fbpx

Artificial intelligence and radical technical innovation: the impact of the Fourth Industrial Revolution on mixed migration

The following essay was originally compiled for the Mixed Migration Review 2019 and has been reproduced here for wider access through this website’s readership.

Artificial intelligence (AI) is a major feature of the nascent Fourth Industrial Revolution (4IR), or “second machine age”, which is predicted to have major impact on the way humans live, work, play and interact with each other, as well as on the environment and technology itself.

This essay explores the current and possible future influence of AI on refugees, immigration and mobility.

The Fourth Industrial Revolution and artificial intelligence

Beginning in the late 18th century, the First Industrial Revolution harnessed water and steam to bring mechanisation to manufacturing; electricity powered mass production in the Second; electronics and information technology delivered the Third through increased automation. “Now a Fourth Industrial Revolution is building on the Third, the digital revolution that has been occurring since the middle of the last century. It is characterized by a fusion of technologies that is blurring the lines between the physical, digital, and biological spheres.”

The 4IR is both an extension of previous revolutions and a new era in its own right, one that is “disrupting almost every industry in every country and creating massive change in a non-linear way at unprecedented speed.” This is the future environment in which the causes and conditions of and responses to mixed migration will take place.

AI encompasses computerised processes that mimic human intelligence, such as image recognition, translation, and decision-making. Impressive progress has been made in AI in recent years thanks to the exponential growth of computing power and the generation and availability of ever-increasing amounts of data, driven by the subfield of machine learning.[1]

According to one estimate, AI is expected to add $15.7 trillion to the global economy and to boost local economies by 26 percent by 2030. But, as we shall see, the benefits will not be evenly spread across the world.

AI anxiety

The accelerating pace of AI development is generating a high degree of anxiety and excitement. While AI already outperforms humans in a limited range of activities, before long it is expected to do better than us in almost all cognitive tasks. Debates about the threats of “super-intelligent” forms of AI, the advancement towards killer robots, or even sexbots, have received wide coverage.[2] Many are wary of the dangers of AI: avowed futurist entrepreneur Elon Musk described it as having the potential to be an “immortal dictator from which we can never escape.” The European Union Agency for Fundamental Rights and others are already warning of the potential human rights risks of AI.

Whatever the verdict, AI is already widely in use every day and has become so ubiquitous that we barely notice its presence and spread, from voice-activated computer characters such as Siri (Apple) and Alexa (Amazon), through driverless cars, drones and weapons systems, to software used to discover new drugs, and algorithms that predict our cultural interests and home heating requirements. Machine learning is already overtaking humans in predicting deaths and heart attacks. This is just the start, we are assured, and already AI is having some initial application in relation to mixed migration, and to refugees in particular.

Al applications

Refugees and asylum seekers

Just as the number of displaced people is at a historic high (more than 70 million as of June 2019), so is the level of technology that can help refugees and displaced people increase their opportunities and quality of life. AI is poised to change the ways the world engages with refugees (and vice versa) at a time when durable solutions are scarce and problematic. Of course, the critical political issues of burden sharing, support and resettlement are not solved by technology alone, but an increasing number of applications have already been developed to assist refugees and migrants, some of which explicitly use machine learning and AI. Here is a selection:

  • Free robot lawyers and information providers
    DoNotPay is a UK-based chatbot initiative that provides free legal advice to refugees through intelligent algorithms and offers customized legal help, including guidance through the UK asylum application process. Refugee Text taps into mobile networks to provide crucial information to refugees via automated SMS messages.
  • Matching refugees with jobs
    Stanford University’s Immigration Policy Lab has developed a machine learning algorithm to help governments and resettlement agencies find the best places for refugees to relocate to in terms of matching their skills, education, language levels, etc. Switzerland pioneered the use of this algorithm in late 2018. The algorithm is predicted to increase employment of refugees by between 40 and 70 percent.
  • Optimising integration
    Using machine learning, optimization algorithms and complex computation of data to improve refugee integration, an award-winning initiative named Annie MOORE (Matching Outcome Optimization for Refugee Empowerment) matches refugees to communities where they will find resources appropriate to their needs and preferences, including employment opportunities. Rafiqi is an interactive online AI tool that connects refugees to mentors and opportunities, also with the goal of accelerating and easing integration. It has been available to a group of refugees in London and Berlin since 2018 and the company is planning to extend around Europe.
  • Psychological support
    Many refugee camps have limited or no mental health services even though refugees are at a higher risk of mental health disorders, including post-traumatic stress, depression, and psychosis.19 Responding to this need, Silicon Valley start-up X2AI, developed Karim, an intelligent chatbot that has personalized text message conversations for emotional support in Arabic, one of the most commonly-spoken languages among refugees.
  • Machine learning and remote sensing data
    Conducting on-the-ground surveys of settlements including structures, can be labour-intensive, time-consuming, costly and dangerous. A 2018 review of refugee and IDP settlement mapping indicated that machine learning autocoding technologies may offer important help to humanitarian organisations. Although these efforts are still ongoing, the use of machine learning and remote sensing data, including satellite high-resolution imagery, promises to improve efficiency.
  • Tracking flows
    Big data and data science are used to track the flows of refugees and displaced people. In mid-2017 the United Nations announced that the Data For Democracy team had won the Unite Ideas Internal Displacement Event Tagging and Extraction Clustering Tool challenge, by “building a tool capable of tracking and analysing refugees and other people forced to flee from or evacuate their homes.” The IBM Watson News Explorer searches the web continually to isolate all news articles relating to refugees. UN Global Pulse, through its lab in Kampala, has been working with the UN and the government on a Radio Content Analysis Tool to “explore how insights from public talk radio can provide real-time information on what is happening on the ground”.
  • Machine learning and potential bias in asylum adjudications
    In a recently multi-country study, researchers used machine learning to analyse almost half a million asylum hearings in 336 locations, rendered by 441 judges between 1981 and 2013. They developed a predictive model based on case data that proved to be 82 percent accurate in refugee cases when tested against actual judicial outcomes. They found that “extraneous factors” may be influencing decisions resulting in potentially “unfair” decisions or at least showed that adjudications were subject to bias.
  • Paying refugees to boost AI development
    REFUNITE, a nonprofit set up to help refugees reconnect with their families, has developed a mobile phone app called LevelApp, which allows refugees to earn money by “training” algorithms for AI and gaining a foothold in the “global gig economy” while helping AI companies dramatically reduce costs. The 5,000 refugees in Uganda involved in the pilot programme are mainly from South Sudan and the Democratic Republic of Congo. REFUNITE hopes to increase involvement to 25,000 refugees by 2020.

Migrants and immigration

In relation to migrants and immigration the three main areas of AI being piloted are processing, prediction and prevention. Compared to the above list of examples of refugee-oriented AI innovations, those relating to migration are arguable more controversial insofar as processing, prediction and prevention are closely associated with control and the potential preparation of restrictions.

Processing migration

For over a decade, Hong Kong’s Immigration Department has been using eBrains, an award-winning AI technology that uses “business rules, data mining, machine learning [and] AI clustering” to process visa applications. AI technology provides “decision support” for millions of annual visa applications.

According to a report released in late 2018, the Canadian federal ministry responsible for immigration has been experimenting with AI since at least 2014. The ministry argues that the system is primarily used by immigration officers as a sorting mechanism to quickly separate complex visitor visa applications from standard ones. However, some human rights experts have concerns about how AI will change the immigration system, and about what it will mean if computers ultimately make some decisions autonomously. Some argue that the nuanced and complex nature of many refugee and immigration claims mean they are ill-suited to “technological experiments” and time-saving automation. “These systems will have life-and-death ramifications for ordinary people, many of whom are fleeing for their lives.” This report’s analysis echoes global human rights concerns when it states “…immigration and refugee law [are] a high-risk laboratory for experiments in automated decision-making.”

Predicting migration

Migration forecasting is “notoriously difficult and unreliable”. Despite more sophisticated techniques, more reliable data and the contemporary use of big data and machine learning, there is little evidence we are significantly closer to finding methods of prediction more reliable than those E.G. Ravenstein attempted in the late 1880s.

A recent comparative analysis of international migration in population projections developed by the Global Knowledge Partnership on Migration and Development (KNOMAD), highlights some consensus, but also a ”considerable amount of disagreement about the size and direction of actual migration flows between major sending and receiving countries. Basic assumptions
about future flows also significantly diverge.”

Traditional human mobility models, such as gravity and radiation models, offer some predictive capacity based on population and distance features. What is needed for predictive capacities to be useful and more reliable is a model that captures more complicated migration dynamics. Although successful machine learning models that incorporate a variety of exogenous features to predict origin/destination migration flows remain elusive, some data scientists are currently attempting exactly that. Some experts claim that their “machine learning models outperform traditional human mobility models on a variety of evaluation metrics…”  The aim is to model human migration under different what-if conditions, such as “potential sea level rise or population growth scenarios”.

Some experts are more optimistic about the powers of predictive analysis to forecast humanitarian crises as well as about how AI can be used to “predict Africa’s next migrant crisis”. Additionally, some NGOs are working with data scientists and the private sector to attempt predictive programmes for mixed migration and internal displacement, but to date AI’s use in this area is limited and successful implementation continues to elude modellers.

Preventing irregular migration

There is evidence of a growing capacity to use AI in border control and border management as part of an expanding security market that in Europe alone is predicted to be worth $146bn by 2020.

In October 2018, the European Union announced it was funding a new automated border control system to be piloted in Hungary, Greece and Latvia. The system, called iBorderCtrl, uses “smart lie-detecting avitars” to question travellers seeking to cross borders.

ROBORDER, a consortium of research institutions, law enforcement agencies and tech firms, is already testing its systems. According to its own publicity ROBORDER “aims at developing and demonstrating a fully-functional autonomous border surveillance system with unmanned mobile robots including aerial, water surface, underwater and ground vehicles which will incorporate multimodal sensors as part of an interoperable network.” Its intention is to implement a “heterogenous robot system” and enhance it with detection capabilities for early identification of criminal activities at border and coastal areas.”

Concerns have been raised over the emergence of such “techno-solutionism in border monitoring systems” and the potential for further human rights violations as “swarms of robots” patrol the EU’s land, air and sea borders and as promotors of such methods conflate security and terrorist threats with irregular migration.

One futurist has proposed replacing border controls and an expensive and inefficient wall between Mexico and the US (estimated cost: $25 billion) with thousands of drones for a fraction of the cost. These drones would monitor the borders day and night, “with loud bilingual speakers to talk to illegal immigrants trying to cross into America, and they can also have facial recognition software to see if immigrants are on criminal lists.” This proposal goes further than border controls, advocating for “very authoritarian measures such as tracking refugees and migrants (particularly those from the Middle East) for years after they entered the country and until they have proven not to be a danger to society.”

There is little doubt that AI and robotics will become integral to future security systems and therefore to border control and preventing irregular migration, not least because technology can significantly enhance current systems that struggle to deal with large numbers of people crossing multiple borders and lengthy distances, frequently in remote or harsh areas. As other essays in this Mixed Migration Review explain, the likely impact of demographic changes and environmental stressors amongst other future changes will lead the number of migrants and refugees in the world to increase significantly. With asylum seekers, refugees and migrants, travelling together in mixed irregular flows, often in vulnerable situations, this raises serious concerns.

‘Creative destruction’

Creative destruction is an oxymoronic term introduced to economic theory in 1942 by the Austrian economist Joseph Schumpeter. He used it to describe the special form of economic growth that entrepreneurs bring to the capitalist system through radical innovations.

Despite the destruction of older industries and economies, the real force that has sustained long-term economic growth recently has, wrote Schumpeter, been the “perennial gale of creative destruction” powered by the technical innovation and industrial mutation that continuously revolutionise the economic structure from within, “incessantly destroying the old one, incessantly creating a new one”. The 4IR is one such transformation that many believe will affect all aspects of society and will bring about the innovation of entire systems. “The business models of each and every industry will be restructured” and with it the impact on all aspects of the
labour market.

Refugees and migrants are likely to be among the demographic groups most affected by this anticipated global disruption of skills and jobs, not only in terms of their prospects of finding employment in destination countries but also in terms of root causes and drivers of conflicts, displacement, and mobility in the first place.

Exacerbating inequalities

According to World Economic Forum Executive Chairman Klaus Schwab, “In addition to being a key economic concern, inequality represents the greatest societal concern associated with the Fourth Industrial Revolution.” Such speculation is perhaps hazardous as no one can be sure how these seismic transformations will unfold in the coming decades. Meanwhile, the debate rages on.

Societies and economies have adapted well to earlier technological innovations and pessimistic predictions have sometimes proven to be exaggerated. However, what many analysts coalesce around is the notion that any economic disruption will favour early adaptors and innovators, meaning those already investing in the 4IR – those providing the intellect and physical capital – will be the initial winners as they concentrate and accumulate technical advances and therefore economic power.

One source estimates that by 2025 global annual revenues from the AI software market alone will be worth $118.6 billion. Of course, if early adaptors or AI businesses are based in less-developed countries, global inequalities may actually decline. Further, innovative producers do tend to get wealthy but they also pass cost-savings on to consumers, who as a whole benefit quantitatively far more, and far more equitably too: when cheaper goods are made we all have access to them as prices fall, as they do over time for new technology. Televisions, personal computers and mobile phones are cases in point. It has been argued that as global productivity rises rapidly, so will wages.

The future of work

The nature of AI’s future impact on the labour market is the subject of particularly fierce debate, and while previous technical revolutions have, after challenging transition years, led to the creation of new kinds of jobs, there is no guarantee, or consensus, that such history will repeat itself.

Optimists speak of AI creating millions more jobs than it will eliminate, or of eliminating just a small fraction of current jobs in the world’s wealthiest countries: “there are not a fixed number of jobs that automation steals one by one, resulting in progressively more unemployment. There are as many jobs in the world as there are buyers and sellers of labor.”

Some pessimists, on the other hand, predict computerisation puts almost half of current US employment at risk. Others have warned of a similar impact in Europe. Even the human medical profession is said to be susceptible to large-scale obsolescence.

It’s still not clear how the current changes will affect job availability for migrants. What is safe to say is that the rewards of the second machine age will be predominantly reaped by the small minority of people and companies that own the machines, and who are unlikely to be located in or benefit countries where mixed flows of migrants and refugees originate. A transition period may be problematic, but in the longer term the impact of technology on migration could equally be positive, rather than negative.

Labour market polarisation and the Global South

As automation replaces human labour across entire national economies, thereby impacting the international economy, the net displacement of workers by machines might exacerbate the gap between returns to capital and returns to labour. Schwab and others see an increasing segregation, or polarisation, of the job market into low-skill/low-pay and high-skill/high-pay sectors, and a deeper “hollowing out” of middle-income jobs.

Winners and losers

In a more winner-takes-all economic system at national and international levels there are strong possibilities that social tensions will increase as the divergence between 4IR winners and losers becomes starker. Countries in the Global South therefore risk being left behind in the 4IR and their inability to be ready to take up AI in time. “Not only will they not reap the potential benefits of AI, but there is also the danger that unequal implementation widens global inequalities.”

In countries where the price of labour is very low and education outcomes are also low the uptake of AI will most likely be slowest. The Government Artificial Intelligence Readiness Index illustrates that many refugee and migrant countries of origin are ill-equipped to make changes towards AI. For example, Somalia, Eritrea, Sudan, South Sudan, and the Democratic Republic of  Congo are all countries that produce both migrants and refugees, are at the very bottom (last 10) of the index, and also in the lowest decile of most human development rankings. By contrast the most preferred countries of destination for migrants and refugees are those with the highest AI readiness ranking.

Nevertheless, the economies of Asia and, increasingly,those of several African states, have shown a fast uptake of and adaptation to technology, often leapfrogging regions in advanced economies. Given good levels of education, investment and vision, it is by no means pre-determined that the Global South will be left behind.

Driven harder to uncertain futures

Still, at regional and global levels these divisions could create yet stronger drivers (unemployment, stagnant economies, poor governance and conflicts – all interlinked) for people in countries of mixed migration origin to move to where they believe they will have better opportunities.

But will there be jobs, and will they be accepted or given access? Many jobs taken up by refugees and migrants in destination countries are low-skilled ones, such as transportation (particularly taxi and delivery drivers) which may be increasingly automated in the future, with the proliferation of driverless trucks, public transport services and cars, and drones.

However, lower-skilled jobs such a domestic workers, and jobs in the hospitality industry as well as the care and health sectors, absorb many refugees and migrants globally, and will continue to be needed and are more resistant to AI substitution.

Room at the top

AI also has applications in a variety of highly educated, well-paid, and predominantly urban industries, including medicine, finance, and information technology. When studies suggest AI could create millions of new jobs, it may be assumed that the majority of these will be high-skilled or specialist positions, especially in the longer term. At one level this means talent, skills and education will have a high premium in tomorrow’s workplace, and where migrant workers are needed, only the high-skilled will gain access. This trend is already happening, with various OECD countries only selecting high-skilled and specialised migrants.

Most governments “either seek to raise (44 percent) or maintain (41 percent) current levels of immigration of highly skilled workers”, while only four percent of governments have implemented policies to reduce the inflow of highly skilled workers into their country. The share of governments with immigration policies focussed on highly-skilled workers doubled from 22 percent in 2005 to 44 percent in 2015. Furthermore, between 2005 and 2015, policies to encourage immigration of highly skilled workers increased across nearly all regions.

Impact on flows

Irregular mixed flows of refugees and migrants comprise many unskilled and partially skilled workers but few highly skilled workers. Immigration policies of destination countries tend to discriminate against low-skilled workers, despite economic demand for them (albeit in the grey economy). This disconnect is likely to boost irregular migration, as it already has for many years in Saudi Arabia for Ethiopians who are deported en mass on a regular basis but repeatedly return irregularly to fill hundreds of thousands of low skilled positions.

Regular pathways, meanwhile, may end up being almost entirely blocked by the effects of AI: “Labour migration, with the exception of certain very highly skilled professions, could soon be a thing of the past. Sooner or later we might face the situation that humans will be subject to immigration laws, whereas non-human systems able to perform the same tasks as humans will only be subject to certain product certification requirements, but not to migration restrictions.”

Will universal basic income close borders further?

Direr – some would say more realistic – predictions of AI’s impact on work often lead to discussions, in the Global North at least, of how to deal with large numbers of jobless citizens and the need for a universal basic income (UBI) to prevent poverty and increase equality.

If wealth gaps between countries continue to widen, UBI might attract irregular migration and, in turn, incentivise greater restrictions in destination states ever more suspicious that outsiders are relocating simply for their social welfare benefits. However, several studies show that the welfare state is not a major pull factor in migration determination. The results of one “reveal no evidence for a magnet effect to the most generous welfare states in the world net of other recognized factors, and even suggest a negative influence linked to the region’s high cost of living.” In any case, it would be very easy for states to exclude recently arrived migrants from UBI schemes, or to have a graduated system depending on duration of presence in the country, accumulation of taxes paid, etc. Similar systems already exist in relation to social benefits in some countries.

Conclusion

Inevitably, mixed migration will be affected in different ways by the radical innovations brought about by 4IR. While some changes may benefit migrants and refugees, others might not. To date, it is hard to see any balance between the advantages and disadvantages; instead, as global and regional inequalities widen, marginalised and vulnerable groups will come under increasing pressure. In turn, growing demand for irregular mobility is likely to occur in a context of diminishing opportunities and more restrictive borders – many of which may be protected by AI technology.


[1] Machine learning is an AI application that enables systems to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it learn for themselves.

[2] See, for example: Chivers, T. (2014) Superintelligence by Nick Bostrom, review: ‘a hard read’ The Telegraph; Human Rights Watch (2018) Heed the Call. A Moral and Legal Imperative to Ban Killer Robots; Wagner, C. (2018) Sexbots: The Ethical Ramifications of Social Robotics’ Dark Side AI Matters; Future of Life Institute (2019) Benefits & Risks Of Artificial Intelligence