed, reskilling initiatives and the ability for other sectors to absorb displaced workers could greatly alter the possible negative societal impacts.
Where these mitigating mechanisms are not in place, the profits of automation and AI adoption will be concentrated in the hands of those who own the technologies, unless invested back into the production process. Also, where AI disadvantages those in automatable middle-skilled work and increases the returns to high-skilled workers, wages may be polarised. High-skilled workers might start earning more while others face falling wages and greater competition for lower-skilled jobs that are resilient to automation.
AI adoption and the digital divide
The potential benefits of AI are harder to reap where access to digital infrastructure and data is lacking, and skills levels and cultural beliefs condition people’s interactions with AI. More than 10 per cent of the world’s population are illiterate, roughly 40 per cent still have no access to the internet and over half don’t own a smartphone. These poorly connected or disconnected populations are unevenly distributed geographically and excluded from the many AI-powered services that make accessing vital information easy for those who are plugged in. This includes e-commerce, education and training, democratic processes and health services.
Beyond this, there is a digital divide in terms of people’s ability to use devices and the internet, understand digital content and create and participate in it. This divide is linked to other socio-demographic characteristics such as gender, race, income, education, location (whether urban or rural) and age. Where economies do not have a sufficient pool of skilled workers or the computing resources to collect, store and process data, adoption of AI to address local needs will be limited.
Additionally, some communities might not have the resources to develop AI solutions, and externally designed ones might not be fit for purpose. They might not reflect a diverse range of perspectives in their design or in the data collected. Where vulnerable and disconnected populations are inadequately reflected in AI design, the existing digital divides can deepen further.
Tackling inequality and improved efficiency
While the inequitable distribution of AI and automation poses challenges, it can also offer many potential benefits and solutions. Sustainable, responsible and accountable AI practices consider a product’s impact on the world and on diverse populations. This thoughtful approach can also avoid deepening inequalities or harming vulnerable people.
AI adoption can increase efficiency and profits, and help facilitate the job market’s shift away from repetitive routine tasks and towards those requiring cognitive and socio-emotional skills. Countries and regions can start to direct efforts towards knowledge-intensive services, therefore avoiding spending their scarce money and resources on costly or inefficient activities that could easily be automated.
While AI adoption may leave some worse off in the short term, having appropriate policies in place could mitigate these risks. Historically, technological disruption has often led to increased inequality temporarily, with some populations winning and others losing, but in the longer term these differences have tended to lessen and the overall level of wealth has risen.
It is crucial that those in charge of developing AI solutions and adopting them are aware of the potential pitfalls and develop frameworks to counter any negative effects. AI enables faster and more accurate processes and its benefits should be harnessed to the benefit of essential and productive sectors worldwide.
I discuss this topic further in a recent report from The Alan Turing Institute.
Sanna Ojanperä is a doctoral student focusing on platform work and AI in the workplace at The Alan Turing Institute and the University of Oxford.
This article originally appeared in our issue The Future of Work: AI and Automation.